rohitg00/agentmemory

Project overview

A local MCP server providing persistent memory for AI coding agents, reducing re-explanation of project context and token costs. It captures tool usage via hooks, manages memory evolution with auto-forgetting, and operates locally using the all-MiniLM-L6-v2 embedding model.

Project type

MCP · AI Agent · RAG

Use cases

Coding & Development

Deployment

Refer to project documentation

License

Apache-2.0

Best for

• Developers and AI engineers using MCP-compatible coding agents who need local memory persistence without relying on external paid services.
• Developers looking to capture tool usage automatically and manage memory lifecycle through TTL expiry and contradiction detection.

Key capabilities

• Records every tool use via hooks with zero manual effort required from the user.
• Manages memory versioning, supersession, relationship graphs, TTL expiry, contradiction detection, and importance eviction.
• Strips API keys, secrets, and private tags before storing memories.
• Ships a stdio MCP server exposing 53 memory tools to any MCP-compatible client.

Limitations and risks

• Requires a specific pinned version of the iii-engine (v0.11.2) and cannot attach to a different version.
• Telemetry practices and data boundary policies are not documented.

Getting started

• Provides a single npx command to install and run the server. Setup is rated as easy due to this single command.

Evidence and sources

• GitHub project description: #1 Persistent memory for AI coding agents based on real-world benchmarks
• README: Your coding agent remembers everything. No more re-explaining.
• README: | **agentmemory** | **95.2%** | **98.6%** | **88.2%** |
• README: | **Automatic capture** | Every tool use recorded via hooks — zero manual effort |
• README: Triple-stream retrieval combining three signals:

Reviewed at: 2026-07-17T10:53:22Z