homeai

aodhan/homeai

Fork 0

Commit Graph

Author	SHA1	Message	Date
Aodhan Collins	af6b7bd945	feat: upgrade voice pipeline — MLX Whisper STT (20x faster), Qwen3.5 MoE LLM, fix HA tool calling - Replace faster-whisper with wyoming-mlx-whisper (whisper-large-v3-turbo, MLX Metal GPU) STT latency: 8.4s → 400ms for short voice commands - Add Qwen3.5-35B-A3B (MoE, 3B active params, Q8_0) to Ollama — 26.7 tok/s vs 5.4 tok/s (70B) - Add model preload launchd service to pin voice model in VRAM permanently - Fix HA tool calling: set commands.native=true, symlink ha-ctl to PATH - Add pipeline benchmark script (STT/LLM/TTS latency profiling) - Add service restart buttons and STT endpoint to dashboard - Bind Vite dev server to 0.0.0.0 for LAN access Total estimated pipeline latency: ~27s → ~4s Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-13 18:03:12 +00:00
Aodhan Collins	c31724c92b	Complete P2 (LLM) and P3 (voice pipeline) implementation P2 — homeai-llm: - Fix ollama launchd plist path for Apple Silicon (/opt/homebrew/bin/ollama) - Add Modelfiles for local GGUF models: llama3.3:70b, qwen3:32b, codestral:22b (registered via `ollama create` — no re-download needed) P3 — homeai-voice: - Wyoming STT: wyoming-faster-whisper, large-v3 model, port 10300 - Wyoming TTS: custom Kokoro ONNX server (wyoming_kokoro_server.py), port 10301 Voice af_heart; models at ~/models/kokoro/ - Wake word: openWakeWord daemon (hey_jarvis), notifies OpenClaw at /wake - launchd plists for all three services + load-all-launchd.sh helper - Smoke test: wyoming/test-pipeline.sh — 3/3 passing HA Wyoming integration pending manual UI config (STT 10.0.0.200:10300, TTS 10.0.0.200:10301). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-03-04 23:28:22 +00:00

Author

SHA1

Message

Date

Aodhan Collins

af6b7bd945

feat: upgrade voice pipeline — MLX Whisper STT (20x faster), Qwen3.5 MoE LLM, fix HA tool calling

- Replace faster-whisper with wyoming-mlx-whisper (whisper-large-v3-turbo, MLX Metal GPU)
  STT latency: 8.4s → 400ms for short voice commands
- Add Qwen3.5-35B-A3B (MoE, 3B active params, Q8_0) to Ollama — 26.7 tok/s vs 5.4 tok/s (70B)
- Add model preload launchd service to pin voice model in VRAM permanently
- Fix HA tool calling: set commands.native=true, symlink ha-ctl to PATH
- Add pipeline benchmark script (STT/LLM/TTS latency profiling)
- Add service restart buttons and STT endpoint to dashboard
- Bind Vite dev server to 0.0.0.0 for LAN access

Total estimated pipeline latency: ~27s → ~4s

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

2026-03-13 18:03:12 +00:00

Aodhan Collins

c31724c92b

Complete P2 (LLM) and P3 (voice pipeline) implementation

P2 — homeai-llm:
- Fix ollama launchd plist path for Apple Silicon (/opt/homebrew/bin/ollama)
- Add Modelfiles for local GGUF models: llama3.3:70b, qwen3:32b, codestral:22b
  (registered via `ollama create` — no re-download needed)

P3 — homeai-voice:
- Wyoming STT: wyoming-faster-whisper, large-v3 model, port 10300
- Wyoming TTS: custom Kokoro ONNX server (wyoming_kokoro_server.py), port 10301
  Voice af_heart; models at ~/models/kokoro/
- Wake word: openWakeWord daemon (hey_jarvis), notifies OpenClaw at /wake
- launchd plists for all three services + load-all-launchd.sh helper
- Smoke test: wyoming/test-pipeline.sh — 3/3 passing

HA Wyoming integration pending manual UI config (STT 10.0.0.200:10300,
TTS 10.0.0.200:10301).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

2026-03-04 23:28:22 +00:00

2 Commits