aidemo
llms.txt32 starsChanged 7 days agoMITView on GitHub
aidemo is built for coding agents to author product demos as code — the hero video on a landing page, a walkthrough in a GitHub README, a what-shipped video attached to release notes for customers. It turns an agent-authored storyboard.json into a polished, narrated, captioned product-demo video (MP4); because the replay is deterministic, that video re-renders itself in CI when the product changes — no re-recording, no API key, about $0 a render. It is an open-source (MIT) alternative to…
- Installs packages
In this repository8 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- aidemo-blogSkill
- blog-waveSkill
- landing-pageSkill
- record-demoSkill
- releaseSkill
- verifySkill
The file
We split this llms.txt into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 66
Use cases: autoplaying GIFs for GitHub READMEs (aidemo render demos/onboarding --headless && aidemo gif demos/onboarding writes …
≈ “Common” is an estimate: at least 1 in 5 of the 3,727 llms.txt files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
# aidemo > aidemo is built for coding agents to author product demos as code — the > hero video on a landing page, a walkthrough in a GitHub README, a > what-shipped video attached to release notes for customers. It turns an > agent-authored `storyboard.json` into a polished, narrated, captioned > product-demo video (MP4); because the replay is deterministic, that video > re-renders itself in CI when the product changes — no re-recording, no API > key, about $0 a render. It is an open-source (MIT) alternative to Screen > Studio, Clueso, or Demosmith for when you'd rather your coding agent make > the demo: the agent writes the storyboard (script + per-scene voice/music > plan + declarative browser action-spec), then a deterministic player drives a > real Chrome — no LLM in the loop during capture — records it, generates a > realistic voiceover (OpenAI TTS), aligns word-level captions (Whisper), trims > dead time, auto-zooms on interactions, and exports a shareable MP4 with > SRT/VTT caption files timed to it (optionally also a soft subtitle track > viewers can toggle). Runs > locally — fully offline with the built-in in-process voice > (AIDEMO_TTS_PROVIDER=local, no API key, no server) or any OpenAI-compatible > local server — and works against localhost and auth-walled apps using your > own Chrome profile. Key facts: - Works with any MCP-capable agent — Claude Code, Codex CLI, Gemini CLI — via the built-in `aidemo mcp` stdio server: authoring tools (engine-served authoring guide, storyboard JSON Schema, structured validation) plus pipeline tools that run as polled jobs (probe/record/render return a jobId; job_status reports per-scene progress, log tails, failure artifacts; job_cancel aborts mid-take and keeps the partial footage). `aidemo repo-init` registers the server (.mcp.json, .gemini/settings.json; one-line codex mcp add for Codex). In Claude Code, the record-demo plugin (`/plugin marketplace add tandryukha/aidemo` then `/plugin install record-demo@aidemo`) bundles the skill and the MCP server in one step. Any human or agent that can write a `storyboard.json` and run `aidemo render` can use it too. - Deterministic replay: recording runs a fixed action-spec at full speed; the agent only authors, it never drives the capture. Same input → same video. - Tweak loop without re-recording: per-scene TTS hash cache means editing one narration line re-voices only that scene; zoom choreography is compose-time, so a bad zoom is a recompose, never a re-record. - Self-updating in CI (no LLM tokens, no API key): the storyboard is committed, and the aidemo GitHub Action (`uses: tandryukha/aidemo@stable`) replays it in CI when the product changes. Deterministic replay plus the in-process voice model mean a re-render needs no key and no tokens, so it costs ~nothing. CI can auto-commit the fresh MP4/GIF, comment it on a pull request, or publish to a stable raw-GitHub embed URL that never changes (`aidemo embed <dir>` prints the snippet). The same storyboard also emits named stills at marked beats (screenshot mode, `still` markers), renders narration in several languages from a single recorded take (`--lang`/`--langs`), and renders per-prospect variants (`--param`/`--variants`); `aidemo probe --golden` diffs the recorded flow against a committed baseline so a broken selector fails CI instead of shipping a stale video. - Private by design: the pipeline's only network calls are TTS narration and caption transcription, sent with your own OpenAI key — or skipped entirely: AIDEMO_TTS_PROVIDER=local runs Kokoro-82M in-process (opt-in `npm install kokoro-js`; ~90 MB model downloaded once, air-gapped after) and captions derive from the script automatically, so the whole render is offline with no API key and no server. Any OpenAI-compatible local server (speaches, Kokoro-FastAPI, LocalAI) via OPENAI_BASE_URL also works, key-free. `aidemo captions --offline` needs no speech-to-text at all (approximate script-based timing). No telemetry or analytics; music is synthesized locally from math, captions/cards are rasterized locally, and the output MP4 never leaves disk. Verified end-to-end both against a bundled mock OpenAI-compatible server and against real local models (speaches: Kokoro TTS + faster-whisper captions, no API key). - Use cases: autoplaying GIFs for GitHub READMEs (`aidemo render demos/onboarding --headless && aidemo gif demos/onboarding` writes `output/final-demo.gif` next to the MP4); muted-autoplay hero videos for landing pages (the hero video on aidemo.top is aidemo output); release / what-shipped demos attached to a release (`gh release upload v1.4.0 demos/whats-new/output/final-demo.mp4`); personalized customer & prospect demos against localhost or auth-walled apps in your own logged-in Chrome, nothing uploaded anywhere; end-of-sprint progress demos — one prompt, the agent authors and renders in the background while you keep working. - Integrations: authoring by any MCP-capable agent (Claude Code, Codex CLI, Gemini CLI) via the aidemo MCP server; narration voice by OpenAI TTS (gpt-4o-mini-tts) behind a swappable VoiceProvider interface, with an opt-in ElevenLabs provider (AIDEMO_TTS_PROVIDER=elevenlabs) and an in-process local provider (AIDEMO_TTS_PROVIDER=local — Kokoro-82M via kokoro-js, no server, no key); captions by OpenAI Whisper (word-level timestamps) — voice and captions both also work against any OpenAI-compatible local server (speaches, Kokoro-FastAPI, LocalAI) via OPENAI_BASE_URL; recording via Playwright + system Chrome; final compose via ffmpeg. All rendering is local. - Requirements: Node 22.12+, Google Chrome, ffmpeg/ffprobe, and an OpenAI API key — or none: AIDEMO_TTS_PROVIDER=local (built-in in-process voice) or an OpenAI-compatible local server via OPENAI_BASE_URL make the render fully offline and key-free. Developed on macOS; Linux expected to work (default capture path). - Made with aidemo (first customers, real-world examples): a 51-second self-narrated demo of the engine driving Wikipedia (the hero video on the landing page and README); aidemo's first customers are MaxFit (maxfit.ee) — a 75-second narrated shopping demo inside its ChatGPT app, an auth-walled widget in nested iframes through to a live checkout — and turg.fitness.ee, a production fitness-gear marketplace whose walkthrough (same record-demo skill, pointed at the live site) is next. ## Docs - [README](https://github.com/tandryukha/aidemo#readme): positioning, design decisions, quickstart, CLI reference, cinematic-polish options - [Authoring guide](https://aidemo.top/AUTHORING.md): the canonical agent-neutral guide — storyboard schema, action vocabulary, demo-director principles, ChatGPT-app recording facts; served by the engine itself via the MCP `get_authoring_guide` tool and `aidemo guide` - [record-demo skill](https://github.com/tandryukha/aidemo/blob/main/.claude/skills/record-demo/SKILL.md): thin Claude Code adapter that fetches and follows the authoring guide - [Releases](https://github.com/tandryukha/aidemo/releases): versioned releases with real rendered MP4s as assets - [Example output (51s, narrated)](https://github.com/tandryukha/aidemo/releases/download/v0.3.0/wikipedia-showcase-demo.mp4): aidemo demoing itself on Wikipedia, authored by Claude - [Example output (75s, narrated)](https://github.com/tandryukha/aidemo/releases/download/v0.3.0/maxfit-chatgpt-demo.mp4): a shopping flow inside a ChatGPT app, authored by Claude - [CONTRIBUTING](https://github.com/tandryukha/aidemo/blob/main/CONTRIBUTING.md) ## Blog - [aidemo blog](https://aidemo.top/blog/): practical, evidence-backed articles on product demo videos — scripting, recording, automation, AI tooling, and demos-as-code. Every article has a raw-markdown mirror at its URL + /index.md. - [Blog llms.txt](https://aidemo.top/blog/llms.txt): per-article markdown index - [Blog full corpus](https://aidemo.top/blog/llms-full.txt): every article in one file ## Install - Claude Code plugin (bundles the record-demo skill + the aidemo MCP server): `/plugin marketplace add tandryukha/aidemo` then `/plugin install record-demo@aidemo` - GitHub Action (render in CI, no key, ~$0): add `uses: tandryukha/aidemo@stable` to a workflow (see docs/CI.md) - Any MCP agent (Codex CLI, Gemini CLI) or CLI use from a product repo: `npx -y github:tandryukha/aidemo#stable repo-init` (registers the MCP server), then `npx -y github:tandryukha/aidemo#stable <cmd>` - Clone + run the engine itself: `git clone https://github.com/tandryukha/aidemo && npm install`, then `node bin/aidemo.mjs render <demo-dir>`
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

