Documentation
How to use fractal
fractal is long-term memory for your agents: it keeps facts organised by topic, updates the ones that change and gives back only what is needed. Here is how to connect it, what it does and how to call it.
Why not just a vector database
A vector database stores pieces of text and finds them by similarity. fractal builds on top of that a tree that organises itself, with two clocks: when an agent learns a fact that contradicts an old one, the old one expires (it is not deleted) and the new one becomes current. So the question «what did we know on that date» still has an answer.
Every agent has its own memory: nobody reads another one's without explicit permission.
Quickstart
- Open the panel (14-day free trial).
- In Keys, create a key: the key is the agent's identity, no other id is needed.
- Connect your agent with one of the methods below.
claude mcp add --transport http fractal https://getfractal.it/mcp
Want to keep the data in-house? An app you install on your own computer or server is on its way: the memories stay with you.
Connect an agent (MCP)
Claude Desktop, Claude Code, Cursor and any client that speaks the Model Context Protocol connect fractal as a remote server, over MCP's StreamableHTTP transport.
The address is https://getfractal.it/mcp, with the key as a Bearer token. For a client configured with a JSON file:
{
"mcpServers": {
"fractal": {
"url": "https://getfractal.it/mcp",
"headers": { "Authorization": "Bearer fr_..." }
}
}
}
Once connected, the agent finds tools such as memory_add, memory_recall and memory_maintain. The exact list depends on the capabilities you gave that key.
HTTP API
Every MCP tool is also a REST route, under /api/v1/<tool>: same permissions, same behaviour, generated from the same source so the two cannot drift apart. The full list is at /api/openapi.json.
curl -X POST https://getfractal.it/api/v1/memory_add \
-H "Authorization: Bearer fr_..." \
-H "Content-Type: application/json" \
-d '{ "text": "Prefers dark interfaces." }'
The rest of the documentation
- Integrations: LangGraph, LangChain, Pydantic AI and the in-process library.