Mimir provides AI agents with persistent memory and task management capabilities by building a knowledge graph from indexed files and stored interactions. It targets developers and multi-agent systems, enabling AI to learn and recall context across sessions, improving complex project management and AI workflow efficiency.

How It Works

Mimir functions as a Model Context Protocol (MCP) server, leveraging Neo4j for its graph database and AI embeddings for semantic search. It stores tasks, context, and relationships, creating a dynamic knowledge graph. An OpenAI-compatible Chat API integrates Retrieval-Augmented Generation (RAG) and MCP tools, allowing AI assistants to access persistent memory and perform complex operations.

Quick Start & Requirements

Requires Docker Desktop, Node.js 18+, and Git. Installation involves cloning the repository, copying .env.example to .env, configuring HOST_WORKSPACE_ROOT, and running npm run start or docker compose up -d. Setup is estimated at 5 minutes. Links to VS Code and IDE integration guides are available.

Highlighted Details

• Graph-RAG: Integrates Neo4j's graph relationships with vector embeddings for context retrieval beyond simple similarity.
• Multi-Agent Orchestration: Facilitates coordinated agent workflows with features like task locking and dependency management.
• MCP Integration: Acts as an MCP server, enabling seamless interaction with AI assistants like Claude and ChatGPT.
• PCTX Integration (Code Mode): Supports sandboxed TypeScript execution for AI agents, drastically reducing token usage (up to 98%) for complex tasks.
• Codebase Indexing: Indexes various file types (including PDF, DOCX) into the knowledge graph, respecting .gitignore.

Maintenance & Community

The provided README does not detail specific contributors, community channels (like Discord/Slack), or roadmap specifics beyond upcoming features.

Licensing & Compatibility

Licensed under the MIT License with additional terms for AI/ML systems; users should review the LICENSE file for specifics. It is designed for self-hosting and integration with various AI assistants.

Limitations & Caveats

The Orchestration Studio is noted as being in beta. The "additional terms" in the license may impose restrictions beyond standard MIT. Full functionality requires configuring external LLM providers (OpenAI, Ollama, Copilot).