KG Extraction cockpit (AU-ECO.connector.git-task-resolver)¶
The 🧬 KG Extraction sidebar tab (geniusbot/qt/extraction_cockpit.py,
view 12) turns a document or URL into a live, force-directed knowledge graph on
the Qt cockpit — backed by the real gateway, not mock data.
What it does¶
- Ingest — paste text or enter a URL, click Extract.
- Stream — facts stream onto a native
QGraphicsViewforce graph (geniusbot/qt/force_graph.py) as they generate. The layout (relax_layout, a pure Fruchterman-Reingold step) is unit-tested without a display. Node keys are NFKC-normalized so variants merge. - Inspect — click an edge to open its fact card: title, triple, description, confidence %, tags, evidence span, and source file.
- Export — the JSONL button downloads the job's facts.
How it works¶
GatewayClient.submit_and_stream_extraction submits to
/api/enhanced/extract/submit, then streams
/api/enhanced/extract/stream/{job_id}. Each event is forwarded to the panel on
the UI thread via the AgentBridgeWorker progress signal, so facts render live
and thread-safely. The backend (KG-2.64 extractor, KG-2.65 GPU-slot scheduler,
KG-2.66 readability reader) is documented in agent-utilities
docs/architecture/document_fact_extraction.md.