Add Reachy Mini OpenShell demo project

[johnnygreco]

Summary

Adds a Reachy Mini conversation demo for OpenShell under projects/reachy-mini-openshell. The project provides a simulator-first local workflow, Gradio browser UI, Reachy movement tools, selectable model backends, and onboarding docs for running the app with minimal setup.

What This Adds

• A reachy_mini_conversation_app package with reachy-mini-conversation-app and reachy-mini-backend-check commands.
• A one-step local launcher, ./scripts/start-local.sh, that syncs dependencies, validates config, starts the Reachy Mini simulator daemon, and launches the browser app.
• A single .env.example template copied to .env; credentials and model routing stay out of the browser UI.
• Backend selection through BACKEND_PROVIDER:
  • openai_realtime for OpenAI Realtime speech, tool use, and speech output.
  • hf_realtime for the Pollen/Hugging Face realtime path.
  • local_stt for OpenAI-compatible STT -> Chat Completions + Reachy tools -> TTS.
• Microphone and text input modes in the UI.
• Reachy profile/tool support for motion, emotion playback, background tool status, and cancellation.
• Local-STT microphone phrase detection, audio normalization, endpoint clients, and app-flow checks.
• A fake OpenAI-compatible backend plus smoke script for local-STT validation without external STT/TTS services.
• README and docs-site tutorial coverage for simulator setup, daemon status verification, app usage, backend selection, provider configuration, and troubleshooting.
• Repo hygiene updates for Python/Rust/TypeScript-style generated files and secret/local env handling.

Local Demo Flow

cd projects/reachy-mini-openshell
cp .env.example .env
export OPENAI_API_KEY=sk-...
./scripts/start-local.sh

Open the printed Gradio URL, then choose Microphone for voice input or Text for typed prompts.

Validation

• Panel review completed with no blocking runtime, docs, or hygiene findings.
• uv run ruff check .
• uv run ruff format --check .
• uv run ty check
• uv run pytest -q (165 passed)
• env OPENAI_API_KEY=dummy uv run reachy-mini-backend-check --env-file .env.example
• git diff --check

[johnnygreco]

Panel Review Update

Launched a three-pass panel review after pushing the final refactor:

• Backend/runtime pass found a P1 Realtime tool-call race: the app requested a follow-up response before sending function_call_output. Fixed in 78193da by waiting until the tool result is sent before queueing the Realtime follow-up, with a regression test covering the no-pre-result-response behavior.
• UX/docs and repo-hygiene passes found a P1 launcher readiness issue: scripts/start-local.sh treated any HTTP-ish daemon status as healthy. Fixed in 78193da by requiring /api/daemon/status JSON with type == "daemon_status" and state == "running".
• Docs pass also noted the local-STT app-flow/--real-reachy validation wording could confuse Realtime users. Fixed in 78193da by scoping that section explicitly to BACKEND_PROVIDER=local_stt.

Remaining note: pyproject.toml intentionally keeps uv lock resolution scoped to Darwin for the local Reachy/MuJoCo demo workflow. This is documented, but Linux CI would need its own lock/platform handling before treating this project as a normal cross-platform CI target.

Validation after fixes:

• uv run ruff format .
• uv run ruff check .
• uv run ty check
• uv lock --check
• uv run pytest -q (166 passed)
• git diff --check
• uv run python -m reachy_mini_conversation_app --help
• reachy-mini-backend-check dry-run validation for OpenAI Realtime, HF Realtime, and local-STT example env files
• uv run reachy-mini-backend-check --env-file .env.openai-realtime.example --live
• scripts/smoke-local-stt.sh