A multi-agent coding fleet with shared memory

One agent learns it. Every agent knows it.

error_outline

Without memory

You run Claude Code in one terminal, Cursor in the editor, and a background agent on CI — and each one rediscovers your codebase from scratch. One agent works out that the integration tests need a dedicated database port; an hour later a different agent burns twenty minutes hitting the same wall.

check_circle

With Rekall

Every agent reads and writes the same graph-backed memory. Lessons learned by one agent are recalled by all of them, and the agent-constellation graph tracks which agent knows what.

How it works

Point every agent at Rekall

The MCP server is the universal rail — Claude Code, Cursor, OpenCode, and any MCP-capable agent share one brain.

Hooks capture as they work

Ambient observation capture means agents don’t have to be taught to remember — fixes, gotchas, and conventions land in memory automatically.

The graph connects the dots

Entities and relationships go into a real knowledge graph (Neo4j), and recall fans out across it alongside vector, episodic and procedural memory — five stores, one fused answer to “what do we know about the payments service?”

The constellation tracks expertise

Rekall models which agent has expertise where, so work can route to the agent that already has the context.

See it in action

One memory, every agent

# Claude Code
claude mcp add rekall

# Cursor / Windsurf / Zed / Cline — same MCP server, same brain
npm install -g @rekall/mcp-server

# Agent A learns: "integration tests need the dedicated
# Neo4j on port 7690" → Agent B recalls it at session start