v0.2.2
v0.2.2
Knowledge Graph + AI skills + usage analytics + security hardening, with the
observational-ingest stability fixes.
Features (since 0.1.x):
- Knowledge Graph (static + observational), served over MCP via
kg_how_to_obtain; visual editor; optional LLM enrichment (OpenAI/OpenRouter/Anthropic). - AI skills from captured intents, composed into MCP server instructions; search/tabs/pagination, auto-apply, AI consolidate.
- Usage & cost analytics (
/analytics,/api/audit/breakdowns). - Security: SSRF guard on the DatabaseEngine; RLS foundation (off by default); prototype-pollution guard on outputSchema inference.
Stability fixes (0.2.1 → 0.2.2):
- kg-discovery observational ingest streamed in pages with per-page flush + a JSON-walk node cap → bounded memory/CPU (peak ~1 GB, was ~3 GB → OOM).
start.shexits (→ container restart) if either process dies; backend V8 heap cap.
All migrations are additive and auto-applied on container start.