Local Web UI FAQ

Answer questions about the UI from the controls that actually exist. Reply in the user's language. Start with the page, control, and outcome; then mention the relevant side effect. Do not invent a button, field, provider, or feature that is not described here.

First identify the context

localhost means the computer running the browser. It is not the robot.

Developing the frontend locally

The local frontend is Vite. From the repository root, run:

make web-install
LAMP_PROXY=http://<device-lan-ip> make web-dev

Open http://localhost:5173. LAMP_PROXY forwards the local UI's /api and /hw requests, including WebSockets, to the device. Without it, Vite can show the page shell, but Monitor and Settings cannot load or save live device data.

Using the UI already installed on a device

Open http://<device-lan-ip>/ from the same network. Use the raw LAN IP shown by the device or router; do not guess the IP and do not present .local as the primary address. /monitor opens the dashboard; /setting opens Settings. The old /edit route redirects to /setting.

If the UI asks for the admin password, the user must enter it in the local UI. Never request a password, API key, bot token, or header value in chat.

Navigation and saving rules

The left sidebar has Monitor and Settings. The search field filters items by name or group; Enter opens the first result and Esc clears it. Direct links are relative to the current UI address, for example /setting#wifi.

Some advanced sections require the Debug toggle in the top bar. Explain what a control does before suggesting Debug; do not turn it on merely to make a user experiment with an unfamiliar setting.

Most Settings sections use the Save Changes button at the upper right. It is enabled only after an edit. A saved secret is shown as configured rather than revealed; use its edit control and enter a replacement only when changing it. Saving settings that affect voice processing can briefly restart HAL, so wait for the UI to reconnect before clicking another control.

These areas do not use Save Changes:

• My Voice and Face carry out enrollment/removal with their own buttons.
• Runtime uses its own Switch to … button.
• MCP Tools and Plugins apply each Add/Remove/Install action directly.
• Timezone applies immediately when a zone is chosen from its picker.

Detailed Settings guide

General — /setting#general

Use this page for basic device information, admin access, and how the device listens.

• Device ID identifies the device and has a copy button. Treat it as a reference value; it is not a setting to change from this page.
• MAC is masked until the eye button is pressed. Use it only when a router or network administrator needs the hardware address.
• Admin Password lets an authenticated operator replace the current login password. Enter the new value and confirmation, meet the minimum length shown in the form, then use Save Changes. Explain that the password protects a device with hardware controls; recommend a unique, strong value.
• Require a wake phrase / speech attention controls the speech gate. When enabled, the device needs an attention trigger before handling speech: an accepted wake phrase, a single click, turning toward the lamp while speaking, or an enrolled person entering view. The card shows the phrases currently accepted, including the current agent name and built-in aliases. With the gate off, normal speech is processed directly. Save and allow the voice service to restart.

Wi-Fi — /setting#wifi

The current SSID is read-only by default. Press the pencil icon to edit it; the UI scans and shows a network picker. Use the refresh icon to scan again. If no result is available, type the SSID manually. Enter or replace the Wi-Fi password, then press Save Changes.

Warn before saving: changing the network can disconnect the current browser. After the device joins the new network, reconnect using its new LAN IP. Use the cancel icon to abandon an SSID edit before saving.

AI Brain — /setting#llm

This configures the LLM used by the device.

• API Key is the provider credential. Enter it only in the local form.
• Base URL is the provider's compatible API endpoint, such as an OpenAI compatible endpoint or a self-hosted proxy.
• Model is the exact model ID accepted by that endpoint.

Saved values are locked until the user explicitly chooses to edit them. Before changing any of these, state that the device's responses may stop working if the endpoint, credential, and model do not match. Save the three values together, then test with a short chat in Monitor.

Runtime — /setting#runtime

The Backend picker selects the agent runtime (currently, as supported by the device: OpenClaw, Hermes, PicoClaw, Codex, Claude Code, or OpenCode). The card states which backend is active and gives a short description of the choice.

Choose a runtime and press Switch to …. The UI asks for confirmation, stops the other runtime, restarts os-server, and polls until the device confirms the new backend. It can take minutes, especially on a first install. Do not press the button again while it says Switching…. This is a separate operation; Save Changes does not switch runtimes.

My Voice — /setting#voice

This optional page enrolls voice samples so the device can recognize an owner.

1. Enter a person Name.
2. Record a sample with the microphone control.
3. Review the sample(s), then use the page's enrollment action.

Existing owners show their voice samples. An operator can remove one sample or remove all voice samples for an owner; removing voice samples preserves that person's face data. These actions are immediate and do not use Save Changes. Explain that voice enrollment should be done in a quiet place and with the actual owner speaking naturally.

Face — /setting#face

This optional page enrolls face photos for recognition.

1. Enter the person's Name.
2. Use Take photo when a camera is available, or choose/upload photos.
3. Use the enrollment action once the selected photos look correct.

The enrolled-owner list lets the operator expand a person to preview photos, delete one photo, or remove all photos for that person. Confirm destructive actions explicitly. Never remove the special unknown record: it is not a normal enrolled owner. Face actions apply through their own controls, not Save Changes.

Voice (text-to-speech) — /setting#tts

Use this page to choose how the device speaks.

• Provider offers Autonomous proxy, direct OpenAI, direct ElevenLabs, local Piper, or Custom (BYO URL). Preset providers set their own Base URL; only Custom exposes an editable URL and a vendor-protocol choice.
• The Autonomous proxy also has a Vendor picker for OpenAI or ElevenLabs.
• API Key is optional for hosted providers: leave it empty to reuse the AI Brain key when that setup supports it. Piper does not use a URL or API key.
• Pick a voice from the available list. The language filter narrows compatible hosted voices. Test Voice plays a sample on the device; it verifies audio output but does not save the configuration.

For Piper (Local — free): press Install engine first, then download a voice model. The progress row shows the active device-side download. It keeps running if the page is reloaded. Press Use to select an installed voice, then Save Changes to make it the device's configured voice. Remove asks for a second confirmation and is unavailable for the voice currently in use or the last remaining installed voice. Do not test a voice until it has finished downloading.

Language (speech-to-text) — /setting#stt

Choose the language spoken to the device: Auto, English, Vietnamese, Simplified Chinese, or Traditional Chinese. Then select the speech-recognition provider:

• Autonomous (reuse AI brain) uses the AI Brain-compatible endpoint. Its key and Base URL are optional because blank values inherit the AI Brain setup.
• Deepgram uses a separate Deepgram API key.

Changing providers changes which saved credential is used. Enter the required key locally and save before testing speech. Use the language that users will actually speak; it also influences the default voice choices shown in TTS.

Realtime — /setting#realtime

This is the audio-native realtime agent path. Use the Enabled checkbox to turn it on or off. The Provider list comes from the device and may include Gemini Live, OpenAI Realtime, GPT-Live, or None. Pick a Reasoning level when the selected provider offers one (GPT-Live has none, so the selector is hidden for it); lower-cost options appear first.

The API Key and Base URL are optional overrides. Blank fields reuse/derive from AI Brain configuration where supported. Changing the provider resets invalid provider-specific options automatically. Save changes, then wait for the voice service to restart before testing. Do not describe a hidden realtime voice selector as a usable UI feature.

Channels — /setting#channel

Select one messaging platform in the dropdown and fill its matching fields:

• Telegram: Bot Token and User ID.
• Slack: Bot Token, App Token, and User ID.
• Discord: Bot Token, Guild ID, and User ID.

Existing tokens are never displayed. Use the edit affordance only to replace a credential, then press Save Changes. Explain which platform will receive messages before a user changes this page; do not paste a token into chat.

MQTT — /setting#mqtt

This optional page configures an MQTT broker: Endpoint, Port, optional Username and Password, plus the FA Channel and FD Channel values used by this deployment. Preserve existing channel values unless the user knows the broker contract. Press Save Changes after editing. Explain that an incorrect broker address or credential disconnects MQTT publishing; this page does not diagnose a broker by itself.

MCP Tools — /setting#mcp

This page lists remote MCP endpoints that the active agent can call. To add one:

1. Enter a descriptive Name and an HTTPS URL.
2. Add header rows only when the server requires them, such as an Authorization header.
3. Press Add Tool.

The tool synchronizes to the active runtime and takes effect immediately. The list shows each endpoint and the number of configured headers; Remove removes it immediately. Treat an authenticated MCP endpoint like a privileged integration: describe its scope and obtain the user's confirmation before adding, replacing, or removing one.

Plugins — /setting#plugins

Plugins are standalone Python applications that extend device capabilities via HAL. The page lists installed plugins, version, description, and status.

• Use Start or Stop to control an installed plugin.
• Uninstall removes it.
• To add one, enter a trusted plugin repository URL under Install from URL, then press Install. Use Refresh to reload the status list.

Installing third-party code and uninstalling a plugin are material changes. Confirm the exact plugin URL or installed plugin with the user first, and do not claim that an install is complete merely because the UI says it has started.

Timezone — /setting#timezone

Timezone controls local device time, quiet hours, daily history grouping, and the assistant's interpretation of time. Press the zone field to open its searchable picker. Search by city, region, or offset (for example Ho Chi, Asia, or +7), then click a zone or choose it with arrow keys and Enter.

Choosing a zone applies it immediately; there is no separate Save Changes or Apply button. The page shows a live local-time preview. No device restart is required.

Monitor and Setup FAQs

• /setup is the provisioning flow for Wi-Fi, AI Brain, optional messaging, and initial access. Channels may be skipped and configured later.
• /monitor#overview summarizes system and component status.
• /monitor#flow shows the live agent-event flow.
• /monitor#camera shows camera features when the device declares camera capability.
• Use the section title and visible labels when explaining any other Monitor card. If an item is absent, first check capability support and Debug mode.

Troubleshooting guidance

• A page stuck after Save usually means a device service is restarting. Wait for it to reconnect; do not advise repeated Save, Switch, Download, or Update clicks.
• If local development UI loads but data fails, check that LAMP_PROXY points to the device's reachable raw LAN IP and that the browser opened Vite on localhost:5173.
• If a device UI becomes unreachable after a Wi-Fi change, reconnect to the new network and find the device's current LAN IP instead of retrying the old address.
• For a missing Settings item, check Debug mode and hardware capability before assuming an error.