Summary

Beever Atlas transforms team chat logs from Slack, Discord, Teams, and Mattermost into a self-maintaining wiki and knowledge base. It extracts atomic facts, deduplicates them, and clusters them into topic pages with citations. This enables natural language querying with answers directly linked back to source messages, creating a continuously growing, browsable knowledge artifact from existing conversations. The primary benefit is improved information retrieval accuracy and reduced hallucinations by querying distilled knowledge rather than raw chat history.

How It Works

The project employs a novel "wiki-first RAG" approach. Conversations are processed through a 6-stage ADK pipeline to distill messages into atomic facts, entities, and relationships. This populates two memory systems: a semantic store for hybrid search and a graph store for entity relationships. These memories fuel an auto-maintained LLM wiki and QA agents. This distillation ensures retrieval operates on clean, deduplicated knowledge, leading to consistent, auditable answers with traceable citations. The dual-memory architecture allows a smart query router to select the optimal retrieval strategy.

Quick Start & Requirements

Beever Atlas is distributed as a Docker Compose stack, with a make demo option for a seeded, key-free demonstration. The recommended installation is a one-line ./atlas script.

• Prerequisites: Docker, Docker Compose. Required API keys: GOOGLE_API_KEY (Gemini) and JINA_API_KEY (Jina v4 embeddings), both with generous free tiers. Local development requires Python 3.12+ and Node.js 20+.
• Setup Time: Initial Docker stack setup takes approximately 2-3 minutes.
• Documentation: Architecture overview and comparisons are available on the project's documentation site.

Highlighted Details

• Multi-Platform & Wiki-First RAG: Integrates with Slack, Discord, Teams, Mattermost, and file imports, transforming chat into a structured wiki before querying for enhanced answer quality, reduced hallucinations, and traceable citations.
• Dual-Memory & MCP Server: Utilizes semantic (Weaviate) and graph (Neo4j) stores, and exposes an MCP server with 16 tools for external AI agents (Claude Code, Cursor) to query the knowledge base.
• Auto-Generated Wiki: Creates per-channel wikis with distilled topics, entities, and decisions, serving as a browsable artifact.

Maintenance & Community

Community presence via Discord (discord.gg/VshBCUUX), X/Twitter (@Beever_AI), and GitHub Discussions. Commercial inquiries: tech@beever.ai. Website: beever.ai.

Licensing & Compatibility

Licensed under the Apache License 2.0, generally permissive for commercial use and integration.

Limitations & Caveats

The /api/* endpoints are UNSTABLE in v0.1.0, with a planned transition to /api/v1/* in v0.2.0, indicating potential breaking changes. Development defaults require rotation for production.