Commit Graph
309 Commits
Author SHA1 Message Date
Mikei386 364ac6b83e Add venv support to FLUX worker image 2026-08-22 17:45:50 +02:00
Mikei386 facc224c1b Install FLUX dependencies in isolated venv 2026-08-22 17:45:27 +02:00
Mikei386 7ac93befc4 Add RTX 5080 FLUX hot-swap worker 2026-08-22 17:40:32 +02:00
Mikei386 8267a85a96 Offload vision projector to RTX 3060 2026-08-22 17:19:56 +02:00
Mikei386 7513a0c912 Guide profiles to verify uncertain facts on web 2026-08-22 12:07:34 +02:00
Mikei386 a9c7a414a0 Enable local web MCP for all profiles 2026-08-22 11:54:50 +02:00
Mikei386 d86727b231 Fix web MCP DNS and crawler runtime 2026-08-22 11:49:47 +02:00
Mikei386 95ed583a66 Provision OpenWebUI workspace model presets 2026-08-22 11:37:30 +02:00
Mikei386 41b9c17f0f Define four-profile production matrix with Medium default 2026-08-22 11:30:38 +02:00
Mikei386 977f8f5c76 Validate Pure quant for 256K Ultra profile 2026-08-22 11:01:13 +02:00
Mikei386 f52cf14069 Add tested 256K Ultra profile 2026-08-22 10:44:14 +02:00
Mikei386 a194a2941d Improve MCP tool selection guidance 2026-08-21 20:34:40 +02:00
Mikei386 6f14f62dd1 Persist Piper settings in restored OpenWebUI 2026-08-21 17:27:26 +02:00
Mikei386 a5cf11582a Add reproducible Piper TTS service 2026-08-21 17:02:32 +02:00
Mikei386 53dfffc289 Reveal animated Open WebUI background 2026-08-21 16:38:34 +02:00
Mikei386 c0cbeb0f15 Add global Midnight Aurora Open WebUI theme 2026-08-21 15:28:39 +02:00
Mikei386 6edf1fe462 Add reviewed OpenWebUI quick actions 2026-08-21 15:23:28 +02:00
Mikei386 16c83ca7d6 Add OpenWebUI stability and privacy guards 2026-08-21 15:12:00 +02:00
Mikei386 422c4fac9e Disable OpenWebUI follow-up generation 2026-08-21 14:59:42 +02:00
Mikei386 9a40758c7f Make OpenWebUI reasoning filter order reproducible 2026-08-21 14:56:39 +02:00
Mikei386 bf90698b99 Document Athena rebuild and clarify tool volume ownership 2026-08-21 14:35:52 +02:00
Mikei386 934a4d8ced Restore OpenWebUI with its matching image 2026-08-21 14:28:44 +02:00
Mikei386 0c21aed7af Automate reference host data restore 2026-08-21 14:11:45 +02:00
Mikei386 a2c2ee7570 Make model permissions usable by inference 2026-08-21 14:07:09 +02:00
Mikei386 bea144c91e Make fresh router startup reliable 2026-08-21 14:03:32 +02:00
Mikei386 868b89be1d Allow router to initialize fresh volume ownership 2026-08-21 14:00:40 +02:00
Mikei386 16be3e606f Start web search only from tool stack 2026-08-21 13:57:01 +02:00
Mikei386 ccf0efd059 Avoid redundant TinySearch browser install 2026-08-21 13:55:07 +02:00
Mikei386 29d65dc223 Use readable Docker build progress 2026-08-21 13:39:06 +02:00
Mikei386 db6e93296d Fix CUDA driver linking in llama.cpp image 2026-08-21 13:38:28 +02:00
Mikei386 d754d35229 Heal legacy NVIDIA keyring permissions 2026-08-21 13:22:46 +02:00
Mikei386 e9c91ea4aa Allow apt to read NVIDIA toolkit keyring 2026-08-21 13:22:12 +02:00
Mikei386 42e2aad3a6 Use embedded MTP tensor for IQ4-MIX profiles 2026-08-21 13:18:45 +02:00
Mikei386 057920b941 Install current NVIDIA driver from official repository 2026-08-21 13:12:37 +02:00
Mikei386 4089bb46e6 Add approval-bound Sonarr episode search 2026-08-21 10:57:33 +02:00
Mikei386 f0d552ef58 Containerize MCP tool services 2026-08-20 23:54:02 +02:00
Mikei386 3ab9628088 Document centralized isolated MCP tool plane 2026-08-20 21:42:40 +02:00
Mikei386 e83c0e2c70 Add reproducible Docker and WireGuard host bootstrap 2026-08-20 21:23:16 +02:00
Mikei386 cb07779f5a Replace vision hotswap with native multimodal profiles 2026-08-20 14:24:56 +02:00
Mikei386 8a45a0d851 Optimize Qwen fast profile for vision and 76K context 2026-08-20 14:11:25 +02:00
Mikei386 fd4a3055e8 Rebuild profile router as secure recoverable v2 2026-08-20 13:41:00 +02:00
Mikei386 bdd6c08643 Document complete recovery requirements and reference state 2026-08-20 13:05:27 +02:00
Mikei386 0e4a9de5ba Expand router into reproducible local AI platform 2026-08-20 12:56:53 +02:00
Mike a84220725a Vision: Q3-Augen-Orchestrierung + Folgefragen-Sanitize
- Automatische Vision-Orchestrierung: temporärer Q3-Vision-Server
  (llama.cpp + mmproj) analysiert Bilder, Hauptmodell (Fast/Medium/Long)
  erzeugt die finale Antwort. Zentrale GPU-Lock (Ausschluss mit FLUX),
  Drain, Timeouts, Restore in jedem Fehlerfall.
- Vision-Analyse wird als interne User-Message injiziert (nie als
  Assistant-Turn in Open WebUI).
- Analyse-Cache (LRU, keyed by Bild-Hash): Folgefragen triggern keinen
  neuen Hotswap.
- Sanitize: alle Bild-Parts (image_url/Base64) werden bei jedem Request
  durch die gecachte Analyse ersetzt, Bilddaten entfernt - das
  Nicht-Vision-Hauptmodell bekommt keine Bilder mehr (kein
  image input is not supported).
- /v1/streams/lookup fail-fast während Vision-Swap; /props-Regression
  behoben; POST /vision/test für direkten Test.
- systemd-Unit: VISION_*-Umgebungsvariablen; README dokumentiert Vision.
2026-08-20 07:48:25 +02:00
Mikei386 fedcb0b6d3 Cleanup: letzter Kokoro-Reference entfernt 2026-08-19 22:02:59 +02:00
Mikei386 800d83a91b Cleanup: alte Kokoro-Dateien entfernt 2026-08-19 22:02:49 +02:00
Mikei386 d399d2b4f7 TTS: Kokoro → XTTS-v2 (CPU-only, Claribel Dervla)
- Neues xtts_worker.py: Coqui XTTS-v2, HTTP-API auf Port 8085
- Router: TTS_WORKER_URL → 8085, TTS_MODEL → xtts-v2, TTS_VOICES → claribel
- deploy: mike-ai-xtts.service, install.sh + deploy.sh aktualisiert
- Tests: 54/54 bestanden (mock_tts_worker + test_local.sh auf XTTS umgestellt)
- README: TTS-Section auf XTTS-v2 aktualisiert
- Kokoro-Service gestoppt und deaktiviert (Dateien bleiben als Backup)
2026-08-19 22:01:58 +02:00
Mikei386 ed471196d4 router: HTTP/1.1 chunked Transfer-Encoding-Unterstützung
- Zentrale _read_body() Helper für Content-Length und chunked encoding
- _read_chunked_body(): hex Chunk-Größen, Chunk Extensions, 0-Chunk,
  Trailer, MAX_UPLOAD_SIZE (50 MB Default)
- Malformed chunked encoding → HTTP 400
- _transcribe(), _speech(), _image_generate(), _forward() nutzen Helper
- Temporäres STT-DEBUG-Logging entfernt
- 16 Regressionstests (dev/test_chunked.py):
  Content-Length, chunked, Multi-Chunk, Boundary über Chunk-Grenzen,
  Chunk Extensions, 0-Chunk mit Trailer, malformed Chunk Size,
  Uploadgrößenlimit, Open-WebUI-artiger WebM-Request
2026-08-19 16:31:22 +02:00
Mikei386 e81f2ab515 router: Multipart-Parser auf email.parser.BytesParser umgestellt
- Robustes MIME-Parsing mit Standardbibliothek (email.parser.BytesParser)
- Handhabt quoted und unquoted Boundaries (Open-WebUI-Kompatibilität)
- Beliebiges Feldreihenfolge, zusätzliche Header, binäre Payloads
- Regressionstests: dev/test_multipart.py (8 Tests)
- Fix für Open-WebUI HTTP 400 bei POST /v1/audio/transcriptions
2026-08-19 14:37:24 +02:00
Mikei386 51ef07c874 router: deutsche Spracherkennung mit whisper.cpp (CPU-only)
- STT-Worker (stt_worker.py): langlebiger HTTP-Service auf Port 8084
  - whisper-cli als Subprozess (CPU-only, 8 Threads)
  - Audio-Vorbereitung via ffmpeg (WebM/Opus/M4A → 16 kHz WAV)
  - Nativ: WAV, MP3, OGG, FLAC
  - Health-Endpunkt: GET /status
  - Transkription: POST /transcribe (Multipart-Form-Data)

- Router-Integration:
  - POST /v1/audio/transcriptions (OpenAI-kompatibel)
  - GET /v1/audio/models (whisper-1, kokoro-german)
  - GET /v1/audio/voices (martin, victoria)
  - /status mit stt-Section
  - model=whisper-1 akzeptiert
  - response_format: json, verbose_json

- systemd-Service: mike-ai-whisper.service
  - Boot-Start, Restart on failure, journald
  - CPU-only, kein GPU-Lock

- Deploy-Dateien aktualisiert (deploy.sh, install.sh)
- Mock-STT-Worker für lokale Tests (dev/mock_stt_worker.py)
- Tests ergänzt: STT Status, WAV, language=de, unbekanntes Modell,
  Worker down, Recovery, Audio-Modelle, Audio-Voices,
  STT+Qwen parallel, STT+TTS parallel
- README.md: STT-Section mit Endpunkten, Benchmarks, Doku

Benchmarks (CPU-only, 8 Threads):
  7.3 s Audio → 8.9 s (RTF 1.22×)
  30 s Audio → 16.8 s (RTF 0.56×)
  50 s Audio → 18.5 s (RTF 0.37×)
  RAM: ~1.7 GB (Modell), Worker: ~20 MB
2026-08-19 13:53:22 +02:00