Thoth

# Row-Bot Docs Full Text

Concatenated public documentation for Row-Bot.

## Row-Bot Documentation

Route: /docs/

A complete public user guide for installing, configuring, and using Row-Bot end to end.

# Row-Bot Documentation Row-Bot is a local-first AI workbench for people who want provider-aware models, parent-led agents, durable documents, memory, tools, workflows, design, code help, integrations, and voice in one controllable system. Run it as a desktop application or a private authenticated server; this guide explains installation, Docker deployment, owner access, the main interface, settings, and every explicit external route. These pages describe Row-Bot 5.0.0, the release represented by this source tree. ## Start Here - [Getting Started](/docs/getting-started/) explains the install path, first launch, and setup choices. - [Row-Bot Interface](/docs/app-shell/navigation) tours the sidebar, Home, a conversation and its details card, panels, the terminal, the K command palette, Buddy and Settings. - [Home: Overview](/docs/home/overview) explains the first page: the Ask box, what needs you, live status and recent work. - [Chat](/docs/chat/) explains conversations, composer controls, attachments, model selection, approvals in place, tool results, and the conversation library. - [Reasoning Controls](/docs/chat/reasoning-controls) explains model-specific effort, thinking toggles, token budgets, and provider-default fallback behavior. - [Settings](/docs/settings/) explains every settings page and what each choice changes. - [Profiles, Goals, And Agents](/docs/profiles-goals-agents/) explains the Agents dialog, goals that keep going until done, delegated agents, and recovery. - [Remote Access And Server Mode](/docs/operations/remote-access) explains invitations, sessions, Tailscale, LAN, HTTPS proxies, and browser-local voice. - [Docker And VPS Operations](/docs/operations/docker/) explains the official container, persistent data and secrets, backup, upgrade, and rollback. - [Computer Use](/docs/computer-use/) explains the opt-in native desktop tool, setup, live controls, and safety boundaries. ## Feature Guides - [Workflows](/docs/guides/workflows) for repeatable background work and scheduled agents. - [Reasoning Controls](/docs/chat/reasoning-controls) for choosing the supported reasoning depth of the active model without changing models. - [Designs](/docs/designer/) for decks, documents, landing pages, mockups and storyboards in a conversation's Design panel. - [Code Folders](/docs/developer/) for binding or cloning a folder from a conversation and working on it in the Developer panel. - [Knowledge](/docs/knowledge/) for local memory, durable document ingestion, graph review, and background organization. - [Computer Use](/docs/computer-use/) for target-window automation in native Windows and macOS applications. - [Phone And Native Desktop](/docs/mobile-native/) for the same owner product on a phone and in the native desktop app. - [Remote Access And Server Mode](/docs/operations/remote-access) for owner invitations, Tailscale, LAN, SSH, Docker, HTTPS proxies, and remote browser voice. - [Docker And VPS Operations](/docs/operations/docker/) for the official image, hardened Compose, credentials, backup, upgrade, rollback, and recovery. - [Monitor](/docs/monitor/) for kept health checks, one-click fixes, activity, history, and logs. - [Insights](/docs/home/insights) for what Row-Bot noticed and the proposals it prepared for your review. - [Skills Hub](/docs/skills/) for browsing, enabling, creating, and reviewing skills. - [Progressive Tools And Skills](/docs/guides/progressive-tools-and-skills) for automatic external-tool discovery, per-task skill selection, compatibility mode, and safety boundaries. - [Channels](/docs/integrations/channels), [MCP](/docs/integrations/mcp), and [Plugins](/docs/integrations/plugins) for integrations. - [Extend Row-Bot](/docs/extending/) to choose safely between Skills, Custom Tools, plugins, MCP, channels, and accounts. - [Operations, Data, And Recovery](/docs/operations/) for backups, restore, updates, repair, uninstall, and diagnostic sharing. - [Voice and Buddy](/docs/voice-and-buddy/) for speech input, Talk, Dictate, read-aloud, and Buddy's native drag-to-undock desktop overlay. ## How To Read These Docs Each major page explains what the feature is, where to find it, the important controls, a common workflow, what is saved, privacy and safety implications, and troubleshooting. Screenshot captions describe the product UI so you can connect the text to what you see in the app. ## References Use [Reference](/docs/reference/) when you need tables of tools, providers, settings pages, channels, skills, MCP servers, plugins, storage, or approval behavior. The guided pages should be your first stop; the reference pages are for lookup.

## Getting Started

Route: /docs/getting-started/

Install Row-Bot, complete first launch, and learn the first choices that matter.

# Getting Started Start here if you are new to Row-Bot. You only need three things for a useful first session: the app installed, one working model path, and a basic understanding of where Row-Bot stores local data. ## The Short Path 1. Install Row-Bot from the release package for your platform. 2. Launch the app and answer **How should Row-Bot think?** 3. Choose one model path: local Ollama, a hosted provider, a subscription account, or a custom endpoint. 4. Send a first chat message. 5. Open Setup Center or Settings later when you are ready for documents, workflows, designs, code folders, channels, MCP, plugins, Skills, Buddy, or voice. ## Important Concepts - **Local app** means Row-Bot runs on your machine and opens a local desktop or browser window. - **Data directory** means the folder where Row-Bot keeps conversations, memories, documents, settings, workflows, logs, designs, code folder records, skills, plugins, and local integration state. - **Model provider** means the service or local runtime that supplies a model. Ollama is local; API and subscription providers usually use the internet. - **Tools** are actions Row-Bot can take, such as reading files, searching documents, using a browser, running commands in a code folder, or sending through a channel. - **Approvals** ask you to approve or deny sensitive actions before Row-Bot proceeds. They wait until you answer. ## Pages In This Section - [Installation](/docs/getting-started/installation) - [First Launch](/docs/getting-started/first-launch) - [Row-Bot Interface](/docs/app-shell/navigation)

## Installation

Route: /docs/getting-started/installation

Install Row-Bot from a release package or run it from source with clear next steps.

# Installation For ordinary use, install Row-Bot from the latest release package for your operating system. Running from source is useful for contributors, testers, and people who want to inspect or modify the app. ## Install From A Release 1. Download the latest Row-Bot release for Windows, macOS, or Linux. 2. Run the installer or unpack the archive. 3. Launch Row-Bot from the Start Menu, Applications, app launcher, or the provided command. 4. Answer **How should Row-Bot think?** to choose your first model. The installed app starts a local Row-Bot process and opens the UI in a desktop window or browser, depending on your platform and launch preference. If the usual port is busy, Row-Bot chooses another local port. ## Run From Source Use the source path if you are developing Row-Bot or testing changes: 1. Clone the repository. 2. Create a Python environment. 3. Install uv and sync the locked Python dependencies with `uv sync --locked --all-extras --group test`. 4. Install the docs-site dependencies only if you are working on documentation. 5. Run the app entry point from the repository. `requirements.txt` is a generated pip export from `uv.lock` for installer compatibility. It is available as a fallback for environments that cannot use uv, but source dependency changes should be made in `pyproject.toml`. Source runs use the same local data concepts as the packaged app. Keep test data separate from personal data when experimenting. ## Local Data Directory Row-Bot stores local app data under the active Row-Bot data directory. This includes conversations, memories, model/provider metadata, workflows, designs, code folder records, documents, logs, skills, plugins, Buddy assets, channel settings, and MCP settings. Advanced users can set `ROW_BOT_DATA_DIR` before launch to use a separate data directory for testing or portable runs. Do this before starting the app, and remember that Row-Bot will treat that folder as the active local data store for the whole process. ## After Installing Continue to [First Launch](/docs/getting-started/first-launch). You can finish optional setup later from the Setup Center and Settings. The standard desktop install stays on local loopback. To connect another trusted computer or phone, finish local setup first and then follow [Remote Access And Server Mode](/docs/operations/remote-access). For a headless or container deployment, read that guide before publishing a port or creating the first owner invitation.

## First Launch

Route: /docs/getting-started/first-launch

Answer "How should Row-Bot think?" to choose your first model, then finish the rest at your own pace in Setup Center.

# First Launch The first time you open Row-Bot, it asks one question: **How should Row-Bot think?** Your answer connects one working model, which becomes the default for new conversations. Everything else can wait for Setup Center. Every desktop, source, and official Docker install uses the same first launch. ## How Should Row-Bot Think? Pick one of three cards. You can add other providers later. | Choice | What it means | Tradeoffs | | --- | --- | --- | | **On this computer** | Private and free, with Ollama. | Needs Ollama and a downloaded model; larger models need more memory and disk space. | | **With my subscription** | ChatGPT, Claude or Grok. | Signs in to an account you already pay for and follows that provider's terms. | | **With an API key** | OpenAI, Anthropic, Google, OpenRouter and more, or your own endpoint. | Needs an API key and internet access; the provider bills per use and receives the prompts you send it. | - **On this computer** walks you through it: **Install Ollama** if it is missing, **Start Ollama** if it isn't running, **Download a model**, then **Pick a model** from the models on this computer. - **With my subscription** lets you choose ChatGPT, Claude or Grok and sign in. - **With an API key** lists the recommended providers first and the rest under **More providers**. Choose one, paste its key, and pick a model. Advanced users can add an OpenAI-compatible endpoint, such as LM Studio, vLLM, llama.cpp, LocalAI, LiteLLM or SGLang. After you pick a model, Row-Bot sends it a short test message. A local model can take a minute to load the first time. If the model doesn't answer, choose **Try again** or **Choose another model**. If Row-Bot finds Hermes Agent or OpenClaw on this computer, it then offers to bring that assistant's memories, persona and skills along. Choose **Import from ...** or **Not now**; you can import later from **Settings Data**. Choose **Set up later** to look around first. Home then stops opening setup for the rest of this window, and Overview reminds you to choose how Row-Bot thinks. ## Setup Center Setup Center is where you finish the rest whenever you like. Open it from Overview's **Finish setup** card or from the command palette (**Setup Center**). 1. **What would you like to use?** Pick any of: chat assistant, research and documents, workflow automation, designs, code, messaging channels, or local and private AI. Recommended areas move to the top. 2. **Continue setup** lists each area with one button that goes where it is set up: Models, Knowledge, Workflows, Designer (starts a design in a chat), Developer, Channels, Accounts, Tools & Skills, MCP & Plugins, Voice, and a Final Check that opens Monitor to run the diagnosis. 3. Each area's **** menu marks it done or skips it. Areas Row-Bot can check, such as a connected model, tick themselves. The links at the bottom import from Hermes or OpenClaw, or set up private knowledge search. ## Install Private Knowledge Search Row-Bot can search your memories and documents by meaning with **Mixedbread Embed Large v1**, a separate local model for semantic memory and document search. It is a 675 MB download from Hugging Face and is not the model that writes chat responses. Install it from **Settings Documents** (Setup Center's **Set up private knowledge embeddings** link goes there). Under **Index health and maintenance**, **Search model files** offers **Download** for the first install, **Retry** to load an already cached model again, and **Repair** to replace a damaged download. Files go into Row-Bot's private cache; documents and memories are not uploaded, and normal recall stays offline after the download. Docker keeps this cache in its named `/data` volume. Until it is installed, Row-Bot continues with bounded lexical and graph fallback, so chat and memory do not stop working. Rebuild the document and memory vectors after changing the search model. ## Privacy, Cost, And Credentials Local model runs can stay on your computer. Hosted, subscription, realtime voice, web search, account, channel, MCP, and plugin features can contact outside services when configured and used. Enter provider keys and account sign-ins only through first launch or the relevant Settings page. Cost depends on the provider or service you choose. ## Troubleshooting - If Row-Bot can't find Ollama, install or start it, then check again from the same card. - If a provider connects but no models appear, refresh it in Settings Providers. - If the test message fails, try again or choose another model; a small local model may need a minute to load. - If the private knowledge model download fails, check internet access to Hugging Face and try **Download** again later. - If you skipped setup, open Setup Center from Overview or the command palette.

## Models And Providers

Route: /docs/configuration/models-and-providers

Configure providers, discover models, choose defaults, and pin quick model choices.

# Models And Providers Providers are where models come from. Models are the specific choices you use in conversations, workflows, designs, code folders, voice, and other Row-Bot surfaces. To understand what leaves the app, when tools run, and where approvals fit, read [How A Request Runs](/docs/concepts/request-lifecycle). ## Recommended Setup Order 1. Open Settings -> Providers (or answer **How should Row-Bot think?** at first launch). 2. Connect one provider path. 3. Refresh provider health. 4. Open Settings -> Models. 5. Refresh the catalog. 6. Choose the **Brain** model, and Vision, Image and Video models if you want those jobs. 7. Pin the models you use most. 8. Test in a chat. ## Choosing A Provider Path Use local Ollama for local-first privacy and no provider billing. Use an API provider for strong hosted models. Use a subscription account when you already have the supported provider account. Use a custom endpoint for advanced local or self-hosted OpenAI-compatible runtimes. ## Choosing Models Keep at least one everyday chat model and one stronger tool-capable model pinned. For code folders, designs, workflows, and complex tools, choose a model that can handle tool calls and enough context. For models with exact reasoning capabilities, Row-Bot exposes a per-thread [Reasoning control](/docs/chat/reasoning-controls) with only the efforts, toggle, or token-budget choices that model supports. Provider default remains available as the compatibility-safe option. For Ollama, Row-Bot prefers explicit tool-calling metadata, then the daemon's reported capability list. A reported `tools` capability enables agent use even for a model family newer than Row-Bot's maintained fallback catalogue; an explicit capability list that omits `tools` remains authoritative. Family fallbacks are used only when the daemon provides no capability metadata. ## Troubleshooting - If a provider connects but has no models, refresh Models. - If a model fails with tools, choose a tool-capable model. - If an Ollama model is unexpectedly excluded from agent use, refresh the local catalogue and inspect the capabilities reported by the installed Ollama daemon. - If a custom endpoint fails, check base URL, model name, API compatibility, and context window.

## Row-Bot Interface

Route: /docs/app-shell/navigation

Find your way around Row-Bot: the sidebar, Home, a conversation and its details card, panels, the terminal, the command palette and Settings.

# Row-Bot Interface Row-Bot has one window with three parts: the **sidebar** on the left, the main area in the middle (Home, a conversation, the conversation library, or Settings), and, in a conversation, the **panels** for its designs and code folders on the right, with the **terminal** docked underneath when you open it. ## The Sidebar - The top row holds **Home** (the house icon, also the Row-Bot logo), which opens the Home tabs: Overview, Workflows, Knowledge, Monitor and Insights (see [Home](/docs/home/)); **Search**, which opens the command palette; and the button that collapses the sidebar. - **New chat** starts a conversation. Its **** starts a chat with one of your favourite agents, or **All agents...** opens the Agents dialog. - **Conversations** lists your conversations, pinned ones first. The icons beside the heading filter by type: All, Chats, Designs, Code and Workflows. A conversation's row shows when it is replying, waiting for approval, or has agents working. - A row's pin button keeps it at the top; its **** menu renames it, exports it as Markdown or PDF, or deletes it. - **Show all** lists every conversation in the sidebar; **Conversation library** opens the full library to search, filter, export and clean up. See [Chat](/docs/chat/#find-export-and-clean-up). - **Agents**, below the conversations, shows your favourite agent profiles as icons (one click starts a chat with that agent) and **All agents**, which opens the Agents dialog with the whole profile library. See [Profiles, Goals, And Agents](/docs/profiles-goals-agents/). - The **attention indicator** at the bottom appears only when something needs you: waiting approvals (with **Approve** and **Deny** in place), problems (opens Monitor), or an available update (opens Updates). - **Buddy** sits above **Settings**. In the native Windows and macOS apps, drag Buddy itself out of the sidebar to undock its desktop overlay. See [Settings: Buddy](/docs/settings/buddy). - **Settings** opens the settings pages. See [Settings](/docs/settings/). Collapse the sidebar to a narrow rail with its toggle; the rail keeps New chat, Home, Agents, the attention indicator, Settings and Buddy as icons. ## A Conversation The conversation's header shows its name (click it to rename), **Find** to search the conversation, **Share or export**, and **Conversation details**. The **details card** beside the conversation lists what the conversation is **Working on** (its designs and code folders), its **Outputs**, its **Goal**, and the **Agents** it started. A delegated agent's own conversation shows the way back to its parent in the header. See [Chat](/docs/chat/). ## Panels And The Terminal A design opens in a **Design panel** and a code folder in a **Developer panel**, beside the conversation. Open them from the details card or the header's **Open panel** menu; drag the divider to resize. See [Designs](/docs/designer/) and [Code Folders](/docs/developer/). In the desktop app, the **terminal** opens docked under the conversation with the terminal button, **Open panel Interactive terminal**, or Ctrl+`. See [Tools, Approvals, And Terminal](/docs/chat/tools-approvals-and-terminal#interactive-terminal). ## The Command Palette Press K (Ctrl+K on Windows and Linux), or the search button in the sidebar's top row, to open the command palette. Type what you want in your own words: - **Requests** such as "connect a model", "phone", "connect telegram", "add mcp server", "install a skill", "change model", "connect gmail" or "back up" go straight to the exact place it is done. Words can come in any order. - **Conversations and messages** are found by title and by words in their messages. - **Commands** open Home tabs, the conversation library, Setup Center, Settings or the Agents dialog, and toggle the sidebar, the details card or the terminal. Each shows its shortcut. - **Settings pages and rows** open at the exact row. - **Switches** flip a setting in place, showing its current state, with **Undo** afterwards: for example "turn off shell", "turn on developer tools", the browser tool, Dream Cycle, learning new skills, or the habit tracker. "dark mode", "light mode" and "follow the system appearance" change this device's appearance. - **Agents** and **workflows**: "Chat with ..." starts a chat with an agent; "Run ..." runs a workflow. Enter runs the top result only when it matches what you typed; otherwise choose a result with the arrow keys. ## Keyboard Shortcuts | Shortcut | Action | | --- | --- | | K / Ctrl+K | Open the command palette | | O / Ctrl+Shift+O | New chat | | . / Ctrl+. | Show or hide the conversation details card | | Ctrl+` | Open or close the terminal | | / | Find a setting, on a Settings page | | Enter / Ctrl+Enter | Approve the focused approval | | Z and Z / Ctrl+Z and Ctrl+Shift+Z | Undo and redo in the Design panel | | Esc | Close the palette, a menu or a dialog | ## On A Phone Or Tablet On a narrow screen, the sidebar becomes a drawer opened from the top bar, panels open as full-screen sheets, and the conversation header's **** menu holds Find, Conversation details, Share or export, the panels and the command palette. See [Phone And Native Desktop](/docs/mobile-native/). ## What Is Saved Conversations, their names, models, profiles, workflows, knowledge, settings and logs are saved in the local Row-Bot data directory. Panel sizes, the terminal's height, favourite agents and appearance are remembered on each device. ## Troubleshooting - If a conversation is missing from the sidebar, use **Show all**, the type filter, or the conversation library's search. - If a panel looks stale, close and reopen it from the details card. - If a button is disabled, check the related Settings page and Monitor for readiness messages.

## Home: Overview

Route: /docs/home/overview

Start from Home Overview: ask Row-Bot anything, answer what needs you, see live status, and pick up recent work.

# Home: Overview Overview is the first page Row-Bot opens. It answers three questions at a glance: what needs you, what is running, and where you left off. Everything on it opens the place where the work actually lives, so you can act without hunting through tabs. ## Where To Find It Choose **Home** in the sidebar. Overview is the first of the five Home tabs: Overview, Workflows, Knowledge, Monitor and Insights. Until a default model exists, opening Row-Bot opens [first launch](/docs/getting-started/first-launch) instead. ## Ask Box And Quick Starts - **Ask Row-Bot anything, or describe what to make...** starts a new chat with your text. Use **Add files in a new chat** to attach files first, then **Start chat**. - **New design** and **New code folder** set up a design or a code folder and open it in a new chat with its panel beside it. See [Designs](/docs/designer/) and [Code Folders](/docs/developer/). - **New workflow** opens the workflow editor on the Workflows tab. - **Continue "..."** reopens your most recent conversation. ## Needs You **Needs you** comes first and counts what is waiting for you: - **Approvals** from any conversation, workflow or delegated agent show **Approve** and **Deny** right in the card, with how long each has waited. Open the conversation or workflow when you want the full context first. - **Interrupted agent work** offers **Resume** or **Dismiss**. - **Failed workflows** offer **Run ... again**. - **Health checks that need attention** carry their one-click fix, such as restarting a stopped channel, reconnecting an account, choosing a model, or checking again. See [Monitor](/docs/monitor/). - **Finish setup** appears while Setup Center still has steps; hide the reminder once you are done. When nothing waits, the card says you're all caught up. ## Live Status The status strip shows four tiles, each opening its tab: | Tile | What it shows | | --- | --- | | **Agents** | Agents working in your conversations right now. | | **Workflows** | What is scheduled next and today's runs. | | **Memory** | A small picture of your most connected memories and what is new this week. | | **Health** | The latest Monitor checks, which run by themselves. | ## Recent Work - **Continue where you left off** shows recent conversations as cards, with pinned ones first and a mark on any that is replying, waiting for you, or was interrupted. - **Since yesterday evening** is a timeline of what ran: workflow runs, finished agents, Dream Cycle runs and new memories. - **Learned this week** lists subjects Row-Bot saved to memory, and a current insight with **Open Insights**. See [Home: Insights](/docs/home/insights). ## What Is Saved Overview stores nothing of its own. It reads your conversations, workflows, approvals, memories and Monitor checks, and every action runs through the same reviewed command as the place it comes from. ## Troubleshooting - If Overview opens Setup instead, finish choosing a model or choose **Set up later**. - If a tile reads **Unavailable**, refresh the page; Monitor shows whether the service behind it is healthy. - Approving from Overview is the same decision as approving in the conversation. Open the conversation first when the request is not obvious from its title.

## Chat

Route: /docs/chat/

Use Row-Bot conversations: the composer, models, attachments, tools, approvals in place, the details card, names, and the conversation library.

# Chat A conversation is the main place to work with Row-Bot. It can stay a simple chat, or use memory, documents, tools, approvals, attachments, skills, voice, agents, goals, designs and code folders. ## Start A Conversation - Choose **New chat** in the sidebar (O / Ctrl+Shift+O), or type into the Ask box on [Home Overview](/docs/home/overview). - Use the **** beside **New chat**, a favourite agent in the sidebar, or **Chat with ...** in the command palette to start with an agent profile. A new conversation is named from the first words of your first message straight away, then gets a short name of its own after the first reply. Click the name in the header to rename it; a name you choose is kept. ## The Conversation - **Header**: the conversation's name, **Find** to search it, **Share or export**, and **Conversation details** (. / Ctrl+.). - **Transcript**: your messages and Row-Bot's replies, each with a small speaker mark (your Buddy's picture for Row-Bot), plus tool steps, how long Row-Bot thought, charts, images, generated media, approval cards and the agents a turn started. - **Composer**: type the next request. Enter sends and Shift+Enter adds a line. **Send** becomes **Stop** while Row-Bot replies. ## Composer Controls - **Model** chooses the model for this conversation. See [Model Picker](/docs/chat/model-picker). - **Thinking** appears beside the model only when that exact model exposes supported reasoning choices. See [Reasoning Controls](/docs/chat/reasoning-controls). - **Approvals** sets how actions are reviewed in this conversation: **Ask** (approve each consequential action), **Block**, or **Auto**. - The **+** menu (**Add files and more**) holds **Attach file**, **Add resource...** (a design or code folder), **Skills...**, **Agent profile**, and **Mode**: **Agent** uses tools and agents, **Chat only** answers without tools. - **Skill chips** above the composer show the skills in use, including default ones; removes one from this conversation. - **Voice** buttons use Dictate or Talk when voice is set up. ## Attachments Attach files with **+ Attach file**, by dropping them on the conversation, or by pasting. Each file shows as a small tile above the text: pictures show a preview, other files their type, name and size. Tiles show upload progress and offer **Retry** if an upload fails. You can keep adding files while others upload; **Send** waits for them. Click a tile to preview it, or to remove it. ## The Details Card **Conversation details** sits beside the conversation on a wide screen. It shows: - **Working on**: the designs and code folders bound to this conversation. Choose one to open its panel, or unbind it from its **** menu. **Add resource** binds another. - **Outputs**: images and videos the conversation made, linked to where they appear. - **Goal**: the conversation's goal and its progress, or **Set a goal**. See [Profiles, Goals, And Agents](/docs/profiles-goals-agents/). - **Agents**: the agents this conversation started, each with its status, **Message** and **Stop**. Its **** menu offers **Manage conversation**, **Manage browser** and **Delete conversation**. ## Approvals When Row-Bot needs your decision, an approval card appears in the conversation with **Approve** and **Deny**. Approvals wait until you answer, and they also show in the sidebar's attention indicator, on Overview and in Buddy, so you can decide from wherever you are. See [Tools, Approvals, And Terminal](/docs/chat/tools-approvals-and-terminal). ## Beginner Workflow 1. Start a new chat. 2. Choose a model that is ready for chat. 3. Ask one clear question or task. 4. Attach files only when the task needs them. 5. Review any approval before allowing Row-Bot to act. ## Power Workflow 1. Start a chat with the agent profile for the kind of work. 2. Pick a stronger tool-capable model. 3. Bind the design or code folder you are working on. 4. Set a goal when the work needs several turns. 5. Let Row-Bot use tools, but approve file writes, browser actions, shell commands, account actions, channel sends, MCP calls, and plugin tools deliberately. ## Continue From The Buddy Overlay In the native Windows and macOS apps, drag Buddy from the sidebar onto the desktop for a compact view of the selected conversation. Its composer shares that conversation's saved draft, model, tools, approval mode, and active response. Simple approvals can appear in the overlay; open the full thread for complex approvals, attachments, transcript history, tool traces, model controls, Talk, or Dictate. See [Settings: Buddy](/docs/settings/buddy). ## Find, Export, And Clean Up - **Share or export** in the header opens the conversation's actions: change its **Name**, **Pin** it to the top of the sidebar, or **Export** it as **Markdown** or **PDF**, then **Save export** to choose where the file goes. - A sidebar row's **** menu exports or deletes one conversation too. - **Conversation library**, at the end of the sidebar's list, lists every conversation. In the library, **Search titles and messages**, filter by type (All, Chats, Designs, Code, Workflows), and select conversations. The selection bar at the top pins or unpins them, exports them as Markdown or PDF, or deletes them; **Clear selection** (Esc) starts again. Exported files go to Row-Bot's Exports folder; **Show in folder** opens it. Deleting a conversation stops its running work, then removes its messages, drafts, attachments, media, tool sessions, workflow state and agent state after you confirm. Bound designs and code folders remain. Developer worktrees with changes and sandboxes with unimported work are kept and reported, so deletion does not discard source work. Export anything you need before confirming because deletion cannot be undone. ## What Is Saved Conversation names, messages, model choices, per-model reasoning selections, approval mode, profile, attachments copied into Row-Bot-managed storage, and tool results are saved locally. External providers receive prompt content only when you choose their models. ## Privacy And Safety Local model runs can stay on your machine. Hosted models, web search, browser actions, account tools, MCP servers, plugins, and channels can send data outside the app when configured and used. Row-Bot should not insert secrets into chat; enter credentials only through Settings or the provider's sign-in flow. ## Troubleshooting - If Row-Bot cannot answer with the current model, choose a more capable model in the picker. - If the Thinking control is missing or a saved choice resets, read [Reasoning Controls](/docs/chat/reasoning-controls#troubleshooting). - If attachments are ignored, check that their tiles finished uploading and that they are relevant to the prompt. - If a tool is unavailable, check the composer's **Mode**, the conversation's agent profile, approval mode, and Settings. - If an external tool is enabled but not selected automatically, use a more specific request or review [Progressive Tools And Skills](/docs/guides/progressive-tools-and-skills). - If a reply is stuck, press **Stop**, then retry with a narrower prompt.

## Model Picker

Route: /docs/chat/model-picker

Choose a conversation's model from one list across local, subscription, pay-per-use and custom providers, and troubleshoot missing models.

# Model Picker The model picker in the composer chooses which model answers in the current conversation. It is where provider setup becomes visible in day-to-day use. ## Where Models Come From Models appear after Row-Bot discovers them from a connected provider or local runtime: - Local Ollama models come from the Ollama service running on your computer. - Subscription models come from a ChatGPT, Claude or Grok account you signed in to. - Pay-per-use models come from a provider connected with an API key in Settings Providers. - Custom endpoint models come from the compatible endpoint details you provide. - Models you pin in Settings Models appear first. The same list is used everywhere a model is chosen: the composer, and the Brain, Vision, Image and Video pickers in Settings Models. ## Using The Picker - **Search models** narrows long lists; **Recent** shows the models you used last. - Models are grouped by provider, each marked **Connected** or **Not connected**, and tagged with how they are paid for: **Subscription**, **Pay per use**, **Credits**, or **Local - free** (runs on this device). - A model that can't be used right now, for example because its provider is disconnected, shows as **Unavailable** with the reason. - **Chat only** marks models that can't use tools. - The choice applies to this conversation only; it doesn't change your default model in Settings Models. ## Choosing Well Use a smaller or local model for quick private chats. Use a stronger tool-capable model for workflows, designs, code folders, long context, or multi-step tool use. For local and self-hosted endpoints, prefer a context window large enough for Row-Bot's instructions and tool schemas. Reasoning choices belong to the exact provider-qualified model, not just its display name. After choosing a supported model, use the conditional Thinking control described in [Reasoning Controls](/docs/chat/reasoning-controls). ## Pinning Models Open **Settings Models**, find a model in the **Catalog**, and pin the ones you use often. Pinning doesn't create a new model; it puts an existing one at the top of every picker. ## Troubleshooting - If the picker says there are no models yet, choose how Row-Bot thinks in Setup, or connect a provider in Settings Providers. - If a model is missing, refresh the provider in Settings Providers, then refresh the catalog in Settings Models. - If a model shows as unavailable, read its reason; reconnecting its provider usually fixes it. - If a model appears but fails with tools, choose a model that isn't marked **Chat only**. - If custom endpoint models look duplicated, use the provider-qualified name.

## Reasoning Controls

Route: /docs/chat/reasoning-controls

Choose provider-aware reasoning effort, thinking toggles, and token budgets for the active Row-Bot model.

# Reasoning Controls Reasoning controls let you change how much supported models reason before answering without switching models. Row-Bot shows only choices that the exact provider-qualified model reports or that Row-Bot can identify from a maintained model route. ## Where To Find The Control - In every conversation, including those working on a design or code folder, the **Thinking** control sits beside the model picker in the composer. In narrow windows, the composer progressively compacts labels while keeping each icon paired with its own menu. - On a phone, the composer's controls fold into a smaller set of buttons; Thinking still appears next to Model when the selected model supports it. - In Chat or a connected messaging channel, use `/reasoning` to inspect the current choice and the valid choices for the active model. The control is intentionally absent when Row-Bot does not have exact, actionable reasoning capability data for the selected model. A generic model name or a provider-wide assumption is not enough. ## What The Choices Mean | Choice | Effect | | --- | --- | | Provider default | Sends no per-thread reasoning override and lets the provider or endpoint choose its normal behavior. This is the safest compatibility choice. | | Low, Medium, High, XHigh, or another effort | Requests one of the exact effort levels supported by the selected model. The list varies by model. | | On or Off | Enables or disables thinking only when that model exposes a true toggle. Models with mandatory reasoning do not offer Off. | | Token budget | Sets a positive reasoning-token budget within the minimum and maximum reported for that model. This appears only for budget-capable models. | More reasoning can increase latency and provider token usage. It can help with planning, coding, analysis, and multi-step tool use, but a higher setting is not automatically better for every request. ## Scope And Persistence A selection is saved locally for one thread and one exact provider-qualified model. Switching models does not apply an incompatible value to the new model. If you return to a model in the same thread, Row-Bot can restore that model's valid saved choice. Changing the global default model does not rewrite existing thread choices. Conversations working on designs and code folders use the same thread-scoped behavior as any other conversation. ## Slash And Channel Commands Run `/reasoning` with no argument to show the active setting and valid choices. Supported examples include: ```text /reasoning high /reasoning default /reasoning on /reasoning off /reasoning budget 4096 ``` Row-Bot validates the command against the active model. Unsupported efforts, toggles, or budgets are rejected without replacing the previous valid selection. Messaging channels use their conversation's active thread and model, so the setting remains isolated from unrelated chats. ## Providers And Custom Endpoints Row-Bot maps native reasoning controls for supported OpenAI and Codex, Anthropic and Claude Subscription, xAI, Google, Ollama and Ollama Cloud, OpenRouter, and compatible endpoint routes when exact capability data is available. Available models and choices can change as provider catalogs change; the control itself is the authoritative list for the selected model. For a custom OpenAI-compatible endpoint, open **Settings -> Providers**, edit or add the endpoint, then expand **Advanced -> Reasoning**. Keep **Reasoning mode** on Auto unless the endpoint's metadata is missing or wrong. Thinking budget, returned reasoning content, replay, and extra request JSON are advanced compatibility settings that apply to every model exposed by that endpoint. Enable replay only when the endpoint explicitly supports receiving preserved reasoning history. ## Responses And Compatibility Fallback When a provider returns reasoning content, Row-Bot can stream it separately and retain it as a collapsed **Thinking** section instead of mixing it into the final answer. If a provider rejects a valid-looking explicit reasoning choice before returning response content, Row-Bot retries once with Provider default, clears the rejected saved override, and shows a notice. Authentication failures, rate limits, timeouts, cancellations, and server failures are not silently retried as reasoning compatibility problems. ## Privacy And Safety The selected control is stored locally with thread settings, but hosted providers receive the resulting reasoning parameter and may bill for additional tokens. Reasoning content returned by a provider can contain sensitive intermediate material. Do not enable reasoning replay for a custom endpoint unless you trust that endpoint and understand its message format. ## Troubleshooting - **The control is missing:** confirm the active model is provider-qualified, refresh Providers and Models, and check that its catalog entry exposes exact reasoning capabilities. - **A choice disappeared after switching models:** choices are model-specific; inspect the new model's menu or run `/reasoning`. - **A saved choice reset to Provider default:** the provider rejected it or refreshed capability metadata no longer supports it. Read the notice, refresh the catalog, and choose from the current list. - **A custom endpoint returns reasoning in the wrong place:** review its Advanced Reasoning settings, especially returned reasoning content and replay. Keep replay off unless required. - **High reasoning is slow or expensive:** choose a lower supported effort, a smaller budget, or Provider default.

## Tools, Approvals, And Terminal

Route: /docs/chat/tools-approvals-and-terminal

Understand Row-Bot tools and tool steps, approvals that wait until you answer, and the interactive terminal docked under a conversation.

# Tools, Approvals, And Terminal Tools are actions Row-Bot can take beyond writing text. They can search documents, read files, use a browser, run commands, work on designs and code folders, inspect knowledge, call MCP servers, use plugins, or send through channels. ## Progressive Discovery With the recommended Auto capability-loading mode, core tools remain available under the active profile while Row-Bot searches enabled MCP, plugin, Custom Tool, and channel tools only when a request needs them. This keeps large external schema catalogs out of unrelated prompts. Discovery does not widen permissions. The selected provider, Agent Profile, workspace policy, approval mode, and execution budget still apply. Live and reopened traces group results under the real integration name. See [Progressive Tools And Skills](/docs/guides/progressive-tools-and-skills) for the complete flow and the eager compatibility option. ## How Tool Use Appears When Row-Bot uses tools, the turn shows a tool step: what Row-Bot tried, what came back, and whether more action is needed. Open it for each call's details. Each result also records whether approval was needed, given or denied, so Row-Bot reports it truthfully. A denied action shows as skipped, never as done, and Row-Bot is told that it did not run. ## Approvals Approval cards appear when an action can change local files, run commands, contact external systems, use accounts, start servers, send messages, call MCP or plugin tools, or do something else that deserves review. The composer's **Approvals** control sets the mode for the conversation: **Ask**, **Block** or **Auto**. **Approvals wait until you answer.** They no longer expire after a set time; a workflow step that sets its own time limit still honours it. Each one shows how long it has been waiting. **Answer them where you are.** Every waiting approval, from this conversation, another conversation, a workflow or a delegated agent, shows with **Approve** and **Deny** in place: - in its conversation, as an approval card; - in the sidebar's attention indicator ("1 approval is waiting"), which opens the list; - in Overview's **Needs you** card; - in Buddy, including the desktop overlay; - in the channel a workflow or channel-started task came from, when that channel supports approvals. A floating notice also says when a new one arrives. An answer in one place resolves it everywhere. When a delegated agent needs approval, its parent conversation shows it too. The card may include a short reason written by the model, but Row-Bot's approval policy still decides whether the action is blocked, allowed, or waiting for you. Before approving, check: - What action is being requested. - What file, folder, account, channel, server, or provider is involved. - Whether the action can send data outside the app. - Whether the proposed command or file change matches your request. - Whether denying is safer until you inspect settings. Approving lets Row-Bot continue that action. Denying stops that action and returns control to the conversation. ## Interactive Terminal In the Row-Bot desktop app, a conversation can open a real interactive terminal docked at the bottom of the window. Open or close it with the terminal button in the header, **Open panel Interactive terminal**, the command palette (**Toggle terminal**), or Ctrl+`. - Type straight into it: colours, line editing, command history (the arrow keys) and Tab completion work as in your own terminal, and interactive programs answer. - **Stop** (Ctrl+C) interrupts the running command, **Clear** clears the screen, **Open in your terminal** opens your own terminal app, and **Close** hides the dock. Closing and reopening keeps the session and its scrollback. - Drag the divider to resize it; the height is remembered on this device. In a phone-sized window it opens as a full-screen sheet instead. - After the shell exits, **Start again** opens a new one. The terminal is for you: what you type runs as your own commands on this computer. Commands Row-Bot runs as tools still follow workspace boundaries, tool availability and approval mode, and appear in the conversation's tool steps. The terminal needs the Row-Bot desktop app: in a browser on this computer the dock says so, and other devices don't offer it. ## Common Workflow 1. Ask for a task that may need tools. 2. Let Row-Bot explain the intended action. 3. Review the approval card, wherever you see it. 4. Approve only if the action matches your intent. 5. Open the tool step to check the result. 6. Ask Row-Bot to summarize what changed. ## Troubleshooting - If no approval appears, the action may be read-only, allowed by the conversation's **Auto** mode, or blocked before running. - If a tool result is confusing, ask Row-Bot to explain the last tool call. - If the terminal says it needs the Row-Bot desktop app, open the conversation there; the button is not offered on other devices. - If you denied by mistake, ask Row-Bot to try again and review the next approval.

## Progressive Tools And Skills

Route: /docs/guides/progressive-tools-and-skills

Understand how Row-Bot discovers external tools and task-specific skills without loading every capability into every prompt.

# Progressive Tools And Skills Row-Bot can keep a large extension library enabled without sending every external tool schema and every skill instruction to the model on each turn. By default, it progressively discovers the capabilities that match the current request. ## How Tool Discovery Works Core Row-Bot tools that are permitted by the active Agent Profile remain available normally. Enabled external tools from MCP servers, plugins, Custom Tools, and channels are kept in a searchable catalog. In **Auto-select external tools** mode, Row-Bot: 1. Searches the enabled external-tool catalog when the request needs an integration. 2. Selects the matching tool without changing its underlying identity or provider contract. 3. Applies the same profile filtering, workspace policy, execution budget, prompt-injection handling, and approval rules as a directly bound tool. 4. Shows the real integration name in live and reopened tool traces rather than a generic invocation wrapper. Discovery does not enable disabled integrations, add credentials, or bypass an approval. If two capabilities conflict with a core tool name, Row-Bot keeps the core boundary intact. ## How Skill Discovery Works Skills are instruction packages, not executable permissions. Enabled skills can be searched when a request clearly matches their purpose. Row-Bot may then load the relevant skill for that task and show a **Using _skill_** receipt plus an active skill chip. Automatically loaded skills are scoped to the current task. Reopening the task restores its active skills, while a different parent task or child Agent keeps independent activation state. Up to five automatically selected skills are retained per task; pinned skills are managed separately. Loading an already active skill is a no-op and should not create a duplicate receipt. ## Parent And Child Boundaries Each child Agent receives its own profile, tool allow-list, approval mode, workspace policy, execution budget, and skill activation state. A child can use the skill discovery bridges when its runtime permits them, but loading a skill cannot grant a tool that its profile denies. Child skill choices also do not leak into the parent or a sibling child. Use **Row-Bot Status** inside a child run when you need to distinguish globally configured tools from the tool groups actually bound to that child. ## Auto And Compatibility Modes Open **Settings -> Tools -> Capability loading**: - **Auto-select external tools (recommended)** keeps the prompt compact and discovers enabled external integrations as needed. - **Load all external tools** is the eager compatibility mode. It sends every enabled external tool schema that survives the same provider and policy filters, which can use substantially more model context. Changing this setting does not alter tool enablement, credentials, approvals, or saved Agent Profiles. ## What You See In Chat - Skill discovery can show `skill_search`, followed by one **Using _skill_** receipt after a successful load. - External discovery can show a search step followed by a result grouped under the real integration name. - Reopened transcripts preserve the real external-tool label. - Direct tools in eager mode keep their existing labels and approval behavior. ## Troubleshooting - If no matching tool is found, confirm the integration is installed, enabled, healthy, and allowed by the active profile. - If a skill is not selected, confirm it is enabled and make the requested task specific enough to match its purpose. - If a child can see a tool in global status but cannot call it, inspect the child's profile-selected tool groups. Global availability is not the same as runtime access. - If a provider rejects a tool schema, try a compatible tool-capable model; discovery does not weaken provider schema validation. - Use eager mode temporarily when diagnosing an older provider or integration that expects direct external-tool binding. See [Tools Settings](/docs/settings/tools), [Skills Settings](/docs/settings/skills), [Skills Hub](/docs/skills/), and [Extensions And Trust Boundaries](/docs/concepts/extensions-and-trust) for related controls and safety boundaries.

## Profiles, Goals, And Agents

Route: /docs/profiles-goals-agents/

Manage agent profiles in the sidebar's Agents dialog, give a conversation a goal that keeps going until done, and follow the agents it delegates to.

# Profiles, Goals, And Agents These three features work together but solve different problems. An **agent profile** is a reusable role and policy. A **goal** is an objective a conversation keeps working toward, turn after turn. An **agent run** is one piece of delegated work, with its own conversation, status and result. For the mental model behind delegation and hand-offs, read [How Profiles, Goals, And Agents Work](/docs/concepts/profiles-goals-and-agents). ## Agent Profiles Open the **Agents** dialog with **All agents** in the sidebar's **Agents** section, **All agents...** under **New chat**'s ****, the Agents icon on the collapsed rail, or **Agent profiles** in the command palette. It lists every profile in groups (Everyday, Creative, Developer, your own profiles, and profiles from workspaces, plugins or imports), with **Search profiles** and a scope filter. Each profile row can: - **Pin to the sidebar**: pinned profiles become favourites in the sidebar's **Agents** section and under **New chat**'s ****, one click from a new chat; - **Start chat** with that profile; - **View**, **Duplicate**, and, for your own profiles, **Edit**, turn on or off, or **Delete**. **Create profile** describes a new role: its name, description and when to use it, instructions, capability (read only, write capable, or orchestrator), allowed tools and skills, context mode, workspace mode and approval mode. Keep the role narrow: describe the outcome, the sources it may use, and what it must not do. To use a profile in an existing conversation, choose it from the composer's **+ Agent profile**. Profiles do not grant capabilities by themselves. A selected tool must still be enabled, available, and allowed by the conversation's approvals. ## Goals A goal is useful when work needs more than one turn. Set it from the conversation's details card: under **Goal**, choose **Set a goal**, describe **What should this conversation achieve?**, and choose **Start goal**. Row-Bot starts at once and keeps going turn after turn; it stops for approvals and when it needs you. - **No turn limit by default.** Overnight and multi-day goals keep going while they make progress. Set an optional **Turn limit** or **Time limit (hours)** when you want one. A default turn limit for new goals is in Settings Models Limits for long work. - **Smart stop.** Row-Bot pauses the goal and says why when it stops making progress (no progress in the last 2 turns) or the same step fails 3 times. When a provider's usage limit stops a turn, the goal waits and continues by itself once the limit resets, if the provider says when. - **Survives restarts.** An active goal continues after Row-Bot restarts or the computer wakes, from its last saved turn, asking again for anything that needed approval. - **Visible usage.** The card shows the goal's state (Working, Continuing, Waiting for your approval, Paused, Needs you, Done or Stopped), the turn it is on ("Turn 7", or "Turn 7 of 30" with a limit), how long it has been running and the tokens it has used. - **Pause**, **Resume** and **Stop** control it. Stopping ends the goal; it does not delete the conversation or its evidence. Starting a new goal replaces the current one, and **Earlier goals** lists the ones before. ## Delegated Agents A conversation can delegate parts of its work to agents. Each agent works in its own conversation. - **In the transcript**, the turn that started agents shows a small stub for each one, with its icon, name and live status (working, waiting for you, done, failed or stopped). A failed agent's stub says why. Click a stub to open that agent's conversation. - **In the details card**, **Agents** lists them, live and failed ones first. **Message** sends an agent a note it reads at its next step; **Stop** ends its run. - **In an agent's own conversation**, the header leads back to the parent conversation. - **When an agent needs approval**, the request appears in the parent conversation and everywhere else approvals show. See [Tools, Approvals, And Terminal](/docs/chat/tools-approvals-and-terminal#approvals). - **If work was interrupted** before it finished, for example by a restart, the card and Overview's **Needs you** offer **Resume agent work** or **Dismiss agent work**. A completed agent's result is evidence for the parent, not automatic permission to write files, send messages, commit code, or publish. For coding work, a parent can assign an agent to an existing local folder. Row-Bot registers that folder as the agent's code folder and applies writer ownership to the resolved folder, not merely the process working directory. Agents assigned to distinct folders or worktrees can write in parallel; writers targeting the same folder wait and run serially. New runs also receive a checkpointed work budget. Settings Models **Limits for long work** shows the recommended steps per run, helper levels, helpers at a time, and the optional helper time limit. Extra agents wait in a first-in, first-out queue when capacity is full; changing a setting affects new runs, not work already in progress. If a run appears stuck, open its conversation to read its last update before stopping it. If several agents edit the same resource, give them distinct folders or worktrees, narrow their ownership, or run them sequentially. Keep consequential final actions with the parent and a human approval point. Row-Bot saves the parent's first continuation checkpoint when asynchronous delegation begins. If a child process outlives its launching shell or the app restarts, resume repairs only genuinely unanswered tool calls with an interruption marker, then wakes the parent from durable child results. It does not replay previously executed work. If a checkpoint cannot be repaired safely, the run remains stopped for review. Repeated model-and-tool states are detected before a run can loop indefinitely. The fourth identical no-progress state is blocked; a fifth ends the run cleanly with a durable reason. Reaching a configured work limit also finalizes the run instead of leaving it marked as active. ## What Is Saved Profiles, goals, agent runs, edges, progress, results, and evidence are stored in the active local data directory. Favourite profiles are remembered on each device. Provider prompts still follow the privacy terms of the model route selected for each run.

## Workflows

Route: /docs/guides/workflows

Create, run, schedule, edit, deliver, and troubleshoot Row-Bot workflows.

# Workflows Workflows are repeatable Row-Bot tasks. Use them for things you want to run again: a morning brief, inbox follow-up, document digest, research check, report draft, or channel delivery. A **reminder** is a simpler kind that just tells you something at a chosen time. For the execution model behind schedules, persistent runs, approvals, and delivery, read [How Background Workflows Run](/docs/concepts/background-workflows). ## Where To Find Workflows Choose **Home**, then the **Workflows** tab. Overview's **New workflow** quick start and the K command palette (**Run ...**) also reach them. ## The Workflow List - Each row shows the workflow's name, when it runs next, its last runs, and small tags for its steps, profile and approvals. **Auto approvals** stands out. - **Run** starts the workflow at once, in one click. The row then shows its live state and **Stop**. - **Enable** or **Disable** turns the workflow's schedule on or off. - **Edit** opens the workflow editor. - The row's **** menu opens **Run history**, **Edit workflow steps**, **Workflow settings** or the workflow's conversation, and can duplicate or delete the workflow. - **Search workflows**, the filter and **Select workflows** help with long lists. - **Delivery defaults** choose the external channels every workflow also sends results to, unless a workflow chooses its own. The app always shows the result. ## Create A Workflow 1. Choose **New workflow**. The editor opens on the page, as editing does. 2. Give it a name, an icon and a short description, and choose **Workflow** or **Reminder**. 3. Write the **Steps**: what Row-Bot should do first, then what next. Each step can use the previous step's output; drag steps, or use the arrow keys, to reorder them. 4. Choose **When it runs**: manually, at a date and time, or repeating (every day, weekdays, chosen days, every few hours or minutes, or a custom cron expression). 5. Choose **Send results to**: follow your defaults, choose channels only for this workflow, or keep results in the app. 6. Under **Model and approvals**, pick an **Agent profile** and **Approvals**: **Ask before actions**, **Block actions**, or **Auto, within the profile**. 7. Save, then **Run** it once and review the result. ## Advanced Workflows - **Edit workflow steps** opens the step editor, with a builder and a step graph. Besides prompts, steps can be conditions, approvals, running another workflow, delegating to an agent, waiting for agents, and notifications. - **Workflow settings** chooses the model, whether runs reuse one conversation, a concurrency group, and the trigger: manual and scheduled runs only, **When another workflow completes**, or **Webhook (HTTP POST)**. A webhook trigger creates a private secret when you save; download its configuration and keep it private. ## Approvals In Workflows An approval step, or an action that needs approval, pauses the run until you answer. It waits as long as needed unless that step sets its own time limit. Answer it from the sidebar's attention indicator, Overview's **Needs you**, the workflow's conversation, Buddy, or a channel that supports approvals; an answer in one place resolves it everywhere. While a run waits, that workflow's next scheduled run is skipped and says so. ## What Is Saved Workflow definitions, schedules, run history, delivery defaults, and status are local Row-Bot data. Results may be sent to channels only when configured and selected. ## Safety Avoid workflows that send messages, write files, start servers, or call external systems without a review point. Use approvals for actions that can affect files, accounts, channels, MCP servers, plugins, or external services. Remember that **Run** starts at once, including any delivery outside the app. ## Troubleshooting - If a workflow does not run on schedule, check that it is enabled and that an earlier run isn't still waiting for approval. - If delivery fails, check the selected channel and its Settings page. - If a run waits for approval, answer it from the attention indicator or Overview. - If results are too broad, split the workflow into smaller steps.

## Designs

Route: /docs/designer/

Create decks, documents, landing pages, app mockups and storyboards with Row-Bot, and edit, present, share and export them in a conversation's Design panel.

# Designs Row-Bot can make visual work with you: decks, documents, landing pages, app mockups and storyboards, with your brand's colours, fonts and logo, charts, and generated images or video. Each design belongs to a conversation and opens in that conversation's **Design panel**, beside the chat, so you can ask for changes and see them land. ## Start A Design - **Ask in a chat:** "Make a five-slide deck about our spring workshop." Row-Bot creates the design, and the finished design opens in the panel by itself when you are not typing. - **Home Overview New design** sets one up and opens it in a new chat. - **In a conversation**, open the composer's **+** menu, choose **Add resource...**, then **Design**. Choose **Create new** (the type, an optional name and brief, a template and canvas, and **Draft it now**) or **Open saved** to bring an existing design into this conversation. The conversation's details card lists it under **Working on**; choose it there, or from the header's **Open panel** menu, to reopen the panel. ## The Design Panel - **Preview** and **Edit** switch between looking and changing. In Edit, click anything on the page to change it, or ask Row-Bot about it; double-click text to edit it in place. Changes save as you go. - **Undo** and **Redo** step through your changes; **Versions** lists saved versions and restores one. - The **page strip** moves between pages, adds a page after the current one, or deletes one. - **Size** chooses the canvas (16:9 widescreen, 4:3, square, A4 document, 9:16 phone, or a custom size). **Device width** previews pages at desktop, tablet or phone width, and the zoom control fits or scales the page. - **Present** shows the design full screen. - **Share** publishes a link (with a QR code), sends to a channel, or prepares an X post. - **Export** saves the current saved version as **PDF**, **PNG**, **PowerPoint** (pictures for high fidelity, or editable text and shapes) or **HTML**, for all pages, the current page or a page range. - The **** menu opens the design command palette to search tools, pages and assets, imports a document, or duplicates the design. ## The Inspector **Inspector** opens three tabs: - **Selection** edits the chosen element: its text, font, size, weight, colour and alignment; an image's file, fit, crop, corners and description; width and borders; and what happens when it is clicked (go to a screen, toggle a state, play media). **Edit as CSS...** is there for exact control. - **Brand** sets the design's **Colours** (primary, secondary, accent, background and text), **Fonts** for headings and body from a font picker (recently used, bundled, and fonts on this computer), and the **Logo**: upload one, then choose its placement, size, spacing, which pages show it, and automatic or manual placement. Save the brand as a preset to reuse it in other designs. - **Library** inserts **Blocks**, **Your images**, saved **Styles** and **Interactions**. **Review** checks the design and lists what to fix. ## Brand And Media Providers Generated images or videos call the image and video models chosen in **Settings Models**. Local design edits stay in local project data. For xAI API image models, Row-Bot uses the provider's live image-generation catalog instead of guessing from the chat catalog. It exposes only the formats, aspect ratios, resolutions, and quality choices reported for that model. If xAI rejects an optional quality field, Row-Bot retries once without it; timeouts and other failures do not trigger an automatic second generation that could duplicate cost. ## What Is Saved Designs, their versions and brand presets are local Row-Bot data. Deleting the conversation keeps the design; open it from another conversation with **Add resource... Open saved**. ## Troubleshooting - If a first draft is thin, give a clearer goal, audience and length, and attach references. - If an export looks wrong, preview each page first; exports use the last saved version. - If image generation is unavailable, choose an image model in Settings Models and check its provider. - If a change says the design changed meanwhile, let the panel reload the current version and try again.

## Code Folders

Route: /docs/developer/

Work on a folder or repository from a conversation: bind or clone it, review changes, run checks and use Git in the Developer panel.

# Code Folders A code folder is a folder or Git repository on this computer that a conversation works on. Bind one to a conversation and Row-Bot can read its files, propose and make edits, run checks and commands you approve, and help with Git. The folder opens in the conversation's **Developer panel**, beside the chat. ## Bring A Folder Into A Conversation **Ask in the chat.** Row-Bot binds folders by itself when you ask: - "Use my Hello-World folder" binds a code folder you already use, by its name. If several names match, a **Which code folder?** card lets you choose; if none does, a **Use an existing folder** card asks you to choose the folder on this computer. - "Clone https://github.com/octocat/Hello-World" shows a **Clone a repository** card with the address. Choose **Choose where** to pick the parent folder, and Row-Bot clones it there. An interrupted clone is kept for inspection and offers **Check clone status**; it is never cloned again automatically. Either way, the conversation shows a **Using code folder ...** card with **Open** and **Undo**. The turn pauses while a card waits for you and continues in the folder once it is bound. Row-Bot never chooses a path on its own: you choose every folder in the picker. **Or set it up yourself.** Open the composer's **+** menu, choose **Add resource...**, then **Code folder**: | Folder setup | What it does | | --- | --- | | **New draft in the configured workspace** | Creates an empty folder under Drafts in your workspace folder. | | **Register an existing folder** | Saves an existing folder's name and location; its files and Git state are unchanged. | | **Create a new empty folder** | Creates one empty folder under a parent you choose. | | **Clone a repository** | Downloads a repository into a new folder under a parent you choose. | **Open saved** brings a code folder you used before into this conversation. Home Overview's **New code folder** does the same in a new chat. Choosing a folder on this computer needs the Row-Bot desktop app. The conversation's details card lists the folder under **Working on**. Its **** menu, or **Undo** on the card, unbinds it from the conversation; the files stay on disk. The Agents tool can also assign a child directly to an existing local folder. Row-Bot validates the path and registers it as that child's code folder; a folder assignment and a pre-existing workspace ID are alternative ways to select the workspace, not settings to combine. ## The Developer Panel - **Changes** lists changed files with their diffs (unified or split), and the agent's own change sets, each marked reviewed or not reviewed, with **Undo** or **Ask Row-Bot to undo** in its **** menu. - **Files** browses the folder and previews files. - **Run** lists the checks Row-Bot detected (such as tests and linters) with their last result, runs a command you type (such as `npm test`) as a reviewed command, shows running processes with their output, and stops them. Sandbox changes waiting to be imported appear here too. - **Git** switches or creates branches, commits with a suggested message, pushes (which asks first), opens a pull request through the GitHub command-line tool, and creates a separate worktree for this conversation. - **Safety boundaries** shows what is allowed, what asks first and what is blocked here, where commands run, and whether the sandbox has network access. - **Custom tools** turns the folder's commands into a reusable tool with the Custom Tool Builder. Your custom tools are also listed in **Settings Tools Custom tools**. ## End-To-End Workflow 1. Bind or clone the folder in a conversation. 2. Check the folder name, branch and changes in the panel. 3. Ask a question about the code, or describe the change you want. 4. Approve only the commands and edits you understand. 5. Run the checks from **Run**. 6. Review **Changes** before committing, then commit and push from **Git**. ## Parallel Child Workspaces Writer ownership follows the resolved workspace folder. Children working in distinct folders or worktrees can edit concurrently, while writers aimed at the same folder are serialized to prevent overlapping changes. Changing only a shell working directory does not create an independent write boundary. Use a separate worktree or folder when two children genuinely need parallel write access. ## Sandbox Modes Local mode lets Row-Bot operate in the selected workspace with your configured file and command permissions. Docker Sandbox mode, when available on a host installation, isolates command execution in a container and requires an import step before changes affect the real workspace. It requires a supported host Docker runtime and can differ from your local environment. Inside the official Row-Bot application container, Developer Docker Sandbox is unavailable and a requested Docker workspace fails closed; Row-Bot never probes a nested daemon or silently runs that workspace locally. Local mode remains an explicit choice and can see only workspace paths deliberately mounted into the application container. An approved risky Custom Tool is a third case: it deliberately executes in Local mode inside the application container against the selected visible Custom Tool path. That behavior is shown in the approval dialog and is not a nested Docker sandbox or fallback from a requested Docker workspace. See [Docker And VPS Operations](/docs/operations/docker#developer-and-headless-boundaries) for the complete container boundary. For a host installation, start with Local mode on a disposable branch. Use Docker Sandbox when you want stronger isolation and the supported host runtime is available. ## Troubleshooting - If Row-Bot can't see your code, check that the folder is listed under **Working on** in the conversation's details card. - If choosing a folder does nothing, use the Row-Bot desktop app; a browser can't open folders on this computer. - If an Agent folder assignment is rejected, confirm that the directory already exists and pass either that folder or an existing workspace ID, not both. - If a command asks for approval, read the exact command and folder before approving. - If Docker sandbox import is offered, review the patch before importing it into the real project.

## Knowledge

Route: /docs/knowledge/

Explore, search, review and correct what Row-Bot remembers in Home Knowledge, and configure memory in Settings Memory.

# Knowledge Knowledge is Row-Bot's local memory: the facts, people, projects, preferences and relationships it has saved from conversations, documents and your own edits, connected as a graph. Row-Bot recalls relevant memories when they help a conversation. For the relationship between thread history, memory, documents, the graph, Wiki Vault, and Dream Cycle, read [How Memory Becomes Knowledge](/docs/concepts/memory-knowledge-and-dream-cycle). ## Where To Find It - **Home Knowledge** is where you see and manage memories. - **Settings Memory** turns memory on or off, shows graph health, configures the [Wiki Vault](/docs/knowledge/wiki-vault), and holds **Delete all knowledge**. - **Settings Documents** configures document search and its embedding model. ## Views Choose a view at the top of the Knowledge tab: - **Graph** draws memories and their connections and lets the layout settle into place. It opens with up to 2,000 memories; **Show all memories** draws up to 5,000. Use **Zoom in**, **Zoom out** and **Fit**, and hide or show memory types from the legend. - **List** shows the whole library, newest first, with search across every memory. - **Review** is the queue of memories Row-Bot was unsure about. **Mark as reviewed**, **Edit memory** or **Archive memory** settles each one. - **Activity** shows recent recalls (memory used or skipped in a conversation) and recent changes. **Filters** narrow the graph and list by source, status and memory type, show or hide you as the central memory, and can hide unconnected memories. ## A Memory's Details Select a memory to see its description, memory type, status, source, confidence and when it was updated, its connections, where it came from, and its history (created, edited by you, refined by Row-Bot, last recalled). From there: - **Edit memory** corrects it. - **Merge or replace** folds a duplicate into another memory and moves its connections. - **Archive memory** keeps it but stops it being recalled; **Restore memory** brings it back. - **Delete memory** removes it, its connections and its search entries after you confirm. Select several memories to delete them together, up to 100 at a time. Use **Add memory** to save something yourself. ## Dream Cycle Dream Cycle tidies memory overnight: it merges duplicates, enriches thin memories and infers missing links. Turn it on and choose its hours in **Settings Preferences Overnight tidy-up**. **Run Dream Cycle** on the Knowledge tab reviews what it would do and runs it now after you confirm. Its history is in [Monitor](/docs/monitor/). ## Common Workflow 1. Open Home Knowledge after a few useful chats or document imports. 2. Search for the topic you care about, or open **Review**. 3. Open a memory and check its source. 4. Correct, merge, archive or delete misleading memories. 5. Rely on recall once the important memories are right. ## What Is Saved Memories are local Row-Bot data. They are sent to a model provider only as relevant context in a conversation that uses that provider. ## Troubleshooting - If Knowledge is empty, turn on memory in Settings Memory or add documents in Settings Documents. - If a memory is wrong, edit, merge or archive it rather than deleting whole topics. - If recall feels noisy, review the **Review** queue and archive stale memories. Continue with [Wiki Vault](/docs/knowledge/wiki-vault) and [Provenance And Repair](/docs/knowledge/provenance-repair) for the file-output and recovery workflows.

## Wiki Vault

Route: /docs/knowledge/wiki-vault

Keep a local Markdown vault of Row-Bot's memories, with readable file names, and sync your edits back.

# Wiki Vault Wiki Vault writes your memories and their connections as linked Markdown articles in a folder you choose, so you can read them in Obsidian or any editor. It is an output of the local knowledge graph, not a replacement for the underlying memory and provenance records. ## Configure The Vault 1. Open **Settings Memory** and find **Wiki vault**. 2. Turn on **Enable Wiki Vault** and choose the vault folder with **Browse**. Choosing a folder needs the Row-Bot desktop app; pick a folder separate from application code and important personal notes while you test. 3. Save the configuration, then use **Rebuild managed wiki files** to write every article. 4. Open the articles in your editor and check links, titles and source notes. Articles have readable names: a memory called "Garden plan" becomes `Garden plan.md` in the folder for its type. When two memories share a name, or a note of your own already uses it, Row-Bot adds a short code to the new file name instead of overwriting anything. A renamed memory's article is renamed with it. ## What Row-Bot Writes And Leaves Alone Row-Bot keeps a record of the articles it wrote and only ever updates or moves those. Notes you create yourself in the vault are never changed or deleted. Articles from an earlier version are adopted in place with the same names, so existing links keep working. Files Row-Bot replaced are moved aside for your review, not deleted; **Show moved files** lists them. ## Your Edits When you edit an article in the vault, **Sync vault edits** brings the change back to the memory. If the memory changed in Row-Bot too, **Resolve versions** lets you choose between the database version and the vault version. Correct important facts in Knowledge first when you can, so the graph remains the source of truth. ## Privacy The vault is local, but it is deliberately easy to open in other software or sync with a third-party folder. Treat the destination as an export boundary. Review it before placing it in cloud storage, a shared repository, or a publishing system. ## Troubleshooting - If articles are missing, check that memory is on and the vault folder reads **Available**, then rebuild. - If **Browse** opens no folder picker, use the Row-Bot desktop app; a browser can't choose folders on this computer. - If links duplicate, run Knowledge repair before rebuilding the vault. - If private material appears, correct or remove its source memory, rebuild, and review the destination before sharing.

## Knowledge Provenance And Repair

Route: /docs/knowledge/provenance-repair

Trace where knowledge came from, correct records, repair relations, and rebuild derived outputs safely.

# Knowledge Provenance And Repair Knowledge is useful only when you can tell where it came from. Review provenance before relying on a memory, document fact, entity, or relation for consequential work. ## Review A Record 1. Open **Home Knowledge** and search or filter to the topic. 2. Select the memory to open its details. 3. Check its source, timestamps, type, confidence or status, and related entities. 4. Compare important claims with the original conversation or document. 5. Correct, merge, or remove the record if it is misleading. ## Repair Order Use the smallest repair that fixes the problem: 1. Correct a title, type, description, or relation. 2. Merge a duplicate entity when both records refer to the same thing. 3. Remove a bad relation without deleting sound entities. 4. Rebuild derived indexes if search still returns stale results. 5. Rebuild Wiki Vault output only after the graph is correct. Monitor's **History** (Dream Cycle and Extraction) and Knowledge's **Activity** view help explain background changes. A repair should preserve useful source evidence and avoid touching unrelated memories. ## Recovery Before a large repair or migration, back up with **Settings Data Back up and restore**, or close Row-Bot and copy the active data directory. Restore the whole related data set together rather than mixing database files from different moments. If the app offers a built-in repair action, read its scope and completion message before manually replacing files. ## Safety Boundary Deleting a thread does not necessarily mean every derived record disappears immediately, and deleting one graph entity can affect linked output. Review the exact record and its relationships first. Never run repair tests against a personal data directory.

## Computer Use

Route: /docs/computer-use/

Set up and safely use Row-Bot's opt-in native Windows and macOS application control.

# Computer Use Computer Use is Row-Bot's provider-neutral tool for operating native Windows and macOS applications. It is separate from browser automation: use Browser for web pages and Computer Use only when the task must interact with a desktop application window. Computer Use is a beta feature. It is off by default, supports one interactive local task at a time, and is unavailable to schedules, channels, background workflows, child agents, plugins, external MCP callers, mobile clients, and headless or server sessions. ## Set It Up 1. Open **Settings System** and choose **Browser and computer use**. 2. Under **Desktop apps**, open the **Computer Use (Beta)** setup and read the capability and privacy summary. 3. Choose **Install**. Row-Bot shows the required Cua Driver telemetry disclosure before it downloads or starts anything. 4. Choose **Continue** only if you accept the disclosure. Row-Bot downloads the pinned Cua Driver 0.20.0 full archive for your platform, verifies its SHA-256, and extracts it into Row-Bot's private data directory. Upgrading from an older reviewed driver requires accepting the expanded version-2 telemetry notice again. 5. On macOS, grant Accessibility and Screen Recording to the Row-Bot process when prompted, then choose **Recheck**. Restart the same process after changing permissions if the status asks you to. 6. Turn on **Computer Use (Beta)** after the setup card reports ready. Install, Repair, Reinstall, Remove, and system-binary controls affect only the optional driver. Row-Bot never runs the upstream installer or updater. A custom system binary is accepted only after explicit opt-in and version verification. ## Run A Native Task Ask from a normal local desktop chat and name the application and desired outcome. Row-Bot discovers applications and windows, acquires one exclusive task lease, and binds actions to the selected target window. It can launch an allowlisted application, observe that window, click, type, press keys, scroll, and drag when the current policy permits the operation. When accessibility data is insufficient, Row-Bot may send one ephemeral target-window screenshot to the Vision model configured in Settings. The live control card shows sanitized state and a shielded thumbnail; screenshot bytes are not written to chat, durable memory, replay history, logs, or tool results. ## Stop Or Take Over - **Stop** cancels queued work and releases the session. - **Take over** cancels queued mutation and pauses the lease so you can interact directly. - **Resume** requires Row-Bot to observe the target again before it can act. Window replacement, permission loss, target changes, or driver failure invalidates stale state. Row-Bot must reacquire and observe before another mutation. ## Safety Boundaries Computer Use blocks terminals, password managers, Row-Bot itself, secure desktops, elevation prompts, security settings, and attempts to handle credentials, one-time codes, CAPTCHAs, biometrics, or operating-system permission dialogs. Consequential actions use approval policy at the point of risk, and external handoff flows stay with the user. The reviewed Cua 0.20.0 telemetry sends pseudonymous installation/process-session identifiers and bounded product, platform, client, tool/outcome, duration/output, aggregate session/config/cursor/recording, permission, and lifecycle categories to Cua's EU PostHog endpoint. Its tagged event builders do not receive prompts, tool arguments/results, typed text, screenshots, accessibility trees, app/window names, URLs, paths, raw config/cursor values, or raw errors. Row-Bot adds no first-party telemetry and keeps its prompts, files, memories, secrets, screenshots, tool arguments, and channel content outside Cua telemetry. For the exact allowlist and reviewed dependency record, see [Computer Use Beta: architecture and security decision](https://github.com/siddsachar/row-bot/blob/main/docs/COMPUTER_USE_SECURITY.md). ## Troubleshooting - If setup reports an archive or checksum error, choose **Repair** and retry on a trusted network. - If macOS reports missing permission, use the provided buttons to open Accessibility and Screen Recording settings, grant the running Row-Bot process, restart it, and recheck. - If the requested app is blocked, do not work around the policy with a terminal or generic MCP tool. Operate it manually or use a narrower supported application. - If a session is busy, stop or finish the active Computer Use task before starting another. - If the target window closes or changes, ask Row-Bot to reacquire it before resuming.

## Monitor

Route: /docs/monitor/

See Row-Bot's latest health checks, fix problems in one click, and read activity, maintenance, history and logs.

# Monitor Monitor shows whether Row-Bot is healthy, what needs attention and what recently happened. Use it when a workflow, channel, account, model, background job, knowledge extraction or Dream Cycle needs investigation. ## Where To Find It Choose **Home**, then the **Monitor** tab. When something needs attention, the sidebar's attention indicator says so ("1 thing needs attention") and opens Monitor. ## Checks Are Kept Row-Bot keeps the last result of every check, with its time, so the tiles stay current when you switch tabs or restart the app. - **Checks on this computer** (disk space, databases, the scheduler, memory extraction, Dream Cycle, the search index, documents, the wiki vault, channels and the chosen model) run by themselves at start-up and every 15 minutes. They make no network requests. - **Connection checks** contact your model providers, accounts and the internet. They run once an hour while **Check connections every hour** is on, and whenever you choose **Run diagnosis**. Turn the switch off to run them only when you ask. - A check that turns red raises the sidebar's attention indicator and appears in Overview's **Needs you**. ## Needs Attention And Fixes **Needs attention** lists each problem with the one fix that addresses it: | Problem | Fix | | --- | --- | | A channel stopped | **Restart** the channel, the same reviewed start as its Settings row. | | An account sign-in expired | **Reconnect** Google or X. | | No default model | **Choose a model**. | | A setting needs a change | **Open ... settings** at the exact row, such as the public link in Devices & remote access. | | Disk, network or Ollama | **Check ... again**. | Failed workflows offer **Run ... again** on Overview. ## Health Tiles Six tiles summarise the areas Row-Bot depends on: **Model runtime**, **Channels**, **MCP and tools**, **Scheduler**, **Knowledge** and **System**. Each shows its worst check and that check's fix. Open a tile for every check in it, when each was last checked, and each problem's fix. **Run diagnosis** runs every check now. **Copy diagnosis report** copies a plain summary you can paste into a support request. **Start-up warnings** lists anything Row-Bot reported when it started. ## Activity, Maintenance And History - **Activity** charts extraction, Dream Cycle, workflow runs and channel events over **24 hours** or **7 days**. - **Maintenance** shows when knowledge extraction and Dream Cycle last ran and what they did: memories saved, conversations read, contradictions blocked, duplicates merged, memories enriched and links inferred. - **History** lists each Dream Cycle and extraction run with its changes and errors. ## Logs **Logs** shows Row-Bot's local log. Filter by level, **Search logs**, **Follow** new lines, **Collapse repeats**, and **Copy** the visible lines. ## Common Workflow 1. Reproduce or observe the issue. 2. Open Monitor and read **Needs attention**. 3. Use the suggested fix, or open the tile for detail. 4. Choose **Run diagnosis** to confirm, then check **Logs** if the problem remains. 5. Copy only non-sensitive details into a support request. ## What Is Saved Check results, logs and maintenance history are local Row-Bot data. Logs may include file names, conversation names, provider names, channel names or workflow names, so review them before sharing. ## Troubleshooting - If a tile shows **Checking...**, the first checks are still running. - If a connection check is old, choose **Run diagnosis** or turn on **Check connections every hour**. - If Dream Cycle never runs, turn it on in **Settings Preferences Overnight tidy-up**. - If a channel keeps stopping, open **Settings Channels** and check its credentials before restarting it.

## Home: Insights

Route: /docs/home/insights

Review what Row-Bot noticed about its skills, settings and memory, and apply or reject its proposals in Home Insights.

# Home: Insights Insights is where Row-Bot reports what it noticed while working: a skill that keeps failing, two skills that overlap, a setting worth changing, a memory that looks wrong, or a problem worth sending as feedback. Nothing changes until you choose; every proposal is a draft for your review. ## Where To Find It Choose **Home**, then the **Insights** tab, or **Open Insights** from Overview's **Learned this week** card. Insights belong to the owner on this computer. ## Reading An Insight Each insight shows its severity (Critical, Error, Warning, Info or Suggestion), what Row-Bot found, when it found it, and a suggested next step. Row-Bot says when an insight may be out of date: it was found while another chat model was in use, or it has not been seen again for two weeks. Use **All** or **Pinned** to filter the feed, **Refresh** to read it again, **Pin** to keep an insight in view, and **Dismiss** to hide it (with **Undo**). ## Proposals - **Suggest a fix** asks Row-Bot to prepare proposals for an insight that has none yet. - **Apply proposal** makes the change, such as creating or updating a skill. Afterwards the insight says exactly what happened. - **Reject proposal** declines it. - Some proposals are review only: merging skills, changing a setting, or correcting a memory. Row-Bot can't make that change itself, so make it yourself if you agree, then reject the proposal. - **Open investigation draft** opens a prepared conversation for a deeper look. - **Copy feedback** copies a feedback report you can send yourself; Row-Bot never sends it. ## Skill Library Report **Analyze skill library** reviews your installed skills and prepares proposals, such as two skills with similar instructions. It prepares proposals for your review; nothing is applied. ## What Is Saved Insights and their proposals are local Row-Bot data. Applying a proposal changes only what that proposal names. ## Troubleshooting - If the feed is empty, Row-Bot has nothing to report yet; new insights appear after it analyses recent work. - If an action's outcome is not confirmed, use **Check outcome** before trying again.

## Phone And Native Desktop

Route: /docs/mobile-native/

Use the full Row-Bot owner product from a phone, connect it from Settings Devices & remote access, and understand native-only desktop behaviour.

# Phone And Native Desktop A phone connected to your Row-Bot gets the whole owner product in a layout made for a small screen: conversations, Home with Overview, Workflows, Knowledge, Monitor and Insights, approvals, and every Settings page. The native desktop app adds operating-system integration such as the window, tray, microphone, file and folder pickers, the terminal, and the updater. Layout never changes Row-Bot's authority or the approval rules of the underlying task. ## The Phone Layout Row-Bot switches to the phone layout when the window is narrow, whatever the device. - A slim top bar opens the sidebar as a drawer and opens the command palette. - In a conversation, the header's **** menu holds **Find in conversation**, **Conversation details**, **Share or export**, **Rename conversation**, the conversation's panels, and the command palette. - Design and Developer panels open as full-screen sheets. - Settings shows a **Settings section** list at the top instead of the side list. It holds every page, in the same six groups as on the desktop: Preferences, Appearance and Buddy; Providers, Models and Voice; Memory, Documents and Tracker; Tools, Skills, Plugins and MCP; Accounts and Channels; System, Devices & remote access, Updates and Data. Features that need the computer running Row-Bot, such as the interactive terminal, choosing folders, and saving files with the system's Save dialog, are available only in the Row-Bot desktop app there. ## Connect A Phone 1. On the computer running Row-Bot, open **Settings Devices & remote access** and choose **Connect a phone or computer**. The command palette finds it as "phone". 2. Choose **How will it reach this computer?**: **Tailscale** (private and encrypted, recommended), **Same Wi-Fi** (at home or the office, over HTTP), **Internet** (from anywhere, through your public link), or another address you set up yourself. 3. Choose how long it should **Stay signed in**: **30 days** or **12 hours**. 4. **Scan the code** with the phone's camera, or open the shown address on it. 5. On the phone, review the connection page and press **Connect** before the code expires. 6. **Your devices** lists every signed-in phone and computer; sign one out when it is lost, replaced, or no longer trusted. The code expires after 10 minutes and works once. Opening it does not use it up until the phone presses **Connect**. The resulting session is a separate HttpOnly browser credential and never appears in the address. Connecting grants access to your local Row-Bot instance. Do not share a code, expose the local server broadly, or use Same Wi-Fi HTTP on an untrusted network. Tailscale or an operator-managed HTTPS address is preferred and is required for browser microphone use away from the computer. See [Remote Access And Server Mode](/docs/operations/remote-access) for the complete connection and recovery model. ## Native-Only Checks Some states cannot be reproduced faithfully in browser automation: OS microphone prompts, native file and folder pickers, tray menus, updater and restart dialogs, Computer Use takeover, and physical-device network permission. The public guide documents their intent; release review must test them on the target operating system and a physical phone. ## Troubleshooting - If the phone cannot connect, check the route you chose on the computer, the port, the firewall, and whether both devices can reach each other. - If a code expired or was already used, show a new code rather than copying cookies or local credential files. - If the connection page opens but the app stays unavailable, confirm the phone pressed **Connect** and that the device wasn't signed out. - If an approval seems missing, open Overview: **Needs you** lists every waiting approval. - If the phone layout appears on a computer, widen the window; the layout follows the window's width.

## How Row-Bot Works

Route: /docs/concepts/

Understand Row-Bot's request, knowledge, agent, workflow, and extension models before configuring them.

# How Row-Bot Works These pages explain the ideas behind Row-Bot without walking through every control. Use them when you want to predict what the app will do, where data goes, why work continues in the background, or what an extension is allowed to change. - [How A Request Runs](/docs/concepts/request-lifecycle) follows a message from a thread through model context, tools, approvals, and saved results. - [How Memory Becomes Knowledge](/docs/concepts/memory-knowledge-and-dream-cycle) separates thread history, memory, documents, the Knowledge Graph, Wiki Vault, and Dream Cycle. - [How Profiles, Goals, And Agents Work](/docs/concepts/profiles-goals-and-agents) explains reusable roles, durable objectives, parent coordination, and child-agent hand-offs. - [How Background Workflows Run](/docs/concepts/background-workflows) explains schedules, run state, approvals, retries, and delivery. - [Extensions And Trust Boundaries](/docs/concepts/extensions-and-trust) compares built-in tools, Skills, Custom Tools, plugins, MCP, channels, and accounts. Each page ends with links to the settings and task guides that change the behaviour it describes.

## How A Request Runs

Route: /docs/concepts/request-lifecycle

Follow a Row-Bot request through context assembly, a selected model route, tools, approvals, and local history.

# How A Request Runs A Row-Bot request is a controlled loop, not a single message sent blindly to every connected service. 1. **A thread provides continuity.** Your message joins the selected conversation. The thread supplies recent history and any thread-specific model or profile choice. 2. **Row-Bot assembles useful context.** Depending on your settings and request, this can include profile instructions, pinned or task-selected Skills, attachments, and relevant local memory or document results. Enabled skills that are not yet selected stay in a compact searchable catalog. It does not mean the whole local data directory is attached. 3. **The selected model route receives the model prompt.** A local model keeps that inference on the configured local runtime. A hosted, API, subscription, or remote custom endpoint sends the assembled prompt to that provider under its terms. 4. **The model may answer or ask for a tool.** Core tools are bound under the active profile. In Auto mode, Row-Bot can search enabled external MCP, plugin, Custom Tool, and channel capabilities when needed. It then checks that the selected tool is enabled, provider-compatible, and permitted by policy. 5. **Consequential actions stop at an approval boundary.** The approval describes the proposed action and target. Rejecting it prevents that action; approving it does not give unrelated future actions permission. 6. **Tool results return to the active run.** A local file tool, browser, channel, account, plugin, or MCP server has its own execution and data boundary. The model can use the result to continue the loop. 7. **The result is recorded locally.** The thread keeps the conversation and run evidence. Optional extraction, memory, and document features decide what becomes reusable knowledge later. ## Three Boundaries To Keep In Mind | Boundary | What crosses it | What controls it | | --- | --- | --- | | Model route | The assembled prompt and later tool results needed for reasoning. | Selected provider/model, thread override, profile or workflow policy. | | Tool execution | Arguments needed by a local or external capability. | Tool enablement, workspace or account scope, approval policy. | | Durable knowledge | Reviewed or extracted information that may be recalled in another request. | Memory, Knowledge, document, embedding, and retention settings. | Local-first describes where Row-Bot and its durable data live. It does not make a hosted model, web tool, channel, remote MCP server, realtime voice provider, or external account local. ## Change How Requests Run - Choose routes and defaults in [Models And Providers](/docs/configuration/models-and-providers), [Provider Settings](/docs/settings/providers), and [Model Settings](/docs/settings/models). - Control tools and approvals in [Tools, Approvals, And Terminal](/docs/chat/tools-approvals-and-terminal) and [Privacy And Safety](/docs/privacy-safety/). - Control reusable context in [Memory Settings](/docs/settings/knowledge), [Documents Settings](/docs/settings/documents), and [Skills Settings](/docs/settings/skills). - Configure automatic or eager external capability loading in [Tools Settings](/docs/settings/tools) and read [Progressive Tools And Skills](/docs/guides/progressive-tools-and-skills).

## How Memory Becomes Knowledge

Route: /docs/concepts/memory-knowledge-and-dream-cycle

Understand thread history, memory, documents, the Knowledge Graph, Wiki Vault, and Dream Cycle as related but distinct layers.

# How Memory Becomes Knowledge Row-Bot has several kinds of continuity. They work together, but none is a magical copy of everything the app has ever seen. | Layer | What it represents | How it is used | | --- | --- | --- | | Thread history | Messages, attachments, tool traces, approvals, and results in one conversation. | Keeps the current conversation coherent and reviewable. | | Memory | Reusable facts, preferences, summaries, and events saved or extracted for later recall. | Supplies selected relevant context to future requests when enabled. | | Documents | Files deliberately added to the document library and split into searchable content. | Finds passages related to a question without placing every document in every prompt. | | Knowledge Graph | Local entities and directed relationships with source and confidence information. | Connects related people, projects, events, facts, and sources during review and recall. | | Wiki Vault | Linked Markdown pages generated from reviewed graph material. | Makes a portable, human-browsable projection in a folder you choose. | ## From Conversation To Recall When extraction is enabled, Row-Bot can identify useful information from eligible conversations or documents and save structured records. Search combines the available lexical, vector, and graph signals to choose candidates relevant to a later request. Recall refreshes useful memories; it does not make every stored item equally likely to appear. Vector search uses an embedding model, which is separate from the chat model. The recommended local one is Mixedbread Embed Large v1, a 675 MB download you install from Settings Documents. Until it is installed, or while it is unavailable or still loading, Row-Bot continues with bounded lexical and graph fallback instead of silently downloading during a chat. Install, retry, or repair the local model later in [Documents Settings](/docs/settings/documents), then rebuild document and memory vectors after changing models. Cloud embeddings are an explicit alternative that sends indexed text to the selected provider. The source record matters. A graph relation or memory should remain traceable to the conversation, document, or process that produced it. Correct the underlying Knowledge record before rebuilding derived indexes or Wiki Vault pages. ## What Dream Cycle Does Dream Cycle is idle-time knowledge maintenance. When enabled and due, it can examine the local knowledge set, merge high-confidence duplicates, refresh summaries from available source context, infer strongly supported relationships, and prune stale low-confidence inferred relations. It records its history so you can review what happened in Monitor. Dream Cycle is not an autonomous second assistant and it does not make external facts true. It should run only during the configured window while the app is idle. Inferred relationships use confidence thresholds, but important knowledge still deserves human review. ## What Wiki Vault Does Wiki Vault writes a readable view of selected graph material. The graph remains the source of truth; the vault is a generated output. If the destination is synced to a cloud drive, repository, or publishing tool, that folder becomes a separate sharing boundary. ## Configure And Review It - Review memories in [Knowledge](/docs/knowledge/) (Home Knowledge) and investigate changes in [Monitor](/docs/monitor/). - Turn memory on or off, check graph health and configure Wiki output in [Memory Settings](/docs/settings/knowledge); Dream Cycle's hours are in Settings Preferences. - Configure indexed files in [Documents Settings](/docs/settings/documents). - Set up the export folder in [Wiki Vault](/docs/knowledge/wiki-vault) and correct mistakes with [Provenance And Repair](/docs/knowledge/provenance-repair). - Back up the related local data as one set using [Operations, Data, And Recovery](/docs/operations/).

## How Profiles, Goals, And Agents Work

Route: /docs/concepts/profiles-goals-and-agents

Understand reusable Agent Profiles, durable goals, parent coordination, and bounded child-agent delegation.

# How Profiles, Goals, And Agents Work Profiles, goals, and agent runs describe different parts of long-running work. - An **Agent Profile** is a reusable operating role: instructions, model policy, Skills, tools, delegation limits, and approval policy. - A **goal** is the durable objective and finish condition for a piece of work that may span several turns. - An **agent run** is one execution attempt with a prompt, model, status, result, and evidence. - A **parent agent** coordinates the objective and combines results. - A **child agent** receives a bounded subtask and returns a hand-off to its parent. ## Why Delegate Delegation is useful when independent research, inspection, or implementation can happen in parallel or needs a specialist profile. It is less useful when agents would edit the same small resource or when only one decision is needed. The parent remains responsible for deciding whether the combined evidence actually completes the goal. Child agents do not become invisible background permissions. Their runs remain linked to the parent, and a returned result is evidence rather than automatic authorisation to write, send, publish, or declare the goal complete. Tool availability and approvals still apply to the run doing the action. Delegation capacity is bounded. Each new run snapshots the application-wide work-round, nesting, per-parent concurrency, app-wide concurrency, and optional child active-time settings. When capacity is full, eligible children wait in a first-in, first-out queue rather than bypassing the limits. ## Status And Hand-Offs An active goal can accumulate progress across turns, with no turn limit unless you set one. After each turn Row-Bot judges whether the goal made progress, is done, or is blocked; two turns without progress, or the same step failing three times, pause it with the reason so you can decide. Completion should mean the finish condition and required checks are satisfied. A block should identify a real impasse, not merely unfinished work. Stopping an agent ends that run without deleting the parent thread, goal history, or already-recorded evidence. The work-round budget is checked at model, tool, and resume boundaries so a restart cannot reset it. Repeated no-progress states are blocked and then terminated cleanly, and every terminal path writes one durable final status and reason. A useful child-agent hand-off says what was checked, what changed, what evidence was found, what remains uncertain, and whether any consequential action still needs the parent or user. ## Configure And Use It - Create and select roles, set goals and follow delegated agents with [Profiles, Goals, And Agents](/docs/profiles-goals-agents/). - Choose a profile in a workflow with [Workflows](/docs/guides/workflows). - Control the models behind runs in [Models And Providers](/docs/configuration/models-and-providers). - Review waiting work and failures in [Home: Overview](/docs/home/overview) or [Monitor](/docs/monitor/).

## How Background Workflows Run

Route: /docs/concepts/background-workflows

Understand saved workflow definitions, schedules, isolated runs, approvals, retries, and delivery semantics.

# How Background Workflows Run A workflow is a saved task definition. A run is one attempt to execute it. Keeping those separate lets Row-Bot preserve history when you edit the next schedule or retry a failed attempt. 1. **A trigger starts a run.** This can be a manual click, a recurring schedule, a one-time time, or another supported trigger. 2. **The run resolves its policy.** Row-Bot chooses the workflow's Agent Profile, model override, tools, Skills, steps, and approval mode. 3. **The run gets its own state.** Status, progress, step output, logs, and any persistent thread link are recorded so the scheduler can recover cleanly. 4. **Approval can pause the run.** A background action that needs a decision becomes a pending approval instead of silently continuing. It waits until you answer, unless a step sets its own time limit, and it shows everywhere: the sidebar's attention indicator, Overview's **Needs you**, Buddy, and the workflow's channels. Approving resumes that action; denying it leaves a visible run outcome. While a run waits, the same workflow's next scheduled run is skipped and says so, so runs don't pile up. 5. **Completion and delivery are recorded separately.** A useful result can complete even if an external channel delivery fails, so the app can show the result and the delivery problem honestly. ## Delivery Defaults The web app always receives run status. External channels are an additional destination. - **Inherit defaults** means the workflow uses the current workflow-level external channel selection. - **An explicit empty selection** means web app only, even when a global external default exists. - **An explicit channel selection** targets those configured running channels. This distinction prevents a workflow that was deliberately set to app-only from beginning to send externally after someone changes the global default. ## Failure And Retry A failed run remains evidence; retrying creates another attempt rather than rewriting history. Fix the underlying provider, model, tool, approval, input, or delivery problem first. For multi-step work, keep steps narrow enough that the status identifies where the run stopped. Scheduled work can outlive the screen where it was created, but it cannot outlive the local Row-Bot process indefinitely. Pausing a workflow prevents future schedule starts; stopping one active run does not necessarily change the saved schedule. ## Configure And Operate It - Create schedules, profiles, approvals, steps, and delivery in [Workflows](/docs/guides/workflows). - Configure external destinations in [Channels](/docs/integrations/channels) and [Channel Settings](/docs/settings/channels). - Review live status in [Monitor](/docs/monitor/) and waiting approvals in [Home: Overview](/docs/home/overview), on the desktop or a [phone](/docs/mobile-native/). - Back up workflow definitions and history with [Operations, Data, And Recovery](/docs/operations/).

## Extensions And Trust Boundaries

Route: /docs/concepts/extensions-and-trust

Compare Row-Bot tools, Skills, Custom Tools, plugins, MCP servers, channels, and accounts by what they add and where they run.

# Extensions And Trust Boundaries Row-Bot can learn a procedure, gain a tool, load code, connect a server, or authorise an account. Those are different changes and deserve different review. | Capability | What it adds | Main trust question | | --- | --- | --- | | Built-in tool | A maintained capability shipped with Row-Bot. | Is it enabled for this profile, and does this action need approval? | | Skill | Instructions, examples, and supporting files that shape behaviour. | Do the instructions match your intent and avoid unexpected data or tool use? | | Custom Tool | Reviewed local code or a command exposed as a reusable tool. | What code runs, in which workspace, with which dependencies and command class? | | Plugin | A packaged extension that can add code, tools, Skills, channels, configuration, or UI. | Do you trust its publisher, source, permissions, dependencies, and update path? | | MCP server | Tools supplied by a separate local process or remote service. | What can the server access, where does it run, and which of its tools are enabled? | | Channel | A bridge for incoming messages or outgoing results. | Who can contact it, which conversation receives replies, and may workflows deliver there? | | Account | Authorisation for a service such as GitHub, Google, or X. | Which service scopes and account data become available to enabled tools? | ## Instructions Are Not Permissions A Skill or profile can tell the model when to use a capability, but it does not create credentials or bypass Row-Bot's tool and approval policy. Conversely, installing a plugin or MCP server can add real executable capability even when no Skill mentions it. Review both the guidance layer and the execution layer. Progressive discovery preserves this separation. Loading a skill adds instructions to one task; it cannot make a profile-denied tool callable. Searching an enabled external-tool catalog does not enable a tool, change its provider schema, or bypass its approval classification. ## Local Does Not Always Mean Harmless A local plugin, command, or MCP process can still read files, start programs, or use locally stored credentials. A remote server or account can send data outside the app even if its configuration is stored locally. Use narrow workspaces, least-privilege accounts, disabled-by-default imports, and fictional test data. Third-party dependency telemetry is not Row-Bot telemetry. Review dependency behaviour before installation and do not expose prompts, files, memories, secrets, screenshots, tool arguments, or channel content unless that transfer is the feature you intentionally selected. ## Browse And Configure Extensions - Compare extension types and browse screenshots in [Extend Row-Bot](/docs/extending/). - Browse and review Skills in [Skills Hub](/docs/skills/) and [Skills Settings](/docs/settings/skills). - Understand automatic capability selection in [Progressive Tools And Skills](/docs/guides/progressive-tools-and-skills) and configure it in [Tools Settings](/docs/settings/tools). - Install and control packaged code in [Plugins](/docs/integrations/plugins) and [Plugin Settings](/docs/settings/plugins). - Browse, add, test, and enable servers in [MCP](/docs/integrations/mcp) and [MCP Settings](/docs/settings/mcp). - Configure external messaging in [Channels](/docs/integrations/channels) and service authorisation in [Accounts And Authorisation](/docs/integrations/accounts).

## Settings

Route: /docs/settings/

Find every Row-Bot settings page: six groups from General to System, with search for any setting.

# Settings Settings is where Row-Bot is configured: where models come from and which one does each job, memory and documents, tools and extensions, accounts and channels, and how Row-Bot runs on this computer and your other devices. Open it with **Settings** at the bottom of the sidebar. The pages are grouped down the left side; **Find a setting** (press / on a Settings page) searches page names and individual settings and jumps to the exact row. The command palette finds them too, and can flip common switches in place. On a phone, a **Settings section** list at the top replaces the side list; every page and setting is still there. ## How Settings Is Organized | Group | Page | What it changes | | --- | --- | --- | | **General** | [Preferences](/docs/settings/preferences) | The assistant's name and personality, learning new skills, how Row-Bot opens, and the overnight Dream Cycle. | | | [Appearance](/docs/settings/preferences#appearance) | Theme, accent colour, density and transparency on this device, and panel sizes. | | | [Buddy](/docs/settings/buddy) | Your companion's look, personality and motion, and the desktop overlay. | | **Models** | [Providers](/docs/settings/providers) | Where models come from: local, subscription, pay-per-use and custom endpoints. | | | [Models](/docs/settings/models) | Which model does each job (Brain, Vision, Image, Video), limits for long work, and the catalog. | | | [Voice](/docs/settings/voice) | Talk, dictation and read-aloud. | | **Knowledge** | [Memory](/docs/settings/knowledge) | Remembering on or off, graph health and the wiki vault. Memories themselves live in Home Knowledge. | | | [Documents](/docs/settings/documents) | Files Row-Bot can search, and where search runs. | | | [Tracker](/docs/settings/tracker) | Habits, symptoms and health events you track. | | **Capabilities** | [Tools](/docs/settings/tools) | Capability loading, retrieval compression, search tools, built-in tools and custom tools. | | | [Skills](/docs/settings/skills) | Installed skills and public ones. | | | [Plugins](/docs/settings/plugins) | Installed plugins and the plugin marketplace. | | | [MCP](/docs/settings/mcp) | Tools from MCP servers on this computer or online, and their runtimes. | | **Connections** | [Accounts](/docs/settings/accounts) | GitHub, Google (Gmail and Calendar) and X. | | | [Channels](/docs/settings/channels) | Telegram, WhatsApp, Discord, Slack and SMS. | | **System** | [System](/docs/settings/system) | The workspace folder, what Row-Bot may do (commands, files, browser and computer use), and logs. | | | [Devices & remote access](/docs/operations/remote-access) | Connect a phone or computer, your signed-in devices, network access and the public link. | | | [Updates](/docs/operations/#update-repair-and-uninstall) | The version you have, the update channel, and checking for updates. | | | [Data](/docs/operations/#back-up) | Back up and restore, import from another assistant, and irreversible clean-up. | Goals belong to one conversation and are set in its details card; agent profiles are managed in the sidebar's **Agents** dialog. See [Profiles, Goals, And Agents](/docs/profiles-goals-agents/). All authenticated owner sessions see the same settings on every device. On phones and tablets, the layout adapts to the screen; the available settings and authority do not change. ## A Good Setup Order 1. Providers 2. Models 3. Documents and Tools 4. Skills 5. System access 6. Integrations: Accounts, Channels, MCP, Plugins 7. Voice and Buddy 8. Preferences and Appearance ## Safety Notes Credentials belong in Providers, Accounts, Channels, MCP, or plugin-specific settings, not in chat messages. Enable only the tools and integrations you intend to use. Review approvals before actions that write files, run commands, use accounts, contact external services, or send messages.

## Settings: Providers

Route: /docs/settings/providers

Connect local, hosted, subscription, and custom model providers.

# Settings: Providers Providers are the model services Row-Bot can call. This page answers whether each provider is connected, what kind of credential it uses, and whether Row-Bot has enough information to use it for chat, tools, designs, code, voice, or embeddings. ## Where To Find It Open **Settings**, then choose **Providers** in the **Models** group, or search for it with **Find a setting**. ## Controls - Connection Status summarizes how many providers are connected, local, API-based, subscription-based, or media-capable. - Local providers show runtimes such as Ollama. Refresh checks what the local service currently exposes. - Subscription accounts show sign-in based providers such as ChatGPT / Codex or Claude Subscription. Use connect, reconnect, test, and refresh actions from the row. - API providers show services that need an API key or compatible endpoint. Their credential buttons open the setup flow for that provider. - xAI API catalog refresh also discovers live image-generation models and their supported output formats, aspect ratios, resolutions, and quality choices. Row-Bot keeps that capability metadata separate from ordinary chat models. - Custom endpoint providers let advanced users point Row-Bot at OpenAI-compatible servers such as LM Studio, vLLM, llama.cpp, LocalAI, LiteLLM, or SGLang. - Custom endpoint Advanced settings can describe reasoning mode, a thinking budget, returned reasoning content, and whether preserved reasoning may be replayed to that endpoint. - Runtime tests check whether a provider can handle the kind of requests Row-Bot needs. A failed test keeps the provider visible but may stop Row-Bot from offering it for agent work. ## Common Workflow 1. Pick one provider path first: local Ollama for private local runs, a subscription account if you already use one, or an API provider if you prefer hosted models. 2. Add credentials or sign in only through the provider row you intend to use. 3. Refresh the provider so Row-Bot can discover available models. 4. Open **Settings Models** to choose which model does each job and pin the ones you use most. ## What Is Saved Provider connection state is global to the local Row-Bot app. Secrets are stored in the operating system key store when available; Row-Bot settings keep masked status and catalog metadata. ## Privacy And Safety Review credential, account, channel, provider, or tool settings before enabling features that can contact outside services. Local-only features stay on your machine until you ask Row-Bot to use a provider, account, channel, MCP server, plugin, or tool that sends data elsewhere. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#providers-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If a provider is connected but models do not appear, refresh Providers, then refresh Models. - If an xAI image model is missing or rejects a media option, refresh Providers so Row-Bot can use the live image-generation catalog. A quality choice that the endpoint rejects is retried once without that optional field; other generation failures are returned without an automatic duplicate request. - If a runtime test fails, read the row message before changing credentials; the model may be chat-capable but not tool-capable. - If local Ollama is missing, start Ollama and make sure at least one model is installed.

## Settings: Models

Route: /docs/settings/models

Choose which model does each job (Brain, Vision, Image, Video), set the reading limit and limits for long work, and pin models from the catalog.

# Settings: Models The Models page decides which model does each job. Providers supply the models (see [Settings: Providers](/docs/settings/providers)); this page picks between them. ## Where To Find It Open **Settings**, then choose **Models** in the **Models** group, or search for it with **Find a setting**. ## Jobs Each job has its own picker, using the same grouped, searchable list as the composer's model picker (provider, how it is paid for, and whether it is available): - **Brain** is the default model for new conversations and background work. A conversation can still choose its own model in the composer. - **Vision** reads images, screenshots and your camera. Leave it as **Same as Brain** if your Brain model can see, or choose another model. Choose a **Camera** here too. - **Image** makes and edits pictures. - **Video** makes short clips and animates pictures. Turn Vision, Image or Video off when you don't want that job done at all. ## Reading Limit **Reading limit** sets how much text the Brain model reads at once. **Automatic** uses what the model supports; **Limit...** sets a smaller number of tokens, useful for local models on smaller computers. ## Limits For Long Work These limits apply to new runs; work in progress keeps the limits it started with. **Recommended limits** restores the defaults. | Limit | Meaning | | --- | --- | | **Steps per run** | Model-and-tool rounds one run may take. | | **Helper levels** | How deep delegation goes; 1 lets helpers work but not start helpers of their own. | | **Helpers at a time, per agent** | Delegated agents one agent runs at once; more wait in line. | | **Helpers at a time, in total** | Delegated agents running across the whole app. | | **Helper time limit (seconds)** | How long one helper may work; 0 means no limit, and time waiting in line doesn't count. | | **Goal turn limit** | The turn limit new goals start with; 0, the default, means no limit. Each goal can change it. | ## Catalog The **Catalog** lists every model your providers offer. Search it, filter by provider and job, and pin the models you use often so they appear first in every picker. **Refresh catalog** asks your providers what is available now. ## Common Workflow 1. Connect a provider, then choose your **Brain** model. 2. Choose Vision, Image and Video models if you want those jobs. 3. Pin one everyday model and one stronger tool-capable model. 4. Keep the recommended limits unless you have a measured reason to change them. ## What Is Saved Job models, limits and pins are global preferences. A conversation keeps its own model when you choose one in it. ## Privacy And Safety A hosted model receives the prompts, images or files its job sends it. Choose local models for jobs that should stay on your computer. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#models-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If a picker is empty, connect a provider first. - If a model is missing, refresh the catalog and check whether its provider is connected. - If a model shows as unavailable, read its reason; reconnecting its provider usually fixes it. - If a small local model fails before answering, lower the reading limit or choose a more capable model.

## Settings: Memory

Route: /docs/settings/knowledge

Turn remembering on or off, check graph health, configure the wiki vault, and delete all knowledge in Settings Memory.

# Settings: Memory The Memory page holds the settings for what Row-Bot remembers between conversations. The memories themselves are in **Home Knowledge**, where you search, review, edit and delete them. See [Knowledge](/docs/knowledge/). ## Where To Find It Open **Settings**, then choose **Memory** in the **Knowledge** group, or search for it with **Find a setting**. Old links to Settings Knowledge or Settings Wiki open this page. ## Remembering **Enable Memory** decides whether Row-Bot saves useful facts from conversations and recalls them later. A link opens Home Knowledge to manage what is saved. ## Graph Health **Graph health** counts your **Memories**, how many **Connected groups** they form, and how many **Unconnected memories** stand alone. A memory graph thumbnail opens Home Knowledge. Many unconnected memories usually means Dream Cycle or a review would help. ## Wiki Vault **Wiki vault** writes your memories as linked Markdown articles in a folder you choose, and brings your edits back. See [Wiki Vault](/docs/knowledge/wiki-vault). ## Delete All Knowledge **Delete all knowledge** permanently removes the saved graph after reviewing exactly what will go. Row-Bot-managed wiki files and local indexes are cleaned up; files outside Row-Bot's managed wiki are kept. Back up first if you might want it back. ## Related Settings - Dream Cycle's hours are in **Settings Preferences Overnight tidy-up**. - Where memory and document search runs, and its search model, are in **Settings Documents**. ## Common Workflow 1. Keep memory on if you want Row-Bot to remember useful details. 2. Install the private search model in Settings Documents before building a large memory. 3. Review and correct important memories in Home Knowledge. 4. Turn on the wiki vault when you want a readable local copy outside the app. ## What Is Saved Memories live in local Row-Bot data. They are sent to a model provider only as relevant context in a conversation that uses that provider. ## Privacy And Safety The wiki vault is a folder other software can read or sync; treat it as an export. Deleting all knowledge cannot be undone without a backup. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#memory-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If recall feels stale, rebuild the memory index in Settings Documents. - If Row-Bot remembers something wrong, edit or archive the memory in Home Knowledge. - If Dream Cycle is noisy, change its hours or turn it off in Settings Preferences.

## Settings: Buddy

Route: /docs/settings/buddy

Use Buddy in the sidebar or as a compact desktop overlay, message the selected thread, handle approvals, and configure its appearance.

# Settings: Buddy Buddy is Row-Bot's visual companion. It starts in the sidebar and, in the native Windows and macOS apps, can be dragged out into a compact always-on-top desktop overlay. The overlay is a smaller view of the thread selected in Row-Bot: it uses that thread's model, tools, approval mode, saved draft, and conversation rather than starting a separate assistant. ## Where To Find It Open **Settings**, then choose **Buddy** in the **General** group, or search for it with **Find a setting**. ## Undock Buddy Onto The Desktop 1. Turn on **Show Buddy** if the companion is hidden. 2. Press and drag Buddy itself away from its sidebar dock. Releasing it back over the dock cancels the move; dragging it beyond the native Row-Bot window tears it off. 3. Drop it near the part of the desktop where you want the overlay. Row-Bot keeps the fixed-size overlay inside the nearest monitor's usable work area, including multi-monitor layouts. 4. Drag the overlay header whenever you want to reposition it. Drag-to-undock requires the native Windows or macOS application. Browser/server mode, Linux browser-first launch, and the phone layout keep Buddy inside Row-Bot and do not create a native overlay window. ## Message The Selected Thread The title at the top of the overlay identifies the thread that will receive your next message. Buddy follows the conversation selected in the main Row-Bot window: switch conversations there and Buddy switches with you. The smaller line shows whether that thread is a Chat, Code, or Design conversation and what it is doing. If no thread is selected, Buddy follows your most recent conversation. - Type into **Message this thread...**. Enter sends; Shift+Enter adds a line. - The full composer and overlay share the saved draft for each thread. Switching threads loads that thread's draft without merging it into another conversation. - Sending captures the thread and the Chat, Code, or Design context shown at that moment, so changing the main-window selection during a run does not silently retarget the request. - After sending, the overlay shows current progress, the latest response, or an error. If a response was interrupted (for example by a restart), **Resume** continues it. - Use **Stop** to stop the selected thread's active response. The overlay does not retry the turn by itself. - Attachments, transcript history, tool traces, model selection, and microphone controls remain in the full Row-Bot thread. Use **Open full thread** when you need them. The overlay is a text composer, not a second microphone session. Talk and Dictate remain in the full thread. A voice turn submitted there belongs to the same conversation and its progress can still appear in Buddy. ## Review Approvals Buddy never creates a separate approval policy. It projects the selected thread's existing gate: - The current approval shows **Deny**, **Details**, and **Approve** in the overlay. Ctrl+Enter (Cmd+Enter on macOS) approves. - **Details** opens the full thread so you can inspect the complete request before deciding. - If the selected thread or pending generation no longer matches, Buddy opens the full thread instead of approving a stale request. ## Overlay Menu And Lifecycle - **Open full thread** restores the main Row-Bot window at the selected conversation. - **Dock Buddy** closes the native overlay and returns Buddy to its sidebar dock. - **Hide Buddy** hides the overlay without docking it. Use the tray's **Show Buddy** action to bring it back. - While Buddy is torn off, closing the main Row-Bot window hides that window instead of quitting the app. Use **Open full thread**, **Open Row-Bot** in the tray, or the tray's quit action as appropriate. - Tear-off is session-scoped. A fresh Row-Bot launch resets placement to the sidebar dock; a saved hidden preference remains hidden until you show Buddy. - The overlay follows Row-Bot's appearance setting (System, Light, or Dark) and your device's reduced-motion setting. ## Appearance Controls The Buddy page follows the conversation you have open, because Buddy uses that conversation's profile and approvals; open a conversation first if it asks. - **Show Buddy** shows or hides the companion. - **Look** chooses a bundled or custom Buddy appearance. **New look...** makes one with your image model; **Refresh looks** reads them again. - **Personality** changes the tone of Buddy's status cues. - **Bubbles** changes how visual status appears. - **Motion** chooses how lively Buddy is: **Calm**, **Normal** or **Lively**. - **Advanced Buddy settings** holds Buddy's **Name**, optional **Style notes** for new looks, and **Compact size**. Buddy's picture is also the speaker mark beside Row-Bot's replies in conversations. ## Common Workflow 1. Select the thread you want and confirm its name in the main window. 2. Drag Buddy out of the sidebar and confirm the same name appears in the overlay header. 3. Type a short message, review any approval, and use **Open full thread** when you need the complete transcript or tool details. 4. Use **Dock Buddy** when you want it back in the sidebar. 5. Make a new look only after an image model is ready in Settings Models; choose **Calm** motion if animation is distracting. ## What Is Saved Buddy visibility, appearance, overlay position, and thread drafts are local app data. The overlay does not copy conversations into a second store. Generating a custom look may call a configured media provider depending on your setup. ## Privacy And Safety The overlay follows the same privacy and safety boundaries as the selected thread. A hosted model, tool, account, channel, MCP server, plugin, or media provider can still send data externally when that thread uses it. Undocking Buddy does not grant new permissions or bypass approvals. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#buddy-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If Buddy does not appear in the sidebar, turn on **Show Buddy** and reload the current page if needed. - If a drag snaps back, start the gesture on Buddy itself and release away from the original dock. Drag-to-undock is unavailable outside the native Windows and macOS apps. - If a hidden overlay is still torn off, use **Show Buddy** from the system tray. - If Row-Bot's main window is hidden, use **Open Row-Bot** from the tray or **Open full thread** from Buddy. - If the overlay shows the wrong conversation, select the intended thread in Row-Bot and verify the overlay title before sending. - If an approval needs more context, choose **Review** or **Details** and decide in the full thread. - Restart Row-Bot to reset desktop placement to the sidebar dock. Then enable **Show Buddy** if a saved hidden preference keeps it out of view. - If making a new look fails, check that an image model is chosen and its provider is connected.

## Settings: Voice

Route: /docs/settings/voice

Configure dictation, realtime talk, read-aloud, voice models, devices, and diagnostics.

# Settings: Voice Voice gives Row-Bot microphone input and optional spoken output. Dictate turns speech into text for the composer. Talk is a conversation mode. Realtime voice uses a low-latency provider path when configured. ## Where To Find It Open **Settings**, then choose **Voice** in the **Models** group, or search for it with **Find a setting**. ## Controls - Talk settings decide whether spoken input submits directly to Row-Bot. - Dictate settings decide whether speech is inserted into the composer for review before sending. - Read-aloud settings control spoken assistant responses. - Local voice controls choose Whisper, the default local STT engine, or the optional FunASR/SenseVoice engine for Talk and Dictate. - SenseVoice readiness shows whether the platform, packaged runtime, and verified local model snapshot are ready. Its install action is always explicit. - Realtime voice controls use provider-backed low-latency voice models. - Device controls select microphone and output devices. - Voice Models shows runtime defaults and provider voice models. - Diagnostics checks local audio and provider readiness. Talk and Dictate controls live in the full chat composer, not the compact Buddy desktop overlay. If Buddy is undocked, use **Open full thread** before starting a microphone session; Buddy will continue to reflect the selected thread's response and approval state. ## Common Workflow 1. Use Dictate first if you want to review text before sending. 2. Use Talk when you want hands-light conversation and are comfortable with immediate submission. 3. Choose local Whisper for the default local path, or install and select SenseVoice Small when you want the optional FunASR engine. 4. Treat the SenseVoice install as a deliberate approximately 940 MB ModelScope download. After verification, normal transcription stays local and cache-only. 5. Choose realtime voice when latency matters and you accept provider requirements, cost, and internet use. ## What Is Saved Voice preferences are global. Transcribed text belongs to the active thread once submitted. The verified SenseVoice snapshot is stored under the active Row-Bot data directory. Provider-backed voice can send audio or transcript data to the selected provider. ## Privacy And Safety Review credential, account, channel, provider, or tool settings before enabling features that can contact outside services. Local-only features stay on your machine until you ask Row-Bot to use a provider, account, channel, MCP server, plugin, or tool that sends data elsewhere. The explicit SenseVoice install contacts ModelScope for the model download; it does not send audio, prompts, or usage data, and ordinary inference does not check for updates. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#voice-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If the microphone is silent, check the selected input device and browser/app permissions. - If SenseVoice is missing or invalid, use its install or reinstall action in Voice settings. Source installs also require the `voice` extra. - SenseVoice is unavailable on Intel macOS because the required CPU PyTorch and Torchaudio wheels are not published for that runtime; use local Whisper there. - If realtime voice is unavailable, configure a compatible provider in Providers. - If Talk submits too quickly, use Dictate mode instead.

## Settings: System

Route: /docs/settings/system

Choose Row-Bot's workspace folder and what it may do on this computer (commands, files, browser and computer use), and manage logs.

# Settings: System System settings describe where Row-Bot works on this computer and what it may do there: run commands, work with files, drive a browser, and, if you opt in, operate desktop apps with Computer Use. ## Where To Find It Open **Settings**, then choose **System** in the **System** group, or search for it with **Find a setting**. ## Workspace **Workspace folder** is where Row-Bot keeps its own files on this computer, such as new draft code folders. Choosing a folder needs the Row-Bot desktop app. ## What Row-Bot May Do Risky actions still ask first, as each conversation's **Approvals** setting decides. - **Run commands** turns the shell tool on or off. The command palette can flip it too ("turn off shell"). - **Work with files** controls the file tools. - **Browser and computer use** opens their setup: - **Browser** turns the browser tool on and chooses its runtime: a supported installed Chrome or Edge, or Row-Bot's exact managed Playwright Chromium. **Install browser runtime** and Repair are explicit actions; normal start-up only checks readiness and never downloads a browser. - **Desktop apps** sets up **Computer Use (Beta)**, which stays off by default and is gated by platform readiness, its telemetry disclosure, and a verified optional runtime. See [Computer Use](/docs/computer-use/). ## Logs **Log detail** sets how much Row-Bot writes to its local log. **Open log folder** opens it. Monitor's **Logs** shows the same log inside the app. ## Advanced System Settings Additional blocked command patterns, app availability for Computer Use, and removing the managed Computer Use runtime live under **Advanced system settings**. ## What Moved - Connecting a phone or another computer, signed-in devices, network access and the public link (tunnel) are on **Settings Devices & remote access**. See [Remote Access And Server Mode](/docs/operations/remote-access). - **Open in** (app window or browser) is on **Settings Preferences**. - Updates are on **Settings Updates**. ## Common Workflow 1. Keep the workspace folder somewhere you back up. 2. Turn on commands, files or the browser only when you intend to use those tools. 3. Turn on Computer Use only on an interactive Windows or macOS desktop after reading its setup and safety guide. 4. Raise **Log detail** while troubleshooting, then lower it again. ## What Is Saved System settings are global for the active Row-Bot data directory. Workspace files are not uploaded to Row-Bot; external providers may still receive content when a model or tool request sends it. ## Privacy And Safety Commands, file access, browser automation and Computer Use widen what Row-Bot can do on this computer. Keep approvals on **Ask** in conversations that use them, and review each request. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#system-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If Row-Bot cannot read a file, check whether it is inside the folder it was given and that **Work with files** is on. - If Browser reports a missing or mismatched managed runtime, use **Install browser runtime** or Repair. Do not copy another Playwright cache into Row-Bot's managed location. - If Computer Use is unavailable, open its setup and follow the reported runtime or operating-system permission step. - If choosing the workspace folder does nothing, use the Row-Bot desktop app.

## Settings: Tracker

Route: /docs/settings/tracker

Configure recurring activity, habit, symptom, and health tracking surfaces.

# Settings: Tracker Tracker is for structured personal logs such as recurring activities, habits, symptoms, or health events. It gives Row-Bot a more organized way to store and review repeated observations than free-form chat alone. ## Where To Find It Open **Settings**, then choose **Tracker** in the **Knowledge** group, or search for it with **Find a setting**. ## Controls - Category controls decide what kinds of events can be logged. - View and chart options change how tracker history is summarized. - Import or cleanup actions manage existing tracker data. - Privacy-oriented status text explains that tracker records stay in local app data unless you ask a provider or channel to use them. ## Common Workflow 1. Choose the categories you actually want to track. 2. Log events consistently from Chat or tracker-aware workflows. 3. Review trends in the tracker view before asking Row-Bot to summarize them. 4. Remove categories you no longer use to keep prompts focused. ## What Is Saved Tracker data is local Row-Bot data. It can influence answers only when Row-Bot is allowed to retrieve or summarize it. ## Privacy And Safety Review credential, account, channel, provider, or tool settings before enabling features that can contact outside services. Local-only features stay on your machine until you ask Row-Bot to use a provider, account, channel, MCP server, plugin, or tool that sends data elsewhere. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#tracker-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If a chart is empty, confirm there are saved events in that category. - If a summary seems wrong, inspect the raw tracker entries first. - If privacy matters, do not include tracker details in prompts sent to hosted providers.

## Settings: Documents

Route: /docs/settings/documents

Manage document ingestion, extraction, and vector indexing.

# Settings: Documents Documents let Row-Bot search files you add to its local document library. This is different from attaching a file to one chat: indexed documents become reusable context for future questions. Mixedbread Embed Large v1 is the recommended local embedding model; it is separate from the chat model and normal use loads it only from Row-Bot's private cache. ## Where To Find It Open **Settings**, then choose **Documents** in the **Knowledge** group, or search for it with **Find a setting**. ## Controls - **Search runs** chooses where search works: **This computer** (your documents never leave it) or **Cloud** (document text goes to the embedding provider). Switching to Cloud asks first. - **Search model** chooses the embedding model. Mixedbread Embed Large v1 is the recommended local one. - **Free memory when idle** releases the local search model's memory after heavy document work. - **Index health and maintenance** shows the state of your document vectors and memory index: - **Dimension override** is for advanced embedding models that need a specific vector size. - **Document vectors** and **Memory index** each offer **Rebuild**, which recreates them with the current search settings. - **Search model files** offers **Download** to install the selected local model, **Retry** to load an already cached copy again, and **Repair** to replace a damaged download. Download and Repair need network access. - **Upload documents** adds files to Row-Bot-managed storage for indexing. - **Your documents** lists what Row-Bot has processed, with each document's status and progress, and can remove a document after review. ## Common Workflow 1. Under **Search model files**, choose **Download** before adding a large document library. 2. Upload a small test document and wait until it is indexed. 3. Ask a question in a chat that should require the document. 4. Rebuild vectors only when you change the search model or suspect stale results. ## What Is Saved Uploaded files, extracted text, vectors, and local embedding model caches are stored under the active Row-Bot data directory. In Docker, the named /data volume preserves the cache. The active chat only sees relevant results when document search is enabled. A cloud embedding provider is opt-in and sends indexed text to that provider; the recommended local model does not. ## Privacy And Safety Review credential, account, channel, provider, or tool settings before enabling features that can contact outside services. Local-only features stay on your machine until you ask Row-Bot to use a provider, account, channel, MCP server, plugin, or tool that sends data elsewhere. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#documents-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If search misses obvious content, rebuild vectors and check the embedding provider. - If the local model is missing, use **Download** under **Search model files**. Use **Retry** for a cached model or **Repair** for a damaged download. - If local indexing uses too much memory, turn on **Free memory when idle**. - If a document contains private material, remove it from the document library before sharing screenshots or logs.

## Settings: Tools

Route: /docs/settings/tools

Configure progressive external-tool loading, retrieval compression, search and knowledge tools, built-in tools, and custom tools.

# Settings: Tools Tools settings control how Row-Bot exposes enabled external capabilities to the model, its search and retrieval helpers, the small built-in tools it can use, and the custom tools you add. ## Where To Find It Open **Settings**, then choose **Tools** in the **Capabilities** group, or search for it with **Find a setting**. ## Capability Loading - **Auto-select external tools (recommended)** lets Row-Bot search enabled MCP, plugin, Custom Tool, and channel capabilities when a request needs them. Core tools remain available under the active profile. - **Load all external tools** is the eager compatibility mode. It sends every enabled external schema allowed by the current provider and policy, which can consume more context. Both modes preserve tool enablement, provider compatibility, Agent Profile restrictions, workspace policy, approval gates, prompt-injection handling, and execution budgets. ## Other Controls - **Retrieval compression** controls how search results are filtered before they reach the model. - **Search & Knowledge Tools** enables or disables configured research and reference tools such as arXiv, DuckDuckGo, web search, Wikipedia, Wolfram Alpha, and YouTube. Provider-specific setup fields store the credentials required by optional search services. ## Built-In Tools **Built-in tools** are small tools for everyday tasks, such as the calculator, weather, charts and the URL reader. Each has its own switch, and the heading counts how many are on. These used to be a separate Utilities page. Developer tools are among them; the command palette can turn them on ("turn on developer tools"). ## Custom Tools **Custom tools** turns a folder's scripts into tools Row-Bot can run. **Add from a folder** (in the desktop app) inspects the folder and drafts the tool; this sends short excerpts of the folder to your chosen model, which may cost money with a paid provider. You can also ask Row-Bot in a chat to make a tool from a repository, or use the Custom Tool Builder in a code folder's Developer panel. Review a custom tool's commands before turning it on. ## Recommended Workflow 1. Keep **Auto-select external tools** enabled for normal use. 2. Enable only the search and integration tools you intend Row-Bot to use. 3. Confirm external services are healthy before relying on them. 4. Use eager mode temporarily only when testing compatibility with an older model or integration. 5. Review approvals before any action that writes, sends, or changes external state. ## What Is Saved The capability-loading choice and tool settings are local global preferences. Per-task skill activation is stored separately and does not change this tool policy. ## Privacy And Safety Auto mode reduces schemas in the model prompt, but it does not make a networked tool local. When Row-Bot actually calls a hosted model, remote MCP server, plugin service, account integration, or channel, the data required by that operation can leave your machine. Disabled or profile-denied tools are not made available by discovery. ## Troubleshooting - If an external tool is not found, check its plugin, MCP, Custom Tool, or channel configuration and the active Agent Profile. - If a direct tool worked previously but discovery does not select it, use a precise prompt and inspect Row-Bot Status for effective tool groups. - If a provider rejects tool schemas, choose a compatible tool-capable model or temporarily test eager mode. - If retrieval is noisy, change compression mode or disable unnecessary search tools. For the full runtime model, see [Progressive Tools And Skills](/docs/guides/progressive-tools-and-skills).

## Settings: Skills

Route: /docs/settings/skills

Enable, disable, pin, browse, and review Smart Skills.

# Settings: Skills Skills are instruction packs that teach Row-Bot how to handle a type of work. They do not run by themselves; they shape how Row-Bot plans, uses tools, and explains a task. Enabled skills may be searched and loaded automatically for a matching task. ## Where To Find It Open **Settings**, then choose **Skills** in the **Capabilities** group, or search for it with **Find a setting**. ## Controls - Enable and disable controls decide whether Row-Bot may use a skill. - Pinning keeps useful skills easy to find and can make them more likely to be suggested. - **Browse public skills** searches GitHub, ClawHub and LobeHub, or a source URL. Open a skill to read its files before **Install skill**; **Available in chats** (on by default) decides whether conversations may use it once installed. - Skill details show purpose, source, status, and whether the skill is safe to activate. - Search and filters help find skills by task, source, or installed state. - Automatically loaded skills appear as an active chip in the task that selected them. ## Common Workflow 1. Browse or search for a skill that matches the work. 2. Open the detail view before enabling unfamiliar skills. 3. Enable the skill, then start a chat or workflow that names the task. 4. Disable skills you no longer want Row-Bot to consider. ## What Is Saved Skill enablement and pins are local Row-Bot preferences. Automatic selections are stored independently for each parent task or child Agent and are restored when that task reopens. Row-Bot retains up to five automatically selected skills per task; pinned skills are separate. ## Privacy And Safety Review credential, account, channel, provider, or tool settings before enabling features that can contact outside services. Local-only features stay on your machine until you ask Row-Bot to use a provider, account, channel, MCP server, plugin, or tool that sends data elsewhere. A loaded skill cannot grant tools, credentials, approvals, or broader workspace access. The active Agent Profile and runtime policy still decide which tools are actually bound. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#skills-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If Row-Bot ignores a skill, check that it is enabled and relevant to the prompt. - If a skill came from outside the app, review its instructions before enabling it. - If skills make responses too specialized, unpin or disable the extras. - If a skill appears in one task but not another, that is expected for automatic per-task activation. See [Progressive Tools And Skills](/docs/guides/progressive-tools-and-skills) for automatic selection, child isolation, and tool boundaries.

## Settings: Accounts

Route: /docs/settings/accounts

Connect account-level integrations used by tools, setup flows, and channels.

# Settings: Accounts Accounts are service logins Row-Bot can use for tools such as email, calendar, or provider-specific subscription flows. They are separate from ordinary API keys and separate from Row-Bot itself, which does not require a Row-Bot account. ## Where To Find It Open **Settings**, then choose **Accounts** in the **Connections** group, or search for it with **Find a setting**. ## Controls - **GitHub** lets tools and code folders use your GitHub access, such as opening pull requests. **Check** confirms what Row-Bot can reach; public sources use anonymous access when needed. - **Google** connects Gmail and Calendar. Add your Google sign-in file, then sign in in the browser window that opens. - **X** connects your X account for posting and reading. - Each section shows whether the account is connected or needs attention, the result of its last check, and its sign-in, check and disconnect actions. Disconnecting removes Row-Bot's local access to that account. - If a sign-in expires, Monitor and Overview offer **Reconnect** for it. ## Common Workflow 1. Connect only the accounts needed for the tasks you plan to run. 2. Finish the provider sign-in flow in the browser when prompted. 3. Return to Row-Bot and confirm the account shows connected. 4. Review tools and approvals before asking Row-Bot to act through that account. ## What Is Saved Account tokens are stored in the operating system key store when available. Row-Bot keeps local metadata so it can show connection status. ## Privacy And Safety Review credential, account, channel, provider, or tool settings before enabling features that can contact outside services. Local-only features stay on your machine until you ask Row-Bot to use a provider, account, channel, MCP server, plugin, or tool that sends data elsewhere. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#accounts-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If sign-in loops, disconnect and reconnect the account. - If an account tool is unavailable, confirm the account is connected and the related tool is enabled. - If secure storage is unavailable, you may need to sign in again after restart.

## Settings: Channels

Route: /docs/settings/channels

Configure Telegram, WhatsApp, Discord, Slack, SMS, delivery, and health checks.

# Settings: Channels Channels let Row-Bot receive or send messages through external apps. Use them when you want Row-Bot available outside the desktop window or when workflows should deliver results somewhere specific. ## Where To Find It Open **Settings**, then choose **Channels** in the **Connections** group, or search for it with **Find a setting**. ## Controls - Each channel expands into credential fields, status, and controls. - Save stores that channel's settings. - Start and Stop control the channel runtime. - Tunnel settings connect the channel to an externally reachable webhook when required. The tunnel itself is the public link in **Settings Devices & remote access**. - DM Pairing Code helps approve a user before private-message access is allowed. - Paired Users shows who is approved and allows revocation. - Setup Guide explains provider-specific prerequisites. ## Common Workflow 1. Set up the public link in **Settings Devices & remote access** if the channel needs a webhook. 2. Add the channel's required token, URL, or account details. 3. Save, then Start the channel. 4. Pair or approve users before trusting inbound private messages. 5. Use workflow delivery defaults to decide where automated results go. ## What Is Saved Channel settings are global. Messages sent through a channel leave the local app and follow that platform's rules. ## Privacy And Safety Review credential, account, channel, provider, or tool settings before enabling features that can contact outside services. Local-only features stay on your machine until you ask Row-Bot to use a provider, account, channel, MCP server, plugin, or tool that sends data elsewhere. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#channels-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If a channel will not start, check required fields and tunnel status. If it stopped, Monitor offers **Restart** for it. - If messages do not arrive, verify webhook URLs and platform permissions. - If a user should no longer have access, revoke them from Paired Users.

## Settings: MCP

Route: /docs/settings/mcp

Add, browse, import, connect, test and troubleshoot MCP servers and their runtimes in Settings MCP.

# Settings: MCP MCP, the Model Context Protocol, lets Row-Bot use tools provided by another program on this computer or a server online. Treat MCP servers like extensions: only connect servers you trust and understand. ## Where To Find It Open **Settings**, then choose **MCP** in the **Capabilities** group, or search for it with **Find a setting**. ## Switches And Runtimes - **Use MCP servers** is the overall switch for MCP. - **Offer MCP tools in chats** decides whether conversations can use MCP tools. - **Runtimes (Node.js, uv)** shows whether the runtimes many servers need are installed, and installs a managed copy when you ask. ## Your Servers Each saved server shows where it runs (on this computer or online) and its status: **Connected**, **Couldn't connect**, or **On - not connected**. Its main button does the next sensible thing: **Connect**, **Turn on & connect**, or **Disconnect**. Its **** menu offers: - **Details**, a drawer with the server's connection, its tools and their permissions; - **Test connection**; - **Edit settings...** and **Rename...**; - **Remove server...**, which deletes its settings and saved keys and stops its connection. ## Add A Server **Add a server** opens three ways to add one: - **Browse** searches a directory of servers and fills in the details for you. - **Manual** asks for the name, the type (a command on this computer, or a URL), the command and arguments or the address, and any headers or keys. - **Paste JSON** imports a configuration from another app. A new server stays turned off until you connect it, so you can check it first. ## Common Workflow 1. Add or import a trusted server; it is saved turned off. 2. Install a runtime if the server needs one. 3. Choose **Connect**, then open **Details** to review its tools and permissions. 4. Leave the server off, or remove it, if its tools can reach more than you want. ## What Is Saved MCP server definitions are local settings, and their keys use secret storage. A server may still access files, accounts, or networks according to its own implementation, so review its command or URL. ## Privacy And Safety An MCP tool call sends the data that call needs to that server, which may be online. Consequential MCP tools still ask for approval. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#mcp-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If a server shows **Couldn't connect**, use **Test connection**, then check its command, arguments, URL and runtime. - If tools do not appear in chats, check **Use MCP servers**, **Offer MCP tools in chats** and the server's own status. - If a tool asks for risky access, deny the approval and review the server's **Details**.

## Settings: Plugins

Route: /docs/settings/plugins

Manage installed plugins, marketplace installs, configuration, and promoted Custom Tools.

# Settings: Plugins Plugins add local bundles of tools, skills, apps, or integrations. They can make Row-Bot more capable, but they should be treated as code that runs with local app permissions. ## Where To Find It Open **Settings**, then choose **Plugins** in the **Capabilities** group, or search for it with **Find a setting**. ## Controls - Installed plugin cards show version, status, and available actions. - Enable and disable controls decide whether plugin capabilities are active. - Configure opens plugin-specific settings when provided. - Update and remove actions manage local plugin installations. - **Plugin marketplace** browses plugins to install, with an install preview. Every download is checked against its published checksum before it is installed. - Custom Tools promotion turns reviewed local tools into reusable plugin-like capabilities. ## Common Workflow 1. Install plugins only from sources you trust. 2. Read the plugin description and requested fields before enabling it. 3. Configure required settings, then test with a low-risk prompt. 4. Disable or remove plugins you no longer use. ## What Is Saved Installed plugins and plugin settings are local files. Plugin secrets should use secret storage when the plugin supports it. ## Privacy And Safety Review credential, account, channel, provider, or tool settings before enabling features that can contact outside services. Local-only features stay on your machine until you ask Row-Bot to use a provider, account, channel, MCP server, plugin, or tool that sends data elsewhere. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#plugins-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If a plugin does not load, check its manifest and dependency messages. - If plugin tools do not appear, confirm the plugin is enabled and restart if instructed. - If a plugin behaves unexpectedly, disable it first, then inspect its configuration.

## Settings: Preferences

Route: /docs/settings/preferences

Set the assistant's name and personality, learning new skills, how Row-Bot opens, the overnight Dream Cycle, and this device's appearance.

# Settings: Preferences Preferences are personal choices: how Row-Bot introduces itself, how it opens, and when it tidies memory overnight. **Appearance**, the page after it in the **General** group, sets how Row-Bot looks on this device. ## Where To Find It Open **Settings**, then choose **Preferences** in the **General** group, or search for it with **Find a setting**. ## Identity - **Name** is the assistant's name. - **Personality** sets the assistant's default tone; a preview shows how it sounds. - **Learn new skills** lets Row-Bot write and refine skills from your work. Turn it off to keep the skill library exactly as you set it. ## Opening Row-Bot **Open in** chooses whether Row-Bot opens in its own app window, in your browser, or asks each time. ## Overnight Tidy-Up **Dream Cycle** tidies memory while you are away: it merges duplicates, enriches thin memories and infers missing links. Choose the **Hours** it may run (start and end hour). Its runs appear in Monitor's **History**, and Home Knowledge can run it at any time. ## Appearance **Settings Appearance** applies to this device only, so a phone and a computer can look different: - **Appearance**: **System** (follows your device), **Light** or **Dark**. - **Colour theme**: **Blue**, **Teal**, **Violet** or **Amber**. - **Density**: **Comfortable** or **Compact**. - **Reduce transparency** turns off the glass effect. - **Workspace layout Panel sizes** restores panel sizes without changing conversations or the look. A preview shows the result as you choose. In the command palette, "dark mode", "light mode" and "follow the system appearance" change it in one step. ## What Moved - **Updates** (the update channel and checking for new versions) is its own page in the **System** group. See [Operations, Data, And Recovery](/docs/operations/#update-repair-and-uninstall). - **Import from another assistant** (Hermes Agent or OpenClaw) is on **Settings Data**, with back up and restore. See [Operations, Data, And Recovery](/docs/operations/#back-up). ## What Is Saved Preferences are global local settings. Appearance is saved on each device. Existing conversations keep their own model, profile and approval choices. ## Privacy And Safety These settings stay on your computer. Dream Cycle works on local memory and uses your chosen model when it needs one, which may be a hosted provider. ## Control-Level Reference The [generated Settings Controls reference](/docs/reference/generated/settings-controls#preferences-controls) lists every searchable setting on this page with its deep link, search keywords, and any dependency or security notes. ## Troubleshooting - If the app opens in the wrong place, change **Open in** and restart Row-Bot. - If Dream Cycle never runs, check that it is on and that its hours include times your computer is awake. - If the look differs between devices, that is expected: Appearance is saved per device.

## Extend Row-Bot

Route: /docs/extending/

Choose between built-in tools, Skills, Custom Tools, plugins, MCP servers, channels, and accounts.

# Extend Row-Bot Start with the narrowest extension type that can solve the task. Every extension adds capabilities; some also add code, dependencies, network access, credentials, or external side effects. For the conceptual differences between tools, Skills, plugins, MCP, channels, and accounts, read [Extensions And Trust Boundaries](/docs/concepts/extensions-and-trust). | Extension | Use It For | Review Boundary | | --- | --- | --- | | Built-in tool | A capability already shipped and maintained with Row-Bot. | Enable only the tools a profile needs; consequential operations still use approval policy. | | Skill | Reusable instructions and workflow knowledge. | Read the full skill and bundled files before enabling or pinning it. | | Custom Tool | A reviewed local command or repo-specific helper made from a folder in Settings Tools or a code folder's Developer panel. | Inspect its source, command classification, dependencies, and workspace scope. | | Plugin | A packaged bundle that can provide tools, skills, channels, MCP configuration, or UI. | Review manifest, source, permissions, dependencies, and update path before enabling. | | MCP server | Tools exposed by an external local or remote process. | Add disabled, inspect overlap and risk labels, test, then enable only trusted tools. | | Channel | A messaging adapter for receiving or delivering work. | Verify account, recipients, pairing, tunnel, and delivery defaults before starting. | | Account | Authorisation for GitHub, Google, X, or another service used by tools. | Review requested scopes and disconnect unused accounts. | ## Safe Evaluation Sequence 1. Read the extension description and source. 2. Check requested permissions, credentials, network destinations, commands, and install steps. 3. Install into the local data directory only from a source you trust. 4. Keep it disabled while you inspect configuration and provided capabilities. 5. Test with fictional data and a non-consequential task. 6. Enable the minimum capabilities and keep approval mode at **Ask** until behaviour is familiar. 7. Disable or remove extensions that are no longer maintained or needed. Third-party dependency telemetry is not Row-Bot telemetry. Review and accept any dependency behaviour before installation, and never allow it to receive Row-Bot prompts, files, memories, secrets, screenshots, tool arguments, or channel content unless that transfer is the explicit feature you chose.

## Accounts And Authorisation

Route: /docs/integrations/accounts

Connect and disconnect external accounts while keeping scopes, tokens, and data flow understandable.

# Accounts And Authorisation Accounts let selected tools act through services such as GitHub, Google, or X. A model provider account supplies models; an integration account supplies service access. They are configured separately. ## Connect Safely 1. Open **Settings Accounts**, or type "connect google", "connect github" or "connect x" in the command palette. 2. Choose the service and read the status and requested purpose. 3. Prefer the supported OAuth or local CLI flow rather than pasting broad tokens. 4. Verify the service, account, and scopes in the external authorisation page. 5. Return to Row-Bot and refresh status. 6. Test a read-only action before approving writes, posts, email, calendar, or repository changes. ## Tokens And Scopes Secrets use the configured operating-system secret store when available. Settings show status, not the secret value. A service may still retain its own authorisation grant until you revoke it there. Use the narrowest scopes that support the feature and disconnect access you no longer use. ## Troubleshooting - Confirm the system browser returned to the same Row-Bot instance that started the flow. - For GitHub CLI, refresh CLI authorisation after signing in or changing accounts. - If a callback fails, repeat the supported flow instead of copying browser storage. - If an action is denied after connection, review both the service scopes and the Row-Bot approval policy.

## Channels

Route: /docs/integrations/channels

Connect Row-Bot to messaging channels, delivery defaults, pairing, tunnels, and safe external behavior.

# Channels Channels connect Row-Bot to external messaging platforms such as Telegram, WhatsApp, Discord, Slack, and SMS-style providers when configured. Use channels when you want to message Row-Bot outside the desktop app or send workflow results somewhere specific. ## Setup Workflow 1. Open Settings Channels. 2. Expand the channel you want. 3. Add the required token, URL, phone/account detail, or provider-specific field. 4. Set up the public link in Settings Devices & remote access if the channel requires an inbound webhook. 5. Save the channel. 6. Start the channel. 7. Pair or approve users before trusting private messages. 8. Test with a low-risk message. ## Controls - **Save** stores channel settings. - **Start** begins the channel runtime. - **Stop** shuts the channel runtime down. - **Tunnel controls** connect a channel to a public webhook URL when needed. - **DM Pairing Code** approves a user before private-message access. - **Paired Users** lists approved users and lets you revoke access. - **Setup Guide** explains platform-specific prerequisites. - **`/reasoning`** shows or changes the reasoning choice for that channel conversation's active model. Run it without an argument to see the exact valid choices. ## Safety Messages sent through a channel leave the local app. Do not enable a channel until you understand who can message it, what Row-Bot can send back, and whether workflows may deliver results there. When work starts from a channel and a child agent asks for approval, Row-Bot sends the approval back to that parent channel conversation when the channel supports approval messages. The same approval stays visible in the parent conversation, the sidebar's attention indicator and Overview's **Needs you**, and an answer in any of them resolves it everywhere. ## Troubleshooting - If Telegram encounters one transient network failure during initialization, Row-Bot cleans up the partial runtime and retries once. Persistent network failures remain stopped with an actionable diagnostic; invalid bot tokens are not retried. - If Telegram starts but its command menu cannot be registered, polling remains available and the menu failure is logged as non-fatal. - If a channel cannot start, check required fields and tunnel state. Automatic tunnel startup runs outside the UI event loop, so a slow tunnel helper should report status without freezing app startup. - If inbound messages fail, verify webhook URLs and platform permissions. - If the wrong person has access, revoke them from Paired Users.

## Skills Hub

Route: /docs/skills/

Browse, enable, pin, create, edit, and troubleshoot Row-Bot skills.

# Skills Hub Skills are instruction packs that teach Row-Bot how to approach a category of work. **Settings Skills** is where you browse installed skills, inspect details, enable or disable them, pin useful ones, create or edit local skills, and find public skills to install (the command palette's "install a skill" goes there). Enabled skills can also be discovered and loaded progressively when a task clearly matches their purpose. ## UI Walkthrough - **Browse/search** finds bundled, installed, local, and external-source skills. - **Filters** narrow by source, installed state, or task type. - **Skill detail** explains purpose, instructions, source, and status. - **Enable/disable** decides whether Row-Bot may use the skill. - **Pin** keeps an important skill easy to reach. - **Create/edit** lets you maintain a local skill for your own workflow. ## Creating A Skill 1. Decide what task the skill should help with. 2. Write concise instructions, examples, and boundaries. 3. Include when the skill should and should not be used. 4. Save it locally. 5. Enable it and test with a small prompt. 6. Revise if Row-Bot overuses or misunderstands it. ## What Skills Change Skills shape Row-Bot's behavior; they do not automatically grant credentials or bypass approvals. A skill can make Row-Bot more consistent for a task, but tools still need to be enabled and approved where appropriate. ## Automatic Selection When progressive discovery is available, Row-Bot can search enabled skills and load a relevant one for the current task. A successful automatic load appears once as **Using _skill_** and adds an active skill chip. - Automatically selected skills are stored per parent task or child Agent, not per shared Agent Profile. - Reopening a task restores its selected skills. - Parent tasks and sibling children do not share automatic activation state. - Up to five automatically selected skills are retained per task; pinned skills remain separate. - Loading an already active skill does not add a duplicate receipt. Automatic selection changes instructions only. It cannot add a denied tool, alter credentials, bypass approvals, expand a workspace, or change an Agent Profile. Read [Progressive Tools And Skills](/docs/guides/progressive-tools-and-skills) for the complete tool and skill flow. ## Troubleshooting - If a skill is not used, make your prompt match its purpose and confirm it is enabled. - If a skill is too aggressive, disable or rewrite it with clearer boundaries. - If an external skill is unfamiliar, inspect it before enabling.

## Plugins

Route: /docs/integrations/plugins

Install, enable, configure, update, remove, and troubleshoot Row-Bot plugins.

# Plugins Plugins add optional capabilities to Row-Bot: tools, skills, apps, settings, and integration surfaces. Install them only from sources you trust, because they can run code inside the local app environment. ## Setup Workflow 1. Open Settings -> Plugins. 2. Review installed plugins and marketplace options. 3. Read the plugin description and requested configuration. 4. Install or enable the plugin. 5. Configure required fields. 6. Test with a low-risk prompt. 7. Disable, update, or remove the plugin when needed. ## Controls - **Installed plugin cards** show version and status. - **Enable/disable** controls whether plugin capabilities are active. - **Configure** opens plugin-specific settings. - **Update/remove** manage the local installation. - **Plugin marketplace** opens plugin discovery and install previews; every download is checked against its published checksum. - **Custom Tools** can be promoted after review so they behave like reusable app capabilities. ## Safety Plugins can add tools that read files, call services, or create outputs. Keep approval mode active for risky actions and disable plugins you do not actively use. ## Troubleshooting - If a plugin does not load, inspect its manifest and dependency message. - If plugin tools are missing, confirm the plugin is enabled. - If a plugin causes errors, disable it, restart Row-Bot if needed, then review configuration.

## MCP

Route: /docs/integrations/mcp

Configure external MCP servers, test tool availability, and manage MCP safety.

# MCP MCP lets Row-Bot use tools exposed by external Model Context Protocol servers. An MCP server can be local or remote, simple or powerful. Treat it like an extension with its own access and trust boundary. ## Setup Workflow 1. Open Settings MCP. The command palette finds it as "add mcp server". 2. Choose **Add a server**, then **Browse**, **Manual** or **Paste JSON**. 3. Review the command, arguments, URL, transport, and expected tools. A new server is saved turned off. 4. Install a runtime under **Runtimes (Node.js, uv)** if the server needs one. 5. Choose **Test connection**, then **Connect**. 6. Open **Details** to review its tools and permissions, and disconnect it if they reach more than you want. ## Controls - **Use MCP servers** turns MCP on or off; **Offer MCP tools in chats** decides whether conversations can use MCP tools. - **Runtimes (Node.js, uv)** installs the runtimes many servers need. - **Add a server** browses a directory, takes the details by hand, or imports pasted JSON. - Each server's button connects, turns on and connects, or disconnects it, and its status says **Connected**, **Couldn't connect** or **On - not connected**. - Each server's **** menu opens **Details** (connection, tools, permissions), **Test connection**, **Edit settings...**, **Rename...** and **Remove server...**. ## Safety MCP tools can read, write, call APIs, or automate services depending on the server. Row-Bot approvals still matter, but you should also trust the server itself before enabling it. ## Troubleshooting - If a server fails to test, check transport, command, arguments, URL, and dependencies. - If tools are missing, check **Use MCP servers**, **Offer MCP tools in chats** and the server's own status. - If a prompt asks for unexpected access, reject it and inspect the server configuration.

## Voice And Buddy

Route: /docs/voice-and-buddy/

Configure Dictate, Talk, local voice, realtime voice, read-aloud, devices, and Buddy.

# Voice And Buddy Voice and Buddy make Row-Bot feel less like a text box and more like a desktop companion. Voice handles speech input and optional spoken output. Buddy reflects app state and, in the native Windows and macOS apps, can be dragged from the sidebar into a compact always-on-top desktop overlay. ## Voice Modes - **Dictate** turns speech into composer text so you can review it before sending. - **Talk** submits spoken input more directly for conversation. - **Read-aloud** speaks assistant responses when enabled. - **Local voice** uses Whisper by default, or the optional SenseVoice Small engine after you install it explicitly. - **Realtime voice** uses a compatible provider for lower-latency spoken conversation. ## Choosing A Mode Use Dictate when accuracy and review matter. Use Talk when you want a faster hands-light conversation. Use local voice when privacy and offline-style behavior matter. Whisper remains the default local transcription engine. SenseVoice is an optional alternative on Windows, Linux, and Apple Silicon macOS; Intel macOS continues to use Whisper. Use realtime voice when latency matters and you accept internet access, provider requirements, and provider cost. ## Optional SenseVoice Setup Open Settings -> Voice, find **SenseVoice** under **Voice models**, and choose **Install SenseVoice Small**. Then choose it as the Talk or Dictation model under **Speech models**. Installation is an explicit approximately 940 MB download from ModelScope. Row-Bot verifies the contained snapshot before saving it under the active data directory, and normal transcription then uses that local snapshot with update checks disabled. Startup, readiness checks, and ordinary transcription never download or repair the model automatically. ## Devices And Diagnostics Select the microphone and output device in Settings -> Voice. Run diagnostics when audio is silent, delayed, or routed to the wrong device. ## Buddy Controls - **Show Buddy** controls companion visibility. - **Drag to tear off** moves Buddy itself from its sidebar dock into the native desktop overlay. Releasing over the dock cancels the drag. - The overlay follows the selected conversation, whether it is a chat, a design or a code folder, and shares its saved draft, current response, tools, and approval mode. - Enter sends and Shift+Enter adds a line. **Stop** stops the selected thread's active response. - **Simple approvals** can be approved or denied in Buddy. **Complex approvals** open the full thread for review. - **Open full thread**, **Collapse/Expand**, **Dock Buddy**, and **Hide Buddy** are in the overlay menu. - Tear-off is session-scoped; restarting Row-Bot returns placement to the normal sidebar dock. Browser/server mode does not create a native overlay. - **Personality** changes the tone of Buddy cues, and **Bubbles** changes how status appears. - **Look** chooses a bundled or custom appearance; **New look...** makes one with your image model. - **Motion** chooses how lively Buddy is: **Calm**, **Normal** or **Lively**. See [Settings: Buddy](/docs/settings/buddy) for the complete drag, messaging, approval, tray-recovery, and troubleshooting walkthrough. ## Talk To Buddy The desktop overlay accepts typed messages for the thread named in its header. It does not contain a separate microphone control: Talk and Dictate remain in the full Row-Bot thread, where device state, transcription, and provider disclosure are visible. Use **Open full thread**, start Talk or Dictate there, and Buddy will continue reflecting that selected conversation. Use Dictate when you want to review speech as text before sending. Use Talk when you want the configured voice path to submit conversation turns. Undocking Buddy does not change which voice provider is used, create a new voice session, or bypass tool approvals. ## Privacy And Safety Voice input can become chat text. Realtime voice and provider-backed speech may send audio or transcript data to the selected provider. Installing SenseVoice contacts ModelScope for the requested model and standard SDK metadata, but Row-Bot does not send audio, prompts, or usage data; inference stays local after installation. Buddy preferences and assets are local, but custom generation may call a configured media provider. ## Troubleshooting - If Dictate records nothing, check microphone selection and permissions. - If Talk sends too quickly, use Dictate instead. - If SenseVoice reports missing packages, install the `voice` extra for a source checkout or repair the packaged app. If its snapshot is missing or incomplete, reinstall it from Voice settings. - If SenseVoice is unavailable on Intel macOS, select local Whisper instead. - If realtime voice is unavailable, check Providers and Voice Models. - If Buddy does not appear, enable **Show Buddy** or use the tray's **Show Buddy** action when it is torn off. - If Buddy will not undock, use the native Windows or macOS app and start the drag on Buddy itself. Remote browsers and browser/server mode keep it docked.

## Operations, Data, And Recovery

Route: /docs/operations/

Back up, restore, update, repair, and uninstall Row-Bot without losing track of local data and credentials.

# Operations, Data, And Recovery Row-Bot is local-first, so operational safety starts with knowing which data directory and workspace are active. Conversations, settings, memory, Knowledge, workflows, documents, designs, code folder records, skills, plugins, logs, channel state, and MCP configuration can live under that directory. Operating-system key stores may hold credentials separately. ## Remote Access And Server Operations Use [Remote Access And Server Mode](/docs/operations/remote-access) for the complete guide to one-time invitations, desktop and compact owner sessions, Tailscale Serve, LAN, SSH forwarding, Docker, HTTPS reverse proxies, browser-local voice, access recovery, and proxy error diagnostics. Use [Docker And VPS Operations](/docs/operations/docker) for pull-first Compose startup, release and digest pins, persistent volumes, offline backup and restore, explicit upgrade and rollback, host Caddy or Tailscale, secret-file mounts, and container-specific Developer boundaries. Remote access remains off by default in ordinary desktop launches. Server operators should back up access state with the rest of the active data directory, keep one worker, publish one canonical origin, terminate remote traffic with HTTPS, and trust only the exact reverse proxy that connects to Row-Bot. ## Back Up **Settings Data Back up and restore** makes a copy of your Row-Bot on this computer, without passwords or sign-ins. Choose **Back up now**; the command palette finds it as "back up". To copy everything by hand instead: 1. Stop running workflows, channels, servers, and active replies. 2. Close Row-Bot so databases are not changing during the copy. 3. Copy the whole active data directory to protected storage. 4. Back up code folders, exported design files, and wiki vaults separately when they live elsewhere. 5. Record the Row-Bot version used with the backup. ## Restore **Restore from backup** on the same page restores a backup made there; it needs the Row-Bot desktop app and lists what to set up again afterwards, such as passwords and sign-ins. To restore a hand-made copy, restore related databases and files as one set while Row-Bot is closed. Point a test launch at the restored directory first, then check conversations, providers, workflows, Knowledge, documents, designs, code folders, plugins, MCP, and channels. Reconnect secrets that were stored in an OS keychain and were not included in the backup. ## Import From Another Assistant **Settings Data Import from another assistant** scans Hermes Agent or OpenClaw data on this computer and lets you choose what to bring in: memories, persona and skills, and, only if you tick it, API keys and tokens. Nothing is written until you confirm the preview. First launch offers it when it finds one of them. The same page's danger zone holds irreversible clean-up, such as removing tracker data. ## Update, Repair, And Uninstall **Settings Updates** shows the version you have, the update **Channel**, and when Row-Bot last checked. **Check for updates** checks now; when an update is available, the sidebar's attention indicator says so and **Remind me later** hides it for a day on this device. Use the supported package or updater flow for your platform. Back up before a major update. Repair may replace application files but should not be treated as a data backup. Uninstalling the application and deleting the local data directory are separate decisions; review both the application location and the data location before removal. ## Logs And Support Use Monitor and the local log folder to diagnose startup, provider, workflow, channel, and background failures. Before sharing diagnostics, remove names, paths, account identifiers, document content, prompts, message content, tokens, and screenshots that expose private data. ## Release-Specific Manual Checks Installer UX, upgrade, repair, uninstall, Windows signing, macOS notarisation, native file dialogs, OS permissions, and clean-machine behaviour require manual testing on release candidates. A successful docs build does not prove those native flows are release-ready.

## Remote Access And Server Mode

Route: /docs/operations/remote-access

Use Row-Bot as one owner from trusted devices through Tailscale, LAN, SSH, Docker, or an HTTPS reverse proxy.

# Remote Access And Server Mode Row-Bot Remote Access lets one owner use the same instance from several trusted browsers. It is a **single-owner, multi-device** system, not a multi-user service or a hostile tenant-isolation boundary. Reachability, authentication, and presentation are separate: - A network route lets a browser reach Row-Bot's neutral connection page. - A one-time invitation authorizes one browser to create a revocable session. - Every authenticated interactive browser represents the same owner. - Desktop or compact presentation changes layout only, never authority. Normal desktop launches remain local-first. They listen on loopback and keep the local owner experience unless you explicitly enable another route. Authenticated server mode is stricter: every browser must connect with a valid session, including a browser on the server itself. ## Configure Remote Access In The App Open **Settings Devices & remote access** from any signed-in owner session, including on a phone. The command palette finds it as "phone" or "connect device". 1. Choose **Connect a phone or computer**. 2. Choose **How will it reach this computer?**: **Tailscale** (private and encrypted HTTPS, recommended), **Same Wi-Fi** (a direct address on your network, over HTTP), **Internet** (from anywhere, through your public link), or **Another address** (a DNS alias, hostname or other browser-facing address you set up yourself, entered as its exact HTTP or HTTPS origin). 3. Choose how long it should **Stay signed in**: **30 days** or **12 hours**. 4. **Scan the code** on the other device, or open the shown address there. Codes work only once; **Show a new code** replaces an expired one. 5. Review the confirmation page on that device, then press **Connect**. The page on the computer then says **Connected**, and **Connect another** starts again. 6. **Your devices** lists every signed-in phone and computer and signs one out. Row-Bot saves a new address, applies its exact Host/origin admission rule immediately, and creates the origin-bound code without a restart. **Same Wi-Fi** first asks to let devices on your network reach Row-Bot, which restarts it to listen on your network. Under **Advanced**, **Network** shows where Row-Bot listens (**This computer only**, or **This computer and my network**) and its **Allowed addresses**, and **Public link** sets up the ngrok tunnel, which opens only when you start it. The choice of routes refreshes from current state when the code is created. Row-Bot refuses a stale or no-longer-reachable selection instead of issuing a misleading link. Saved trusted addresses remain in `access_routes.json` across restarts and appear among the ways to reach this computer. Remove one from **Allowed addresses** when it is no longer needed; new HTTP and WebSocket requests through that exact address are rejected immediately. Removing the address can interrupt current access through it. Enter one exact `http://` or `https://` origin with an optional port, such as `http://row-bot.lan:8080` or `https://row-bot.example.com`. Credentials, paths, queries, fragments, wildcards, malformed ports, and non-HTTP(S) schemes are rejected. A non-default port trusts only that exact authority, not another port. Trusting an address changes only Row-Bot admission and invitation routing. Row-Bot performs no DNS lookup or reachability probe and does not create DNS records, provision TLS, configure a proxy, open a firewall, or change the listen address. HTTP traffic is unencrypted. HTTPS still requires external DNS and TLS setup; a TLS-terminating reverse proxy must also use the existing exact trusted-proxy configuration. Origins supplied through `ROW_BOT_PUBLIC_ORIGINS` are listed as **Managed externally** and cannot be edited in the UI. When `ROW_BOT_ALLOWED_HOSTS` is set, deployment configuration owns Host admission, so trusted-address add/remove controls are read-only and do not override it. ## Invitations, Sessions, And Layouts | Item | Meaning | | --- | --- | | Invitation | A secret link that expires after 10 minutes and can be consumed once. Opening or previewing it does not connect the device; the recipient must press **Connect**. | | Session | A separate HttpOnly browser credential created after acceptance. Raw session secrets are never placed in the URL. | | Trusted device | A session lasting up to 30 days. An active owner UI checks at startup and every 12 hours, renewing the session to 30 days when it enters the final seven days. | | Temporary access | A session lasting up to 12 hours. | | Desktop layout | The full owner product in its wide-screen presentation. | | Compact layout | The same owner authority in a phone-safe presentation, including every Settings page. Design and Developer panels open as full-screen sheets. | Row-Bot stores only hashed invitation and session secrets in the versioned access database. Sessions survive normal restarts and can be revoked individually. Revoking a device revokes its sessions. An inactive trusted browser can still expire after 30 days; recover by creating a new one-time invitation from a trusted local terminal, SSH session, or container exec. Temporary 12-hour sessions and migrated legacy sessions retain their fixed expiry and do not renew. Row-Bot has no guest or multi-user sharing boundary. Device names, user agents, routes, and layout choices are session metadata, not entitlements. Existing action approval gates continue to govern sensitive tools and operations. Treat an unused invitation like a password. Do not paste it into chat, logs, issue reports, screenshots, shell history you plan to share, or a public URL. ## Choose A Connection Route | Route | Best fit | Important boundary | | --- | --- | --- | | Local desktop | Browser and Row-Bot are on the same computer. | Default desktop mode stays on loopback and does not expose a remote listener. | | Tailscale Serve | Recommended private HTTPS access across your tailnet. | Tailnet membership does not replace a Row-Bot invitation or session. | | Direct LAN | A private, trusted local network. | Remote LAN HTTP is unencrypted and does not provide a secure browser context for microphone capture. | | Trusted hostname or address | An existing exact DNS alias, hostname, or browser-facing IP and port. | Row-Bot trusts only the saved origin; it does not provide DNS, TLS, proxy, firewall, or reachability setup. | | SSH tunnel | An operator can SSH to a remote host and wants no published Row-Bot port. | Keep Row-Bot on loopback and create the invitation for the workstation-facing forwarded origin. | | Docker | A headless, isolated, reproducible instance. | The supplied Compose file publishes to host loopback by default and persists `/data`. | | HTTPS reverse proxy | An operator owns DNS, TLS, proxy configuration, backups, and ingress policy. | Trust only the exact proxy address or CIDR that connects to Row-Bot. | An active route means the connection page is reachable. It does not mean the browser is trusted. ## Tailscale Serve Tailscale Serve is the recommended desktop route because it provides a private HTTPS origin without changing Row-Bot's default bind address. Row-Bot does not install Tailscale, sign you in, enable Funnel, reset the complete Serve configuration, change firewall rules, or overwrite an unrelated route. Status checks and planning are read-only. Enabling a route requires you to review the exact private Serve change, accept the third-party disclosure, and confirm it in the UI. Row-Bot refuses to mutate when Funnel is active, a conflicting route exists, ownership cannot be verified, or the Tailscale CLI outcome is ambiguous. It records ownership only after it verifies the exact route it created, and it removes only that unchanged owned route. After a verified enable or disable, Row-Bot asks its launcher to restart the app. That restart activates the matching proxy policy: - only loopback connections from the local Tailscale proxy are trusted; - only the verified owned `.ts.net` host and HTTPS origin are added; - the owned route must still target this exact Row-Bot app port. If the launcher is unavailable, the UI tells you to restart Row-Bot manually. A manually managed Tailscale or reverse-proxy route is not treated as Row-Bot-owned; configure its public URL, allowed host, and exact proxy trust explicitly. After Tailscale reports **Active Owned**, create the invitation for its verified private origin. The recipient still needs the one-time invitation and resulting Row-Bot session. ## Direct LAN Use **Allow local-network connections** only on a network you trust. The launcher restarts Row-Bot after the durable listen setting changes. Anyone who can reach the port can see the neutral connection page, but cannot use the application without a valid session. LAN HTTP is unencrypted. Prefer Tailscale or an operator-managed HTTPS origin on shared, guest, workplace, or otherwise untrusted networks. Row-Bot does not change the host firewall. ## SSH Tunnel Start Row-Bot on loopback: ```bash row-bot serve --host 127.0.0.1 --port 8080 ``` Forward a workstation port: ```bash ssh -N -L 18080:127.0.0.1:8080 user@row-bot-host ``` Create the invitation for the origin the browser will actually use: ```bash row-bot access invite \ --layout desktop \ --origin http://127.0.0.1:18080 ``` ## Advanced / Managed Deployments Environment variables and CLI flags remain useful when deployment tooling must own policy. `ROW_BOT_PUBLIC_ORIGINS` entries are visible but read-only in Settings. Setting `ROW_BOT_ALLOWED_HOSTS` makes Host admission externally managed and disables trusted-address mutation in the UI; durable UI addresses never override that explicit policy. ### Authenticated Server Mode `row-bot serve` starts one authenticated headless instance. It defaults to loopback, one worker, no tray, no splash, no opened browser, and no automatic Ollama startup. For a host-installed HTTPS reverse proxy: ```bash row-bot serve \ --host 127.0.0.1 \ --port 8080 \ --public-url https://row-bot.example.com \ --allowed-host row-bot.example.com \ --trusted-proxy 127.0.0.1/32 ``` The equivalent core environment settings are: ```text ROW_BOT_DEPLOYMENT_MODE=server ROW_BOT_PUBLIC_URL=https://row-bot.example.com ROW_BOT_ALLOWED_HOSTS=row-bot.example.com ROW_BOT_TRUSTED_PROXY_CIDRS=127.0.0.1/32 ROW_BOT_WORKERS=1 ``` Use the address that actually connects from the proxy to Row-Bot. Do not trust every private address, a whole Docker bridge range, or caller-supplied forwarding headers. Do not buffer the event stream (`/api/v1/events`, `text/event-stream`) and terminate remote traffic with HTTPS. The reviewed [Caddy example](https://github.com/siddsachar/row-bot/blob/main/deploy/reverse-proxy/Caddyfile.example) and [systemd service](https://github.com/siddsachar/row-bot/blob/main/deploy/systemd/row-bot.service.example) are starting points, not substitutes for operator-owned firewall, rate-limit, logging, backup, and recovery policy. The deprecated `row-bot --server --no-open` form remains compatible, but new scripts and services should use `row-bot serve`. ## Docker The supplied Docker image contains the complete supported server feature set: all canonical Python extras, matching headless Playwright Chromium, native media libraries, `uv`/`uvx`, and pinned Node.js with `node`/`npm`/`npx`. It does not contain invitations, sessions, credentials, downloaded voice models, arbitrary MCP servers, or Tailscale. Compose publishes `127.0.0.1:8080` by default and persists private state and explicit model/runtime downloads under `/data`. Every browser must claim an invitation, including a browser on the Docker host. Docker bridge and gateway addresses never grant owner identity. Follow [Docker And VPS Operations](/docs/operations/docker) for the pull-first public runbook, and use the repository's [Docker operator notes](https://github.com/siddsachar/row-bot/blob/main/deploy/docker/README.md) when building from source. The public guide covers: - invitation bootstrap with `docker compose exec`; - intentional LAN or HTTPS publication; - multiple isolated instances; - health and readiness checks; - read-only secret files; - backups, upgrades, restore, and rollback; - trusted-session renewal and inactive-session recovery; - browser, MCP runtime, local-model cache, and headless feature boundaries. Never mount the host Docker socket merely to satisfy a tool or MCP server. ## Voice From A Remote Browser Local browser voice uses the requesting browser's microphone and speaker. The browser sends bounded audio to Row-Bot for transcription with the selected local STT engine: Whisper by default, or SenseVoice after its explicit installation. Row-Bot returns local Kokoro audio to that same browser. A headless server or container does not need access to the host microphone or speakers. Browsers permit microphone capture on `localhost` or in a secure HTTPS context. Plain HTTP on a remote LAN address is insufficient. Use Tailscale HTTPS or a correctly configured HTTPS reverse proxy for remote dictation or talk. Whisper, SenseVoice, and Kokoro model installation remains an explicit owner action and can use several gigabytes of persistent cache. SenseVoice alone is an approximately 940 MB ModelScope download and is unavailable on Intel macOS. OpenAI Realtime voice is a separate provider/network feature; local browser voice does not silently fall back to it. ## Offline Access Management And Recovery Use a trusted local terminal, SSH session, or `docker compose exec`: ```bash row-bot access invite --layout desktop --origin https://row-bot.example.com row-bot access invite --layout compact --temporary --origin https://row-bot.example.com row-bot access list row-bot access revoke DEVICE_ID row-bot access revoke SESSION_ID --session row-bot access revoke-all --yes row-bot access doctor ``` `list` and `doctor` never print reusable session secrets. Invitation commands are the only commands that print a raw one-time link. If every owner session is lost, create a new owner invitation from a trusted terminal. Back up the complete active `ROW_BOT_DATA_DIR` while Row-Bot is stopped. It contains conversations, configuration, instance identity, invitations, devices, sessions, and other private state; operating-system keyring credentials may require a separate recovery plan. ## Troubleshooting - **`untrusted_forwarding_headers`**: a proxy supplied forwarding headers from an address Row-Bot does not trust. For a Row-Bot-owned Tailscale route, restart once if the UI reports that restart is required. For an operator proxy, set only its exact connecting address in `ROW_BOT_TRUSTED_PROXY_CIDRS`. Do not fix this by trusting a broad network. - **`unexpected_host`**: on a desktop launch, open **Settings Devices & remote access**, add the exact browser-facing origin (as **Another address** when connecting a device, or under **Advanced Network Allowed addresses**), and retry without restarting. In a managed deployment, the host may instead be missing from `ROW_BOT_ALLOWED_HOSTS`, or the proxy may have forwarded a different host. - **Origin or invitation mismatch**: create the invitation for the exact browser-facing scheme, host, and port. Do not reuse an invitation after changing routes. - **Tailscale conflict or unowned route**: inspect the existing Serve configuration. Row-Bot will not reset or overwrite it. - **Phone reaches the page but cannot use Row-Bot**: reachability is working; create or reclaim the correct invitation and confirm the device was not revoked. - **Remote microphone unavailable**: use HTTPS, grant microphone permission in that browser, and install the local voice model explicitly. - **Server starts but is not ready**: check `/healthz`, `/readyz`, local logs, and `row-bot access doctor`. For command lookup, see the generated [CLI reference](/docs/reference/generated/cli) and [environment reference](/docs/reference/generated/environment-and-config).

## Docker And VPS Operations

Route: /docs/operations/docker

Run one persistent Row-Bot server with Docker Compose, private host access, or an operator-managed HTTPS edge.

# Docker And VPS Operations The official container runs one **single-owner, multi-device** Row-Bot instance. It is not a multi-user service or a hostile tenant-isolation boundary. Every browser, including one on the Docker host, must claim a one-time invitation before it receives a revocable owner session. The default Compose topology publishes only `127.0.0.1:8080`, runs the application as UID/GID 10001 with a read-only root filesystem, and keeps private state in separate named data and encryption-key volumes. It does not mount the Docker socket, host devices, a desktop display, or host audio. ## Prerequisites - Docker Engine or Docker Desktop with Linux containers and Docker Compose. - Enough disk for the full image plus the `/data` volume. Models, managed runtimes, browser outputs, and caches can grow after explicit use; Row-Bot does not promise one fixed capacity figure. - For public VPS access: Linux, Docker Compose 2.24.4 or newer, DNS you control, and a host-managed HTTPS proxy or Tailscale installation. Release images are built from tagged source in GitHub Actions with the locked dependency set. Native amd64 and arm64 jobs run the isolated authenticated-container smoke before the release manifest is created, and the registry exposes a digest for immutable pins. The smoke uses bounded request timeouts and requires two consecutive readiness samples; it retries only safe GET probes, never repeats a state-changing POST, and limits failure output to redacted container state and log tails. ## Download A Release-Pinned Compose File You do not need to clone the repository. Choose a released version, create a private deployment directory, and download the Compose file from the same tag: ```bash mkdir -p row-bot-docker cd row-bot-docker curl -fsSLo compose.yaml \ "https://raw.githubusercontent.com/siddsachar/row-bot/v${ROW_BOT_VERSION}/deploy/docker/compose.yaml" docker buildx imagetools inspect "${ROW_BOT_IMAGE}" docker compose -f compose.yaml config ``` Replace `X.Y.Z` with a release that lists the Row-Bot container in GitHub Packages. A source tag can exist without a published container, so confirm the image before starting. Pull requests and manually dispatched container checks verify builds but do not publish them; publishing happens only after a GitHub Release is published. If the registry reports `manifest unknown` for the version you chose, select a release with a container or use [Build From Source](#build-from-source). Keep `ROW_BOT_VERSION`, the downloaded file, and `ROW_BOT_IMAGE` aligned. The Compose default is `ghcr.io/siddsachar/row-bot:latest`, but an explicit version is easier to audit and roll back. Stable releases also update `latest`; prereleases do not. For an immutable image pin, replace the tag with the digest published for that release: ```bash docker compose -f compose.yaml config ``` ## Account Credential Persistence Is Automatic Headless containers do not have a desktop keyring. On every `docker compose up`, a short-lived, network-disabled initializer creates a random `ROW_BOT_SECRET_STORE_KEY` only when the project-scoped `row_bot_secrets` volume is empty. The application receives that volume read-only and uses the key to encrypt ChatGPT/Codex tokens and other owner-entered secrets under `/data/secure-secrets`. There is no key to create or environment value to set for the normal Compose path. Keep both named volumes for restart, container replacement, upgrade, and rollback, and back up the key separately from `/data`. Advanced unattended deployments can replace the generated key volume with an operator-managed read-only directory as described in [Read-Only Secret Files](#read-only-secret-files). ## First Start And Owner Invitation A fresh `up` pulls the image only when it is absent locally, creates the data and encryption-key volumes, initializes the key without network access, and starts the foreground server: ```bash docker compose -f compose.yaml up -d docker compose -f compose.yaml ps curl -fsS http://127.0.0.1:8080/healthz curl -fsS http://127.0.0.1:8080/readyz ``` The unauthenticated root shows a neutral connection flow. Create the first invitation from a private terminal; the output is a one-time secret: ```bash docker compose -f compose.yaml exec row-bot \ row-bot access invite --layout desktop --origin http://127.0.0.1:8080 ``` Open the printed URL, review the confirmation, and select **Connect**. Use `--layout compact` for phone presentation. Both layouts grant the same owner authority. Useful secret-free checks are: ```bash docker compose -f compose.yaml exec row-bot row-bot access list docker compose -f compose.yaml exec row-bot row-bot access doctor --host 127.0.0.1 docker compose -f compose.yaml logs --tail 200 row-bot ``` Do not paste invitation URLs, cookies, provider credentials, or unredacted logs into issues or chat. ## Finish Setup And Install Private Knowledge Search First launch is the same in desktop, source, and Docker installations: answer **How should Row-Bot think?** to choose your first model. Private knowledge search uses **Mixedbread Embed Large v1**, a separate local model for semantic memory and document search, as a 675 MB download you start from Settings -> Documents (Setup Center links there). The download requires internet access to Hugging Face and does not upload documents or memories. In Docker, the downloaded files live in the `/data` cache, so the named volume preserves them across restarts and container replacement. Install it for the recommended semantic search experience. Until then, Row-Bot still works and uses bounded lexical and graph fallback; install the model from [Settings -> Documents](/docs/settings/documents) under **Search model files**. After changing an embedding provider or model, rebuild the document and memory vectors from this screen. A cloud embedding provider is a separate opt-in choice and sends the indexed text to that provider; the recommended default stays local after the initial download. ## Stop, Restart, Reboot, And Replace Normal stop and start retain the container and named volume: ```bash docker compose -f compose.yaml stop row-bot docker compose -f compose.yaml start row-bot ``` `restart: unless-stopped` lets an existing service return when the Docker daemon starts after a host reboot. After every reboot, verify `ps`, `/healthz`, `/readyz`, and an existing authenticated browser. Replacing only the container preserves the same instance identity, settings, sessions, conversations, and selected caches in `/data`: ```bash docker compose -f compose.yaml up -d --force-recreate row-bot ``` ## Multiple Isolated Instances Use a different Compose project name and host port for each instance. Compose project names isolate the containers, networks, and named data and encryption-key volumes: ```bash docker compose --project-name row-bot-main \ -f compose.yaml up -d ROW_BOT_HOST_PORT=8081 docker compose --project-name row-bot-lab \ -f compose.yaml up -d ``` Create invitations for `http://127.0.0.1:8080` and `http://127.0.0.1:8081`, respectively. Every project automatically receives its own data and encryption-key volumes. Do not copy one instance's access database or encryption key into another unless you are deliberately restoring that same instance. Each fresh instance has a different identity, so its cookies and owner sessions remain isolated even when the hostname is the same. ## Consistent Offline Backup Treat a backup as private user data. Stop the service before copying the whole volume so related SQLite files are one consistent set. Use a disposable helper container to create archives; do not use `docker compose cp` for the complete `/data` tree because model caches can contain symbolic links that a Windows host cannot reproduce. The helper has no network, receives the source volumes read-only, and adds only the capability needed to read every file in the stopped volume: ```bash backup_dir="$(pwd)/backups/row-bot-$(date -u +%Y%m%dT%H%M%SZ)" mkdir -p "$backup_dir" docker compose -f compose.yaml stop row-bot row_bot_container="$(docker compose -f compose.yaml ps -a -q row-bot)" docker run --rm \ --network none \ --read-only \ --user 0:0 \ --cap-drop ALL \ --cap-add DAC_READ_SEARCH \ --security-opt no-new-privileges \ --volumes-from "${row_bot_container}:ro" \ --mount "type=bind,src=${backup_dir},dst=/backup" \ --entrypoint sh \ "${ROW_BOT_IMAGE}" \ -c 'set -eu; tar -C /data -czf /backup/data.tar.gz .; tar -C /run/secrets -czf /backup/secrets.tar.gz ROW_BOT_SECRET_STORE_KEY' docker compose -f compose.yaml start row-bot ``` Run the final `start` command even if archive creation reports an error. On Windows PowerShell, use `$backupDir = (New-Item -ItemType Directory -Force "backups\\row-bot-$((Get-Date).ToUniversalTime().ToString('yyyyMMddTHHmmssZ'))").FullName`, `$rowBotContainer = docker compose -f compose.yaml ps -a -q row-bot`, backticks instead of backslashes for line continuation, `$backupDir` in the bind mount, and `$env:ROW_BOT_IMAGE` as the image value. The archive work still runs inside Linux and therefore preserves cache symlinks. Encrypt and protect both `data.tar.gz` and `secrets.tar.gz` with your normal backup system. Record the image tag and digest beside them. Encrypted server credentials under `/data/secure-secrets` are usable only with the matching `ROW_BOT_SECRET_STORE_KEY`; protect both archives. Other credentials supplied through an operator-managed read-only directory need their own recovery procedure. Test a restore under a different Compose project name and port. This creates a separate named volume and does not overwrite the live instance: ```bash restore_backup_dir="$(cd ./backups/RESTORE_SET && pwd)" docker compose --project-name row-bot-restore -f compose.yaml \ create secret-store-init row-bot restore_app_container="$(docker compose --project-name row-bot-restore -f compose.yaml ps -a -q row-bot)" restore_init_container="$(docker compose --project-name row-bot-restore -f compose.yaml ps -a -q secret-store-init)" docker run --rm \ --network none \ --read-only \ --user 0:0 \ --cap-drop ALL \ --cap-add CHOWN \ --cap-add FOWNER \ --cap-add DAC_OVERRIDE \ --security-opt no-new-privileges \ --volumes-from "${restore_app_container}" \ --mount "type=bind,src=${restore_backup_dir},dst=/backup,readonly" \ --entrypoint sh \ "${ROW_BOT_IMAGE}" \ -c 'set -eu; tar -C /data -xzf /backup/data.tar.gz' docker run --rm \ --network none \ --read-only \ --user 0:0 \ --cap-drop ALL \ --cap-add CHOWN \ --cap-add FOWNER \ --cap-add DAC_OVERRIDE \ --security-opt no-new-privileges \ --volumes-from "${restore_init_container}" \ --mount "type=bind,src=${restore_backup_dir},dst=/backup,readonly" \ --entrypoint sh \ "${ROW_BOT_IMAGE}" \ -c 'set -eu; tar -C /run/secrets -xzf /backup/secrets.tar.gz' docker compose --project-name row-bot-restore -f compose.yaml up -d curl -fsS http://127.0.0.1:18080/readyz docker compose --project-name row-bot-restore -f compose.yaml exec row-bot \ row-bot access doctor --host 127.0.0.1 ``` Set `ROW_BOT_IMAGE` to the recorded image before restoring. On PowerShell, resolve the backup directory with `$restoreBackupDir = (Resolve-Path '.\\backups\\RESTORE_SET').Path`, capture the two container IDs with `$restoreAppContainer` and `$restoreInitContainer`, and use those variables in the same helper commands. Inspect the restored data and authentication before changing any proxy or production route. Remove the isolated project with `docker compose --project-name row-bot-restore -f compose.yaml down --volumes` only after you have finished the restore test. ## Explicit Upgrade And Rollback Upgrades are never automatic. Take an offline backup, choose the new version or digest, pull it explicitly, recreate, and run checks: ```bash docker compose -f compose.yaml pull row-bot docker compose -f compose.yaml up -d row-bot docker compose -f compose.yaml ps curl -fsS http://127.0.0.1:8080/readyz docker compose -f compose.yaml exec row-bot \ row-bot access doctor --host 127.0.0.1 ``` Also verify an existing browser session and a normal streamed response. For rollback, use the previous image digest with its matching pre-upgrade data backup in an isolated project first. Database changes can make an old image incompatible with state already opened by a newer one. ## Read-Only Secret Files The default stack already provides encrypted persistence for secrets entered in the UI. For centrally managed or pre-provisioned deployments, you can instead keep one allowlisted setting value per file in an absolute private host directory outside the repository. This override replaces the automatic key volume, so the directory must also contain `ROW_BOT_SECRET_STORE_KEY`. Download the example override from the same release tag: ```bash curl -fsSLo compose.secrets.yaml.example \ "https://raw.githubusercontent.com/siddsachar/row-bot/v${ROW_BOT_VERSION}/deploy/docker/compose.secrets.yaml.example" sudo install -d -o 10001 -g 10001 -m 0700 "$ROW_BOT_SECRETS_HOST_DIR" openssl rand -hex 32 | sudo tee \ "$ROW_BOT_SECRETS_HOST_DIR/ROW_BOT_SECRET_STORE_KEY" >/dev/null sudo chown 10001:10001 \ "$ROW_BOT_SECRETS_HOST_DIR/ROW_BOT_SECRET_STORE_KEY" sudo chmod 0400 \ "$ROW_BOT_SECRETS_HOST_DIR/ROW_BOT_SECRET_STORE_KEY" docker compose -f compose.yaml -f compose.secrets.yaml.example config docker compose -f compose.yaml -f compose.secrets.yaml.example up -d ``` Restrict the host directory to the Docker operator and container UID/GID 10001. The override mounts only `/run/secrets:ro`; it does not put values into Compose environment, labels, command lines, or image layers. Externally managed provider and channel values remain read-only in Settings and are never copied into `/data`. `ROW_BOT_SECRET_STORE_KEY` is different: it is a 32-byte master key encoded as exactly 64 hexadecimal characters. The default initializer generates it automatically in the project-scoped key volume; the override above makes you its operator. When a desktop keyring is unavailable, Row-Bot uses it to encrypt in-app OAuth tokens and other owner-entered secrets under `/data/secure-secrets`. This allows ChatGPT/Codex and other rotating OAuth credentials to survive a process restart or container replacement without placing the tokens in environment variables. The master key itself remains only in `/run/secrets` and is never copied into `/data`. Keep the same key for backup, restore, upgrade, and rollback. A missing key leaves newly entered credentials session-only; an invalid or changed key fails closed and cannot decrypt, replace, or delete existing encrypted records. There is no automatic key rotation. To replace a lost key, reconnect the affected accounts and re-enter secrets after deliberately starting with an empty encrypted secret directory. ## Build From Source Source builds are opt-in and require a repository checkout. The build override changes only the image source; the base file remains authoritative for runtime and security settings: ```bash git clone https://github.com/siddsachar/row-bot.git cd row-bot docker compose \ -f deploy/docker/compose.yaml \ -f deploy/docker/compose.build.yaml \ build row-bot docker compose \ -f deploy/docker/compose.yaml \ -f deploy/docker/compose.build.yaml \ up -d ``` Record the source revision and resulting image digest before relying on a source build. ## Linux VPS With Host Caddy This topology uses host networking only in the VPS override. Row-Bot binds `127.0.0.1:8080`; host Caddy terminates HTTPS on ports 80/443 and connects from IPv4 loopback. Nothing publishes container port 8080 externally. 1. Point the dedicated DNS name at the VPS. 2. Allow inbound TCP 80/443 in the host and provider firewalls; do not expose 8080. 3. Install and operate Caddy on the host. 4. Download the release-matched override and Caddy example: ```bash curl -fsSLo compose.vps.yaml \ "https://raw.githubusercontent.com/siddsachar/row-bot/v${ROW_BOT_VERSION}/deploy/docker/compose.vps.yaml" curl -fsSLo Caddyfile.row-bot \ "https://raw.githubusercontent.com/siddsachar/row-bot/v${ROW_BOT_VERSION}/deploy/reverse-proxy/Caddyfile.example" ``` Set the exact browser-facing origin and host, then validate the merged configuration: ```bash docker compose -f compose.yaml -f compose.vps.yaml config docker compose -f compose.yaml -f compose.vps.yaml up -d curl -fsS http://127.0.0.1:8080/readyz ``` The override sets `ROW_BOT_TRUSTED_PROXY_CIDRS=127.0.0.1/32` exactly. Replace the example hostname in `Caddyfile.row-bot`, integrate that site block into the host's Caddy configuration, then validate and reload Caddy using the host package's service procedure. Do not broaden proxy trust to a private range or accept caller-supplied forwarding metadata. Create the invitation for the exact HTTPS origin: ```bash docker compose -f compose.yaml -f compose.vps.yaml exec row-bot \ row-bot access invite --layout desktop \ --origin https://row-bot.example.com ``` From an external browser, confirm the HTTPS connection page, claim the invitation, keep the UI open through a streamed response, and verify its event stream stays connected. Reboot the VPS during a maintenance window and repeat DNS, TLS, readiness, authenticated-session, and streaming checks. ## Host-Managed Tailscale Serve Tailscale runs on the Linux host, not in the Row-Bot container. Use the VPS override so host Tailscale reaches Row-Bot over exact loopback, but do not install or invoke the Tailscale CLI inside the container. Set the host's actual private HTTPS name: ```bash docker compose -f compose.yaml -f compose.vps.yaml up -d sudo tailscale serve --bg http://127.0.0.1:8080 tailscale serve status ``` Create an invitation for that exact `https://...ts.net` origin. Tailnet membership supplies private reachability; it does not replace Row-Bot's invitation, session, or revocation controls. The container neither contains nor controls the host Tailscale CLI, Funnel, login, firewall, or route lifecycle. ## Trusted Sessions And Recovery A trusted browser session lasts up to 30 days. While an authenticated owner UI remains active, the browser checks at startup and every 12 hours. In the final seven days, the server atomically renews that trusted session to 30 days. Temporary 12-hour sessions and migrated legacy sessions never renew. An inactive trusted browser can still expire after 30 days. If all sessions are lost, recover only from a trusted host terminal or SSH session: ```bash docker compose -f compose.yaml exec row-bot \ row-bot access invite --layout desktop --origin YOUR_EXACT_ORIGIN ``` Revoked, expired, or device-revoked credentials stay invalid. ## Developer And Headless Boundaries There are three distinct Developer execution cases: 1. **Row-Bot installed on a host:** Docker Sandbox can use the host's supported Docker runtime when it is available. Local mode remains an explicit workspace choice. 2. **The official Row-Bot application container:** Developer Docker Sandbox is unavailable and fails closed. The image contains no Docker daemon or CLI, and you must not mount the host Docker socket. Local mode can use only a workspace path you explicitly mount into the application container. 3. **An approved Custom Tool inside the application container:** risky approved execution deliberately uses Local mode inside that same container against the selected visible path. It is not a nested Docker sandbox and is not a silent fallback from a requested Docker workspace. The container has no native tray/window, native desktop Computer Use, physical host microphone or speaker, camera, display server, or GPU by default. Browser-local voice and bundled headless browser automation remain available. Selected local models and runtimes are explicit downloads into `/data`; cache growth depends on what the owner installs and uses. ## Remove The Deployment Ordinary removal keeps private state: ```bash docker compose -f compose.yaml down ``` To delete both named volumes and all instance state deliberately, first verify the Compose project and backup, then run: ```bash docker compose -f compose.yaml down --volumes ``` Those volume deletions remove both private data and the generated encryption key and are not recoverable unless you have a tested backup. External secret directories and backup directories are separate and are never removed by these Compose commands.

## Privacy And Safety

Route: /docs/privacy-safety/

Understand local data, provider calls, credentials, approvals, channels, MCP, plugins, and sharing.

# Privacy And Safety Row-Bot is local-first: the app and its data live on your machine. Local-first does not mean every feature is offline. Hosted models, web search, browser actions, account tools, channels, MCP servers, plugins, realtime voice, and media providers can send data outside the app when you configure and use them. ## Local Data Conversations, memories, documents, workflows, logs, designs, code folder records, skills, plugins, Buddy assets, and settings are stored in the active local Row-Bot data directory. Buddy's native overlay reuses the selected conversation and does not create a second transcript store. ## External Calls External calls happen when you choose or enable something that needs them: hosted models, subscription providers, API providers, web search, browser automation, account tools, messaging channels, MCP servers, plugin tools, realtime voice, and media generation. OpenRouter requests identify Row-Bot as the app making them (its website address and name, in the headers OpenRouter asks apps to send), which lets OpenRouter show Row-Bot in its public app rankings; these headers add no user, prompt or account data. Computer Use is a distinct opt-in boundary. Row-Bot downloads the pinned Cua Driver only after an explicit Install or Repair action, verifies the selected archive, and requires the current telemetry disclosure before any executable invocation. The reviewed 0.20.0 upstream telemetry includes pseudonymous identifiers and bounded product, platform, client, tool/outcome, duration/output, aggregate usage, permission, and lifecycle categories; its tagged event builders exclude prompts, tool arguments/results, typed text, screenshots, accessibility trees, app/window names, URLs, paths, raw values, and raw errors. See [Computer Use](/docs/computer-use/) for the full boundary. ## Credentials Enter credentials only on the relevant Settings page or in the provider sign-in flow. Row-Bot stores secrets in the operating system key store when available and keeps local metadata for status and diagnostics. An explicitly configured server deployment can use a read-only external master-key file to encrypt owner-entered secrets in its persistent data directory; see [Docker And VPS Operations](/docs/operations/docker#read-only-secret-files). ## Remote Access Remote Access is single-owner and multi-device. A Tailscale, LAN, SSH, Docker, or reverse-proxy route provides reachability only; each non-local browser still needs a one-time invitation and a revocable Row-Bot session. Every authenticated interactive browser has full owner authority. Phone/tablet and computer choices select compact or desktop presentation only; device metadata and screen size are not permission boundaries. Row-Bot does not provide guest or multi-user sharing. Invitation links expire after 10 minutes and work once, but should still be treated like passwords until they expire. Row-Bot stores only hashed invitation and session secrets. Session credentials use HttpOnly cookies and are not placed in URLs. Prefer Tailscale or HTTPS for remote access. Plain LAN HTTP is unencrypted and does not support a remote browser microphone. For an operator-managed proxy, allow only the canonical browser-facing origin and trust only the exact address that connects from the proxy to Row-Bot. Never resolve a forwarding-header error by trusting every private address or a whole container network. See [Remote Access And Server Mode](/docs/operations/remote-access) for layouts, lifetimes, revocation, Tailscale ownership, server mode, Docker, and recovery. ## Approvals Use approvals to review file writes, command execution, browser actions, account actions, channel sends, MCP calls, plugin tools, code folder changes, and other sensitive actions. Deny anything that does not match your request. Approvals wait until you answer; they no longer expire on their own, unless a workflow step sets its own time limit. Every waiting approval shows in its conversation, in the sidebar's attention indicator and on Overview, with **Approve** and **Deny** in place, so background work can't slip past you. Each tool result records whether approval was needed, given or denied, and a denied action never runs. Approval mode is policy, not a model choice: blocked actions stay blocked, ask-mode actions wait for you, and auto-approved actions follow the configured policy. Some prompts include a short model-written reason for readability, but the underlying approval gate and action details are system-controlled. ## Sharing Logs Or Screenshots Before sharing logs, screenshots, documents, thread exports, or review packages, check for names, file paths, account names, message contents, tokens, private documents, or misleading real data. ## Safer Defaults - Start with local models when privacy matters most. - Keep channels, MCP, and plugins disabled until needed. - Keep Computer Use disabled until a local interactive task needs native application control. - Keep Remote Access off until another trusted device needs it; use the narrowest route and the appropriate layout. - Use narrow workspaces for file tools. - Review approvals before external or destructive actions. - Keep a final human review step before publishing screenshots or docs built from a personal app state.

## Troubleshooting

Route: /docs/troubleshooting/

Resolve setup, model, chat, approval, workflow, design, code folder, Knowledge, Monitor, Settings, channel, MCP, plugin, skill, voice, and Buddy issues.

# Troubleshooting Start with what Row-Bot already tells you. The sidebar's attention indicator appears when something needs you, Overview's **Needs you** lists it, and [Monitor](/docs/monitor/) shows every health check with a one-click fix. Most issues fall into one of four categories: setup is incomplete, a provider or tool is off, the wrong model is selected, or Row-Bot is waiting for your approval. ## Setup Problems - If Row-Bot keeps opening setup, finish **How should Row-Bot think?** or choose **Set up later**. Open Setup Center from Overview or the command palette to continue optional setup. - Check Settings Providers and Settings Models before troubleshooting a conversation. - Check Settings System if local files, browser automation, or command execution are involved, and Settings Devices & remote access for other devices or the public link. - For native application control, open the Computer Use setup in Settings System and resolve its driver or operating-system permission status before retrying. ## Model Problems - Refresh the provider in Settings Providers, then refresh the catalog in Settings Models. - A model marked **Unavailable** in a picker says why; reconnecting its provider usually fixes it. - Choose a tool-capable model for workflows, designs, code folders, and multi-step tool use. - For local and custom endpoints, use enough context for Row-Bot's instructions and tool schemas; lower the **Reading limit** in Settings Models on smaller computers. - If reasoning choices are missing, rejected, or reset, use the active model's conditional control or `/reasoning`, then follow [Reasoning Controls](/docs/chat/reasoning-controls#troubleshooting). ## Chat, Tool, And Approval Problems - Check the model picker, the composer's **Mode** (Agent or Chat only), the conversation's agent profile, its **Approvals** setting, and the enabled tools. - If an action is waiting, approve or deny it in the conversation, from the sidebar's attention indicator, or from Overview. Approvals wait until you answer. - If a step shows as skipped, it was denied and did not run. - Ask Row-Bot to explain the last tool result if the transcript is unclear. - If Computer Use is paused, choose Resume only after reviewing the target; if its lease is busy, stop or finish the active native task first. ## Workflow Problems - Confirm the workflow is enabled, and that an earlier run isn't still waiting for approval: while one waits, the next scheduled run is skipped. - Run it once with **Run** before trusting a schedule. - Check delivery defaults and the channel's status in Monitor. ## Design And Code Folder Problems - Designs: give a clearer goal, audience and length, and attach references. Exports use the last saved version. - Code folders: check that the folder is listed under **Working on** in the conversation's details card. Choosing or cloning a folder needs the Row-Bot desktop app. - Review commands and changes before approving or importing them. ## Knowledge And Monitor Problems - If Knowledge is empty, turn on memory in Settings Memory or add documents in Settings Documents. - If a memory is wrong, edit, merge or archive it in Home Knowledge. - Use Monitor's **Needs attention**, **History** and **Logs** for background issues, and **Copy diagnosis report** when asking for help. ## Integrations - Channels need credentials, sometimes a public link, a running state, and user pairing. A stopped channel shows **Restart** in Monitor. - Accounts whose sign-in expired show **Reconnect** in Monitor and Overview. - MCP needs **Use MCP servers** on, a trusted server, and a successful **Test connection**. - Plugins need installation, enablement, configuration, and sometimes a restart. - Skills need to be installed, available in chats, and relevant to the prompt. ## Remote Access And Server Mode - If the browser shows `untrusted_forwarding_headers`, a proxy supplied forwarding metadata from an address Row-Bot does not trust. Restart after enabling a Row-Bot-owned Tailscale route. For another proxy, configure only its exact connecting address or CIDR; do not trust a broad network. - If the browser shows `unexpected_host`, add the exact canonical browser-facing host and verify the proxy preserves it. - If a code cannot be used, create it for the exact scheme, host, and port currently used by that browser. Expired, used and stale-route codes must be recreated. - If Tailscale reports a conflict, Funnel, or an unowned route, inspect that configuration manually. Row-Bot deliberately refuses to reset or overwrite it. - If a remote microphone is unavailable, use `localhost` or HTTPS, grant browser permission, and install the local voice model explicitly. - Run `row-bot access doctor` for a secret-free server and route safety report. The full route, authentication, Docker, proxy, and recovery guidance is in [Remote Access And Server Mode](/docs/operations/remote-access). ## Voice And Buddy - Check microphone and output device selection and permissions. - Use Dictate before Talk if you need review. - Check provider readiness for realtime voice. - If docked Buddy is missing, enable **Show Buddy** in Settings Buddy. If a torn-off overlay was hidden, use the tray's **Show Buddy** action. - Drag Buddy itself out of the sidebar in the native Windows or macOS app. Browser/server and phone layouts cannot create the native overlay. - Use **Open full thread** for microphone controls, complex approvals, attachments, tool traces, and complete conversation history. - Restart Row-Bot to return Buddy placement to the sidebar dock; a saved hidden preference must still be turned back on.

## Reference

Route: /docs/reference/

Look up Row-Bot tools, providers, settings, channels, skills, MCP, plugins, data storage, and approvals.

# Reference Use the reference pages when you need a table or lookup after reading the guided docs. These pages are more compact and more technical than the walkthroughs. - [Tools](/docs/reference/generated/tools) - [Providers](/docs/reference/generated/providers) - [Settings](/docs/reference/generated/settings) - [Settings Controls](/docs/reference/generated/settings-controls) - [Home Tabs](/docs/reference/generated/home-tabs) - [Channels](/docs/reference/generated/channels) - [Skills](/docs/reference/generated/skills) - [MCP](/docs/reference/generated/mcp) - [Plugins](/docs/reference/generated/plugins) - [Data Storage](/docs/reference/generated/data-storage) - [Safety And Approvals](/docs/reference/generated/safety-approvals) - [Environment And Config](/docs/reference/generated/environment-and-config) - [CLI](/docs/reference/generated/cli) - [Screenshots](/docs/reference/generated/screenshots)

## Tools

Route: /docs/reference/generated/tools

Generated reference for Row-Bot tools and tool guides.

# Tools

## Providers

Route: /docs/reference/generated/providers

Generated reference for model providers and provider risk labels.

# Providers

## Settings

Route: /docs/reference/generated/settings

Generated reference for the React settings pages.

# Settings

## Settings Controls

Route: /docs/reference/generated/settings-controls

Generated reference for every searchable setting, with deep links, search keywords, dependency notes, and security notes.

# Settings Controls

## Home Tabs

Route: /docs/reference/generated/home-tabs

Generated reference for Home tab coverage.

# Home Tabs

## Channels

Route: /docs/reference/generated/channels

Generated reference for messaging channels.

# Channels

## Skills

Route: /docs/reference/generated/skills

Generated reference for bundled skills and tool guides.

# Skills

## MCP

Route: /docs/reference/generated/mcp

Generated reference for recommended MCP servers.

# MCP

## Plugins

Route: /docs/reference/generated/plugins

Generated reference for plugin manifest behavior.

# Plugins

## Data Storage

Route: /docs/reference/generated/data-storage

Generated reference for local data paths and storage.

# Data Storage

## Safety And Approvals

Route: /docs/reference/generated/safety-approvals

Generated reference for approval policy and safety modes.

# Safety And Approvals

## Environment And Config

Route: /docs/reference/generated/environment-and-config

Generated reference for important environment variables.

# Environment And Config

## CLI

Route: /docs/reference/generated/cli

Generated reference for launcher, access, server, and plugin command-line options.

# CLI

## Screenshots

Route: /docs/reference/generated/screenshots

Generated reference for automated screenshot coverage.

# Screenshots

## Agent Profiles

Route: /docs/app-shell/agent-profiles

Agent profiles now live in the sidebar's Agents dialog; this page points to the current guide.

# Agent Profiles Agent profiles are managed in the sidebar's **Agents** dialog, and a conversation picks one from the composer's **+** menu. Read [Profiles, Goals, And Agents](/docs/profiles-goals-agents/).

## Channels And Voice Guide

Route: /docs/guides/channels-and-voice

Compatibility page pointing to the current Channels guide.

# Channels And Voice Guide This guide has been split into focused pages so each feature has one authoritative walkthrough. - [Channels](/docs/integrations/channels) - [Plugins](/docs/integrations/plugins) - [MCP](/docs/integrations/mcp) - [Voice And Buddy](/docs/voice-and-buddy/) - [Settings](/docs/settings/)

## Designer Studio Guide

Route: /docs/guides/designer-studio

Designer Studio is now a conversation's Design panel; this page points to the current guide.

# Designer Studio Guide There is no separate Designer Studio any more: each design opens in its conversation's Design panel. Read [Designs](/docs/designer/).

## Developer Studio Guide

Route: /docs/guides/developer-studio

Developer Studio is now a conversation's Developer panel; this page points to the current guide.

# Developer Studio Guide There is no separate Developer Studio any more: each code folder opens in its conversation's Developer panel. Read [Code Folders](/docs/developer/).

## Skills, Plugins, And MCP Guide

Route: /docs/guides/skills-plugins-mcp

Compatibility page pointing to the current Skills Hub guide.

# Skills, Plugins, And MCP Guide This guide has been split into focused pages so each feature has one authoritative walkthrough. - [Skills Hub](/docs/skills/) - [Plugins](/docs/integrations/plugins) - [MCP](/docs/integrations/mcp) - [Voice And Buddy](/docs/voice-and-buddy/) - [Settings](/docs/settings/)

## Home: Designer

Route: /docs/home/designer

Home no longer has a Designer tab; designs open in a conversation's Design panel.

# Home: Designer Home no longer has a Designer tab: a design now opens in its conversation's Design panel. Read [Designs](/docs/designer/) for how to start, edit and export one.

## Home: Developer

Route: /docs/home/developer

Home no longer has a Developer tab; code folders open in a conversation's Developer panel.

# Home: Developer Home no longer has a Developer tab: a code folder now opens in its conversation's Developer panel. Read [Code Folders](/docs/developer/) for how to bind, clone and work on one.

## Home

Route: /docs/home/

Home's five tabs: Overview, Workflows, Knowledge, Monitor and Insights.

# Home Home is the start page for everything that isn't a single conversation. Choose **Home** in the sidebar, then one of its five tabs: - [Overview](/docs/home/overview) asks Row-Bot anything, lists what needs you, and shows live status and recent work. - [Workflows](/docs/home/workflows) lists saved workflows and reminders, runs them in one click, and opens the workflow editor. - [Knowledge](/docs/home/knowledge) shows the memory graph and the full memory library, with review, editing and Dream Cycle. - [Monitor](/docs/home/monitor) keeps the latest health checks with one-click fixes, plus activity, history and logs. - [Insights](/docs/home/insights) lists what Row-Bot noticed and the proposals it prepared for your review. Designs and code folders are not Home tabs: each opens in its conversation's panel. See [Designs](/docs/designer/) and [Code Folders](/docs/developer/). ## Full Guides - [Workflows guide](/docs/guides/workflows) - [Knowledge guide](/docs/knowledge/) - [Monitor guide](/docs/monitor/)

## Home: Knowledge

Route: /docs/home/knowledge

Use Home Knowledge to explore the memory graph, search and review memories, and run Dream Cycle.

# Home: Knowledge The Knowledge tab is where you see and manage what Row-Bot remembers: the memory graph, the full memory library, the review queue, recent recalls and changes, and Dream Cycle. ## What You Can Do Here - **Explore the graph.** It opens with up to 2,000 memories; **Show all** draws up to 5,000. - **Search and filter** the whole library by words or status, including the review queue. - **Open a memory** to see its details, connections, sources and when it was recalled. - **Add, edit, merge, archive, restore or delete** memories, one at a time or in bulk. Deleting is always reviewed first. - **Run Dream Cycle now** to tidy memory outside its overnight hours. Settings for remembering, graph health and the wiki vault are in **Settings Memory**. ## Next Step Open the [Knowledge guide](/docs/knowledge/) for the full walkthrough.

## Home: Monitor

Route: /docs/home/monitor

Use Home Monitor to see Row-Bot's latest health checks, fix problems in one click, and read activity, history and logs.

# Home: Monitor The Monitor tab answers: is everything healthy, what needs attention, and what happened recently? Row-Bot keeps the latest result of every check, so the tiles stay current after you switch tabs or restart. ## What You Can Do Here - **Needs attention** lists problems with a one-click fix, such as **Restart** a stopped channel or **Reconnect** an account. - **Health tiles** show each area's latest checks; open one for every check and its fix. - **Run diagnosis** runs every check now; **Copy diagnosis report** copies a summary for support. - **Check connections every hour** controls the checks that contact your providers, accounts and the internet. - **Activity**, **Maintenance**, **History** and **Logs** show what ran and why. ## Next Step Open the [Monitor guide](/docs/monitor/) for the full walkthrough.

## Home: Workflows

Route: /docs/home/workflows

Use Home Workflows to run saved workflows in one click, create or edit them on the page, and set delivery defaults.

# Home: Workflows The Workflows tab lists your saved workflows and reminders: when each runs next, how its last runs went, and what it is doing now. ## What You Can Do Here - **Run** starts a workflow at once; **Stop** appears while it runs. - **New workflow** and **Edit** open the workflow editor on this page. - **Enable** or **Disable** a workflow's schedule. - **Delivery defaults** choose the channels results are also sent to. - The **** menu on a row opens its run history, steps, settings or conversation, or duplicates or deletes it. ## Next Step Open the [Workflows guide](/docs/guides/workflows) for the full walkthrough.

## Reference Tables

Route: /docs/reference/generated/

Compact lookup tables for Row-Bot features and settings.

# Reference Tables

## Settings: Utilities

Route: /docs/settings/utilities

Settings no longer has a Utilities page; the built-in tools are part of Settings Tools.

# Settings: Utilities The built-in tools that used to live under Utilities are now in **Settings Tools Built-in tools**. Read [Settings: Tools](/docs/settings/tools).

## UI Tour

Route: /docs/ui-tour/

A short guided tour of Row-Bot: the sidebar, Home, a conversation, panels, the command palette and Settings.

# UI Tour Use this short tour if you want the quickest mental map before reading detailed pages. 1. Start with the [Row-Bot Interface](/docs/app-shell/navigation): the sidebar, Home, a conversation and its details card, panels, the terminal, and the K command palette. 2. Read [Home: Overview](/docs/home/overview), the first page Row-Bot opens: the Ask box, what needs you, live status and recent work. 3. Read [Chat](/docs/chat/) for conversations, the composer, models, attachments, tool traces and approvals. 4. See how a conversation works on a design or code folder in [Designs](/docs/designer/) and [Code Folders](/docs/developer/). 5. Open [Settings](/docs/settings/) when you are ready to connect providers, choose models, and set up memory, tools, accounts, channels and devices. 6. Keep [Privacy And Safety](/docs/privacy-safety/) nearby when enabling external services or action tools.

More agent context in siddsachar/Thoth

3 other files this repository gives its agents.