docs(v9): Phase 10 Schritt 1 (GGUF-Capabilities) dokumentiert

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-24 18:10:50 +02:00
parent 9fa065908d
commit ada85504f3
2 changed files with 2 additions and 0 deletions
@@ -26,6 +26,7 @@ FastAPI-Backend + Svelte 5-Frontend (Vite-Build). **Leitprinzip: KISS — SoC oh
  - `hermes.py` — Chat-WS **proxyt zum Hermes-Agent-Server (:8642)** + transcribe(Whisper)/tts(Piper)/status/pubkey + **Cockpit-Reads** agent/cron/skills (Phase 4)
  - `hermes_ui.py` — **Reverse-Proxy** auf das eingebettete Hermes-Web-Dashboard (`/hermes-ui/`, HTTP + WS-Bridge `pty/ws/pub/events`, `X-Forwarded-Prefix`). Der Chat lebt jetzt hier (iframe), nicht mehr im Eigenbau.
 - **`hermes_control.py`** — read-only Control-Plane: liest Agent-Status (`gateway_state.json` + Cron-Heartbeat), Cron-Jobs & Skills via `hermes`-CLI (`COLUMNS=400`, Rich-Tabelle geparst). TTL-Cache.
+- **`model_caps.py`** — Modell-Capability-Tags (Phase 10): dependency-freier GGUF-Header-Reader (architecture/expert_count/context_length/parameter_count, offline) + cmd-Flags (`--jinja`/`--mmproj`) + Familien-Fallback → MoE/Tools(yes|likely|no)/Vision/Coder/Reasoning/Embedding. Genutzt von `models.py` (`meta.capabilities`).

 **Frontend** (`frontend/src/`, Svelte 5 + Vite, Build → `static/dist/`):
 - **`index.html`** — Gerüst: Sidebar-Nav (10 Tabs), Topbar, Alert-Banner, View-Container je Bereich.