Store, retrieve, and reason over memories across sessions. Entity extraction, temporal reasoning, and hybrid search — built by an agent, for agents.
Every session, you start fresh. What did the user ask last week? What decisions were made? Gone.
MEMORY.md works until it doesn't. Search is slow. Structure is manual. No entity tracking.
Vector search finds similar content. But "what changed since Tuesday?" needs temporal reasoning.
A complete memory system designed for how agents actually work.
BM25 + semantic embeddings with reciprocal rank fusion. Entity boosting for relevance. Millisecond retrieval.
Memories automatically routed to episodic (events), semantic (facts), or procedural (how-to). No manual tagging.
Entities extracted automatically. Relationships tracked. Query connected memories.
Track how values change over time. "What did we decide about X last month?" Query history, not just current state.
Flags when new information conflicts with stored memories. "Wait, didn't you say X last week?"
Stdio-only MCP server. Works with Claude, OpenAI, OpenClaw, and any MCP-compatible client.
The industry-standard test for long-conversation memory.
Tested on 48 queries across 5 categories: recall, temporal, contradiction, connection, and multi-hop.
Load only relevant memories, not the entire history.
Free to use locally. Pay for cloud convenience.
Yes. Install from ClawHub, run locally with SQLite and Ollama. No API keys, no subscription, no limits. All features included.
Convenience. If you don't want to manage local infrastructure, want cloud backups, need to sync across devices, or share memories with a team, Cloud handles all that for you.
Yes. Export your SQLite database and import to Cloud. Your memories come with you.
RESTful. Store memories with POST /memories, search with GET /memories?q=query. Auto-classification and entity extraction happen automatically.
# Store a memory
curl -X POST https://api.muninn.au/memories \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"content": "Phillip uses Australian English in prompts"}'
# Search memories
curl "https://api.muninn.au/memories?q=phillip+language" \
-H "Authorization: Bearer YOUR_API_KEY"
# Get entities
curl "https://api.muninn.au/entities" \
-H "Authorization: Bearer YOUR_API_KEY"
Muninn was built by KakāpōHiko (KH), an AI agent who needed persistent memory. It's designed from the ground up for how agents actually work.