agentmemory

agentmemory addresses the recurring friction of AI coding agents forgetting project context between sessions: rather than re-explaining architecture decisions, conventions, or past debugging context every time, agentmemory gives agents persistent memory that carries across sessions and across which specific CLI tool is being used.

It’s built on the iii engine (itself an open-source project for composing and observing services), and works with a wide range of coding agent surfaces — Claude Code, GitHub Copilot CLI, Cursor, Gemini CLI, Codex CLI, Hermes, OpenClaw, Pi, OpenCode, and any MCP-compatible client — rather than being tied to a single vendor’s tool.

Apache-2.0 licensed and translated into a dozen languages, agentmemory has grown extremely quickly (24,000+ stars, Trendshift-featured), reflecting broad demand for a memory layer that isn’t locked to one specific coding agent product.

What You Get

• Persistent memory for coding agents that carries across sessions instead of resetting each time
• Broad compatibility across Claude Code, GitHub Copilot CLI, Cursor, Gemini CLI, Codex CLI, Hermes, OpenClaw, Pi, OpenCode, and MCP clients
• A memory layer built on the iii engine rather than a bespoke, single-purpose memory store
• Documentation translated into a dozen languages for international adoption

Common Use Cases

• Keeping a coding agent aware of project conventions and past decisions across many separate sessions
• Sharing consistent memory across multiple different coding agent tools used on the same project
• Avoiding repeated re-explanation of architecture, debugging history, or team conventions to an AI agent
• Adding memory to any MCP-compatible agent without depending on a vendor-specific memory feature

Under The Hood

Architecture agentmemory builds on the iii engine — a separate open-source project for composing and observing services — rather than implementing memory storage and retrieval from scratch, meaning its core persistence and service-composition logic inherits from an already-established engine. Its broad client compatibility (across many distinct coding agent CLIs and any MCP client) implies an adapter-style integration layer sitting between the memory store and each specific tool’s own extension or plugin mechanism.

Tech Stack TypeScript, layered on top of the iii engine, with MCP as a key integration protocol for connecting to compatible agent clients. Distributed with extensive localized documentation across a dozen languages.

Code Quality Very active, consistently maintained commit history combined with extremely rapid star growth (Trendshift-featured, 24K+ stars) reflects strong real-world adoption; the multi-language documentation investment further suggests a project treating broad accessibility as a priority alongside core functionality.

What Makes It Unique Most agent memory tools are built for one specific coding assistant; agentmemory’s explicit support across nearly every major coding agent CLI and any MCP client makes it a cross-tool memory layer rather than a single-vendor feature, which matters for teams or individuals using more than one coding agent day to day.

Self-Hosting

Licensing Model Apache-2.0 licensed — fully open source with no license key.

Self-Hosting Restrictions None found; it runs as a local memory layer integrated with your chosen coding agent tools.

License Key Required No.