Every session your team runs is automatically distilled into knowledge entries — architecture decisions, debugging patterns, API contracts, environment quirks. Your codebase builds its own documentation over time.
AI agents don't just read the knowledge base — they write back. Discoveries, fixes, and patterns flow in automatically. The next agent picks them up.
Every session makes the next session smarter. Fully automatic.
Daemon watches 8 AI tools and captures sessions in real time. Messages, tool calls, file changes, and command outputs — all recorded.
Deterministic summarization extracts files changed, tests run, commands executed, packages installed, and key decisions — no LLM cost.
Extracted facts become knowledge entries — tagged by topic, linked to source sessions, and indexed for search. Backlinks connect related entries.
Related entries are compiled into wiki pages — one page per topic. Pages update automatically as new sessions add information.
Cross-session patterns emerge as concepts. "Auth always uses PyJWT" or "Tests need Redis running" — learned from real sessions, not guesses.
AI agents write discoveries back to the knowledge base during sessions via MCP. The next agent reads them. The loop closes.
When an AI agent discovers something new — a debugging pattern, an API quirk, a configuration requirement — it writes the finding to the knowledge base via the add_knowledge MCP tool. No manual curation needed.
Everything your AI agents need to read, search, and contribute to your team's knowledge.
Knowledge entries compile into topic pages. Each page links to its source sessions and related topics.
| Page | Entries | Sources | Last updated |
|---|---|---|---|
| auth/middleware | 8 | 5 sessions | 2h ago |
| database/migrations | 12 | 9 sessions | 4h ago |
| api/rate-limiting | 5 | 3 sessions | 1d ago |
| deploy/prod-checklist | 15 | 11 sessions | 2d ago |
| testing/fixtures | 6 | 4 sessions | 3d ago |
The health check flags stale entries, conflicting information, and orphaned pages. Keep your knowledge base accurate without manual curation.
Every session your team runs makes the next one smarter. Free during beta.