Memory you can branch.
Axonio gives AI agents and robots persistent memory with hybrid vector and knowledge-graph recall, and lets them fork it, test a change, and merge it back. Git, for what your agents know.
npm install @clawdb/sdk- Claude
- Cursor
- Windsurf
- GitHub Copilot
- VS Code
- OpenAI Agents
- Gemini
- LangChain
- CrewAI
- Vercel AI SDK
- Ollama
- Hugging Face
- ROS 2
- NVIDIA Isaac
Fork what an agent knows. Merge what worked.
- › clawdb_search({ query: "refund policy" })→ "Refunds: 14-day window" (score 0.94)
One URL. Every MCP client gets memory.
import clawdb from '@clawdb/sdk';
const db = await clawdb(); // zero config; provisions on first run
await db.memory.remember('Customer prefers email over phone.');
await db.memory.remember('On Pro plan since March 2025.');
const hits = await db.memory.search('how should we contact them?');
// → "Customer prefers email over phone." · score 0.97 · hybrid vector + graphfrom clawdb import clawdb
db = await clawdb()
await db.memory.remember("Customer prefers email over phone.")
await db.memory.remember("On Pro plan since March 2025.")
hits = await db.memory.search("how should we contact them?", top_k=5)
# → "Customer prefers email over phone." · score 0.97{
"mcpServers": {
"axonio": {
"url": "https://api.axonio.xyz/api/v1/mcp",
"headers": { "Authorization": "Bearer YOUR_API_KEY" }
}
}
}- 7Memory
- 8Branching
- 7Knowledge graph
- 7Reflection
- 5Transaction
- 5Sync
- 4Object storage
- 1Status
- P99 retrieval
- 0.38ms1
- Package installs
- 140,000+
- GitHub stars
- 4,800
- MCP memory tools
- 44
Five kinds of memory. One engine for agents and machines.
What happened, in order
Past debugging sessions and user decisions
Waypoint logs and past obstacle encounters
Same engine, on the machine.
Spatial maps, object permanence and learned recoveries live on the robot, offline. One unit learns a reroute; the fleet has it on the next sync.
Run it where your data has to live.
Or just describe it.
Questions, answered.
What is Axonio?
Axonio is an open-source memory engine and hosted MCP server that gives AI agents and robots persistent, branchable memory: hybrid vector and knowledge-graph retrieval with Git-style fork, diff and merge.
What does “memory you can branch” mean?
An agent can fork its memory the way you fork a Git repository: try a new policy or a risky plan on a branch, diff what changed, then merge it into main or discard it. Branches are zero-copy, so forking a large memory costs nothing until something is written.
How is Axonio different from Mem0, Zep or Letta?
All of them persist memory for agents. Axonio adds versioning of that memory (fork, diff, merge), ships as an MIT-licensed engine you can embed or self-host, and exposes 44 memory tools over MCP. See the side-by-side comparisons for details.
Does it work with Claude, Cursor and other MCP clients?
Yes. Point any MCP client at https://api.axonio.xyz/api/v1/mcp with an API key, or run npx -y @clawdb/mcp-adapter locally. Agents get tools such as clawdb_remember, clawdb_search, clawdb_branch_fork and clawdb_query_graph.
Can I move my existing memories over?
Yes. Built-in importers read from Mem0, Zep, Pinecone, Qdrant, Chroma, Weaviate, MongoDB, Postgres/pgvector and JSON/CSV exports, and map content, embeddings and tags automatically.
Can it run on a robot without internet?
Yes. The engine embeds into edge hardware and runs fully offline. When a connection is available it syncs with cloud or fleet memory using clawdb_sync.
Is there a free tier?
Yes. Free forever: three memory databases, 100,000 vector memories and full MCP access, no credit card. Self-hosting the open-source core is also free.