Beever Atlas turns chat into wiki

Wiki-first RAG is a strong bet for messy team-chat knowledge, but it shifts the hard problem from retrieval tuning to extraction quality and source-grounded consolidation.

• –The architecture is more durable than chunk-first RAG for repeated decisions, because facts are deduped and organized before query time.
• –Weaviate plus Neo4j is a sensible split: semantic search handles fuzzy recall, while graph traversal captures people, projects, and decision relationships.
• –The cost tradeoff is real: ingestion needs LLM extraction and consolidation, so local-model quality will decide whether this pattern can scale cheaply.
• –MCP support makes the project more interesting for developers, because the wiki becomes usable directly inside coding agents rather than only through a dashboard.