v0.2.0 — wiki layer, typed agent terminations, writer pipeline

What's new in v0.2.0

This is the first release with the wiki layer — an always-on background pipeline that turns BrainDB's entity graph into self-maintaining, human-readable pages. Drop a file into data/sources/, the watcher extracts facts, the scheduler decides whether each new fact belongs in an existing wiki / starts a new one / merges duplicates, and the writer agent keeps every page grounded, cited, and self-correcting. Same hands-off posture as the file watcher — no manual driving required.

Headline features

• Autonomous wiki maintenance. Wikis are first-class entities in the graph, not separate files. The maintainer agent triages every orphan (attach / create / consolidate / skip / ambiguous) and the writer agent rewrites the affected page with [[ref:UUID]] citations for every claim. Duplicate wikis auto-merge; conflated identities self-heal.
• Typed agent terminations. Every agent run — main, maintainer, writer, subagent — ends with a validated Pydantic payload via final_answer. A Layer-4 retry-with-correction recovers transparently when a model forgets to call the termination tool.
• Keyword-mediated recall with diversity quotas. Both the fuzzy (pg_trgm) and the embedding pathway in /memory/context match against keyword entities first; facts surface via tagged_with. A two-level diversity quota stops one popular keyword from monopolising results. Multi-item reads return short previews to keep LLM context tight.
• Self-hosted vLLM profiles (advanced / offline). Local Gemma & Qwen profiles next to the hosted deepinfra and nim ones.
• CI in place. GitHub Actions runs the typed-final + handoff unit tests against a pgvector postgres service on every PR.

Try it

git clone https://github.com/dimknaf/braindb.git
cd braindb
cp .env.example .env
# edit .env: set DATABASE_URL + DEEPINFRA_API_KEY (deepinfra is the recommended default)
docker network create local-network
docker compose up -d --build
curl http://localhost:8000/health
# drop a file into data/sources/ and watch the watcher do its thing:
docker logs braindb_watcher -f
# enable the wiki maintainer (opt-in, off by default):
echo WIKI_ENABLED=true >> .env && docker compose up -d --force-recreate wiki_scheduler