The Memory OS for
Autonomous Agents
Graph-Powered Persistent Memory + Multi-Agent Orchestration
Give your AI a brain that remembers, learns, and builds knowledge over time. Free, open-source, and enterprise-ready.
Why Mimir Exists
Most AI agents forget everything between conversations. Mimir changes that.
Graph-Powered Memory
Not just a vector store—a living knowledge graph that understands relationships between tasks, files, and concepts. ACID transactions guarantee your data is always consistent.
Multi-Agent Coordination
PM → Worker → QC workflows out of the box. Multiple agents share a single brain, preventing conflicts with built-in locking and coordination.
Code-Aware Intelligence
Automatically indexes your codebase, tracks architecture decisions, and surfaces relevant context. Perfect for developers managing complex projects.
Semantic Search
Find tasks by meaning, not keywords. Vector embeddings + graph traversal = discover connections you didn't know existed.
Enterprise-Ready Security
RBAC, OAuth 2.0, audit logs, multi-tenant isolation. Built on Neo4j's battle-tested transactional guarantees.
100% Self-Hosted
Your data stays on your infrastructure. No vendor lock-in, no usage limits, no subscription fees. MIT licensed and truly free.
What Makes Mimir Different
The only open-source solution combining Graph-RAG with multi-agent orchestration
🕸️ Native Graph Intelligence
Competitors store isolated vectors. Mimir understands how everything relates—tasks, files, concepts, and decisions.
🤝 Multi-Agent by Design
Built-in PM/Worker/QC patterns. Most frameworks have no concept of shared agent memory or coordination.
💾 ACID Transactions
Neo4j guarantees data consistency. Vector databases don't—your data could be corrupted and you'd never know.
🔌 MCP Protocol Native
Direct integration with Claude, ChatGPT, and any MCP-compatible assistant. No custom adapters needed.
📊 Visual Debugging
Mimir Studio lets you see agent memory, task dependencies, and knowledge graphs in real-time. Competitors give you logs.
💰 Truly Free Forever
MIT licensed with no catches. Pinecone starts at $70/mo, Weaviate at $25/mo. Mimir: $0.
Feature Comparison
See how Mimir stacks up against other solutions
| Feature | Mimir | Pinecone | Weaviate | Milvus | Qdrant |
|---|---|---|---|---|---|
| Graph Relationships | ✅ Native | ❌ None | ⚠️ Limited | ❌ None | ❌ None |
| Vector Search | ✅ Yes | ✅ Yes | ✅ Yes | ✅ Yes | ✅ Yes |
| ACID Transactions | ✅ Full | ❌ None | ❌ None | ❌ None | ❌ None |
| Multi-Agent Support | ✅ Built-in | ❌ None | ❌ None | ❌ None | ❌ None |
| MCP Integration | ✅ Native | ❌ None | ❌ None | ❌ None | ❌ None |
| Self-Hosting | ✅ Free | ❌ Cloud-only | ✅ Yes | ✅ Yes | ✅ Yes |
| Open Source | ✅ MIT | ❌ No | ✅ Yes | ✅ Yes | ✅ Yes |
| Starting Cost | 💰 $0 | 💰 $70/mo | 💰 $25/mo | 💰 $0 | 💰 $0 |
Get Started in 60 Seconds
One command to launch. That's it.
# One-Command Remote Installation (Recommended)
curl -fsSL https://raw.githubusercontent.com/orneryd/Mimir/main/install.sh | bash
# Or Manual Installation:
# Clone the repository
git clone https://github.com/orneryd/Mimir.git
cd Mimir
# Run setup script (checks dependencies, creates .env)
./scripts/setup.sh
# Start all services (auto-detects your platform)
npm run start
# You're done! 🎉
# - Portal: http://localhost:9042/portal
# - Studio: http://localhost:9042/studio
# - Neo4j: http://localhost:7474Built for Scale
Vision Intelligence Built-In
See how Mimir understands images with multimodal embeddings
1. Image Input
Your game character, screenshot, or any visual content
2. AI Understanding
"The image depicts a character dressed in a dark, Victorian-era outfit, standing in front of a traditional Asian-style building with a tiled roof. The character is wearing a top hat, round glasses, and a long coat adorned with various accessories, including a belt with multiple pouches and what appears to be a small weapon or tool attached to the belt. The character's attire suggests a blend of Western and Eastern influences, possibly indicating a fantasy or steampunk setting..."
Vision model extracts rich semantic meaning
3. Knowledge Graph
Searchable by meaning, not just metadata
🚀 Why This Matters
Find images by describing what they contain: "Victorian character with top hat" or "Asian architecture steampunk"
AI automatically connects related concepts across your entire knowledge base
Agents understand visual context when making decisions or generating content
Powered by Qwen2.5-VL • 2B model (fast, 8GB RAM) or 7B model (detailed, 16GB RAM) • Self-hosted • No API costs
Perfect For
Software Development Teams
Track decisions, link tasks to code, maintain architectural knowledge, and never lose context across sprints.
AI Researchers
Build multi-agent systems with persistent memory. Study how agents collaborate and learn over time.
Enterprise Automation
Deploy self-hosted AI with RBAC, audit logs, and transactional guarantees. No data leaves your network.
Knowledge Management
Build living knowledge graphs that grow smarter over time. Perfect for research teams and consultants.
Ready to Give Your AI a Brain?
Join the community building the future of AI memory
MIT Licensed • Free Forever • No Strings Attached
View License →