Beever-AI/beever-atlas

Your First LLM-Wiki Conversation Knowledge Base

What it solves

Beever Atlas transforms noisy, fragmented chat history from platforms like Slack, Discord, Microsoft Teams, and Mattermost into a structured, self-maintaining wiki. It solves the problem of "knowledge loss" in team communications, where important decisions, facts, and project details are buried in thousands of messages and difficult to retrieve via standard search.

How it works

The system uses a multi-stage ingestion pipeline to distill raw messages into "atomic facts," entities, and relationships. This data is stored in two complementary systems: a 3-tier semantic store (channel, topic, and atomic fact) for hybrid search and a graph store for tracking relationships between people, projects, and decisions.

When a user asks a question, a smart query router chooses between semantic or graph retrieval to provide a cited answer. The system also automatically generates a browsable wiki for each channel, which serves as a pre-digested knowledge base that LLMs can reason over more effectively than raw chat logs.

Who it’s for

Teams using chat-based collaboration tools who want an automated way to maintain a knowledge base without manual documentation effort, as well as developers who want to integrate their team's internal knowledge into AI agents via the Model Context Protocol (MCP).

Highlights

  • Multi-Platform Support: Connects to Slack, Discord, Teams, and Mattermost.
  • Wiki-First RAG: Distills conversations into a structured wiki before querying, reducing hallucinations and improving consistency.
  • Dual-Memory Architecture: Combines a semantic vector store (Weaviate) and a graph store (Neo4j) for complex relational reasoning.
  • MCP Server: Provides 28 tools allowing external AI agents (like Claude Code or Cursor) to query the knowledge base.
  • Traceable Citations: Every claim in the wiki and every answer provided by the QA agent links back to the original source message.