feat: reachy_voice on Reachy Mini SDK 1.8.0 becomes the app (replaces reachy_claw)#8
Merged
Merged
Conversation
Bring the reachy_voice app from the separate reachy-voice repo into this repo as the packaged app, replacing the old reachy_claw package. - src/reachy_voice: full working tree (incl. untracked static/app.js, profiles/, all .py); main.py gains a main() console entry point (byte-equivalent to the existing __main__ block). - tests/voice: migrated reachy-voice tests (imports unchanged — already absolute `from reachy_voice...`). - legacy/reachy_claw: old package moved here for later feature ports; not packaged, not imported by the new app. - tests/conftest.py: defer reachy_claw imports into fixtures so the moved package no longer breaks collection of the tests/voice tree. - deploy/jetson/voice: add the reachy-voice APP image (Dockerfile/compose/ entrypoint, COPY src/reachy_voice, build context = repo root, FROM reachy-claw:slv-v7, hf-hub + entrypoint preserved). The pre-existing SLV speech-backend compose is preserved as docker-compose.slv-backend.yml. - pyproject.toml: name reachy-voice, deps reconciled to reachy-voice's (reachy-mini>=1.8.0 + voice deps), hatch wheel packages=[src/reachy_voice], scripts + reachy_mini_apps entry point -> reachy_voice.main:ReachyVoiceApp, vendored openvoicestream-agent source kept pointing at the in-repo wheel. Tests: tests/voice -> 66 passed, 7 skipped (intrinsic SDK/gi importorskip). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…res)
Port the self-contained dashboard endpoints the shared frontend
(static/app.js) calls but the reachy_voice backend lacked. All are
HTTP reverse-proxy / Docker Engine socket calls — no reachy-mini SDK,
no conversation core, no deferred feature.
New module reachy_voice/tier_a.py holds the logic (decoupled from
FastAPI/websocket plumbing so it's unit-testable with a mocked httpx
transport):
- GET /api/ollama/models?url= -> proxy {url}/api/tags, return model
names, fall back to a default list on error.
- WS restart_services -> restart docker containers in order via
/var/run/docker.sock, poll health before advancing, broadcast
restart_status. Default order uses reachy-voice (NOT reachy-claw):
vision-trt -> reachy-daemon -> reachy-voice; remote vision-trt is
skipped.
- WS get_capture_info -> capture_info {path,count}; WS clear_captures
-> capture_reset; GET /api/captures/list + /api/captures/image/{f}
reverse-proxy vision-trt :8630. Vision host derived from
config.vision_mjpeg.
main.py wires these into the settings app's HTTP routes and /ws loop;
all existing handlers (get/set_volume, /stream, /debug/*, /language)
untouched.
Tests under tests/voice (29 new, all green; mock httpx + docker socket
+ vision-trt, no hardware).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…il handling Self-review fixes before the SDK 1.8.0 baseline: - compose: mount /var/run/docker.sock (dashboard "restart services" was a silent no-op without it) + align image tag to 0.2.0 - pyproject + ci.yml: scope tests to tests/voice (44 legacy reachy_claw tests under tests/ root import the moved-away package and broke CI collection) - tier_a.py: broadcast restart "error" when all restarts fail (not "done"); asyncio.get_running_loop() instead of deprecated get_event_loop() - .gitignore: hf-hub/ (never commit the 107MB pre-staged HF cache) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ail-fast off The voice test suite imports the audio stack (sounddevice), which dlopens PortAudio at import time; the runner lacked libportaudio2, breaking collection of tests/voice/test_conversation.py. Also drop the 3.10 matrix entry (uv ignores it under requires-python>=3.11) and disable fail-fast for clean per-version signal. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The two GStreamer-backed import checks guarded only on `gi`, but CI has `gi`
without the `Gst` typelib, so they raised ValueError instead of skipping. Guard
on `gi.require_version("Gst","1.0")` so they skip on dev/CI and run on the robot.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Architecture (5-container runtime + data flow), repo layout, build/test/deploy, SDK 1.8.0 conventions (no_media, set_target_*, play_move, daemon :38001), and hard constraints. Deliberately excludes the upstream JS/Web/scaffolding "golden path" — this is an on-robot Python app. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ction) master's branch protection requires the "test (3.10)" status check; dropping it from the matrix left PR checks incomplete (mergeStateStatus BLOCKED). uv runs the 3.10 job on 3.12 (requires-python>=3.11) so it passes under that job name. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Establishes reachy_voice on reachy-mini SDK 1.8.0 as the new app baseline, replacing the old
reachy_clawapp.src/reachy_claw/→legacy/reachy_claw/(kept for reference; not packaged/imported). Its high-value features (diary, skills, Home Assistant, voice-cloning, conversation modes, multi-LLM backends) will be ported onto the new base in follow-ups.reachy_voiceapp is nowsrc/reachy_voice/— the packaged app + entry-point + deploy.pyproject:reachy-mini>=1.8.0,packages=["src/reachy_voice"],reachy_mini_apps → reachy_voice.main:ReachyVoiceApp; vendored openvoicestream-agent wheel kept.deploy/jetson/voice/(Dockerfile/compose/entrypoint), buildsFROM reachy-claw:slv-v7.Why
The old
reachy_clawapp is pinned to an outdated SDK and monkey-patches internal SDK GStreamer/media classes — fragile across SDK upgrades.reachy_voiceis a clean re-implementation on SDK 1.8.0 + ovs_agent that sidesteps that (no_media + vision-trt + stable motion APIs), and adds SDK-calibrated gaze + an engagement-gated attention tracker.Also: Tier-A dashboard APIs
The shared dashboard frontend is unchanged, but the reachy_voice backend was missing several endpoints. This wires up the pure-port ones (no SDK / no conversation-core dependency), in
src/reachy_voice/tier_a.py:GET /api/ollama/models,restart_services(Docker socket), capture proxies (get_capture_info/clear_captures//api/captures/*).Testing
tests/voice/: 95 passed, 7 skipped (skips are SDK/GStreamer conditional, off-Jetson).ruff check .clean. CI scoped totests/voice(legacy reachy_claw tests kept undertests/root as reference, not run).gi): builtreachy-voice:0.2.0from this branch, swapped it in, verified live — motion compositor (100 moves), vision, audio, edge-LLM, daemon/SDK control, and the new/api/ollama/modelsall working.reachy-voice:v0.1kept for rollback.Self-review fixes included
/var/run/docker.sock(restart-services was a silent no-op) + image tag0.2.0errorwhen all restarts fail (notdone);get_running_loop()tests/voice(legacy tests broke collection).gitignore:hf-hub/Not in this PR (follow-up)
🤖 Generated with Claude Code