agentralabs/agentic-memory

View on GitHub ↗

Persistent cognitive graph memory for AI agents — facts, decisions, reasoning chains, corrections. 16 query types, sub-millisecond. Rust core + Python SDK + MCP server.

22 ★5 forksRustUpdated 2mo ago

What you need to know

AgenticMemory is an immortal cognitive memory for AI agents, storing facts, decisions, reasoning chains, corrections and skills as a navigable graph in a single binary .amem file, exposed to multi-client agents through 13 MCP tools.

Install

cargo install agentic-memory-cli agentic-memory-mcp
pip install agentic-brain
curl -fsSL https://agentralabs.tech/install/memory | bash

Key features

  • Graph-based cognitive memory in a single .amem file
  • 13 MCP tools
  • Five indexes: temporal, semantic, causal, entity, procedural
  • Append-only log with BLAKE3 integrity chains
  • Hybrid BM25 and vector search
  • Sixteen query types
  • Multi-client auto-sync

Caveats

  • Remote HTTP and SSE transport is planned but not yet available
Clients: Claude Code · Claude Desktop · Cursor · Windsurf · Cody

Topics

agent-memoryagenticai-agentsbelief-revisionbinary-formatclaudecognitive-architecturecursorgraph-databaseknowledge-graphllmmcpmemorymodel-context-protocolpythonrust
Stars
22★
Forks
5
Language
Rust
License
MIT
Created
2026-02-18
Last push
2026-06-13