agentmemory: Persistent Memory for AI Coding Agents

Introduction

agentmemory is a groundbreaking open-source repository that provides persistent memory for AI coding agents. It addresses the common problem of agents forgetting context between sessions, which often leads to repetitive explanations and wasted time. By silently capturing agent actions, compressing them into searchable memory, and injecting relevant context, agentmemory ensures your coding agent remembers everything, from architectural decisions to specific code patterns.

Built on the iii engine, agentmemory offers impressive retrieval accuracy (95.2% R@5), significant token savings (up to 92% fewer tokens), and supports a wide array of agents through MCP or REST API. It operates without external databases, relying on SQLite and the iii engine for its robust memory management.

Installation

To get started with agentmemory, you can install it globally via npm or use npx for a quick run. First, ensure you have Node.js (>= 20) and iii-engine or Docker installed.

npm install -g @agentmemory/agentmemory
agentmemory # start the memory server on :3111
agentmemory demo # seed sample sessions + prove recall

Alternatively, for a quick start without global installation:

npx @agentmemory/agentmemory

For detailed agent-specific wiring and configuration, please refer to the official documentation.

Examples

To see agentmemory in action, run the demo command:

npx @agentmemory/agentmemory demo

This command seeds three realistic sessions (e.g., JWT auth, N+1 query fix, rate limiting) and performs semantic searches against them. You'll see it find "N+1 query fix" when you search "database performance optimization", a capability beyond simple keyword matching. You can also open the real-time viewer at http://localhost:3113 to watch memory build live and replay sessions. agentmemory supports integration with various agents like Claude Code, Cursor, and Codex CLI, allowing them to leverage its persistent memory capabilities.

Why Use agentmemory

AI coding agents often forget context between sessions, leading to repetitive explanations and wasted time. agentmemory solves this by providing a robust, searchable memory system that silently captures and compresses agent interactions. Unlike built-in agent memory solutions, which often have limited capacity (e.g., 200-line caps) and basic search, agentmemory offers unlimited scale, advanced hybrid search (BM25, vector, and knowledge graph), and significant token savings (up to 92% fewer tokens).

It supports multi-agent coordination, ensuring that all your agents share a common, evolving understanding of your projects. Real-world benchmarks show 95.2% retrieval accuracy and a cost of approximately $10/year for token usage, making it a highly efficient and effective solution for enhancing your AI coding workflow. With agentmemory, your agent just knows, eliminating the need for constant re-explaining.

Links

For more information, examples, and detailed usage instructions, visit the official resources:

• GitHub Repository
• npm Package