Icm, developed by Rtk AI, is an mcp server tool built with Rust, designed to provide permanent memory for AI agents. Distributed as a single binary, it integrates with numerous AI coding assistants and development environments. This free tool stores agent memories locally in a SQLite database, supporting a local-first operational model. Icm aims to enhance AI agent capabilities by allowing context to persist across different tools and sessions, reducing repetitive tasks for developers. Its architecture supports various installation methods, making it accessible across multiple platforms.

What does Icm connect to?

Icm functions as an AI agents skill and permanent memory system, connecting to a variety of AI coding assistants and development tools. It integrates with Claude Code, Codex CLI, ChatGPT, and Gemini CLI. Other supported tools include Cursor Agent, Aider, Roo, Amp, OpenCode, VS Code, Windsurf, Zed, Amazon Q, Cline, and Kilo Code. For compatible AI clients, Icm works with Claude Code, Gemini CLI, Codex CLI, Copilot CLI, Cursor Agent, Aider, Roo, Amp, OpenCode, VS Code, Windsurf, Zed, Amazon Q, Cline, Kilo Code, and Claude Desktop.

Core features and tools

Icm exposes several CLI commands for direct interaction, including "icm store", "icm recall", "icm init", "icm uninstall", "icm extract", and "icm embeddings download". For integration via hooks, it offers "icm hook start", "icm hook pre", "icm hook post", "icm hook compact", and "icm hook prompt". Additionally, the "icm_memory_store" tool is available for specific integrations like Codex. Icm supports JSON-RPC over stdio and streamable HTTP as transport protocols, ensuring flexible communication with connected AI clients.

Who is it for?

Icm is primarily for developers and users of AI coding agents. It helps those who need AI agents to retain context and knowledge across different sessions and tools. The system is designed for environments where agents benefit from persistent memory and automated context injection. It is suitable for users seeking to reduce the need for AI agents to re-read files or re-discuss decisions. This is not the right fit if you require an explicit authentication model or cloud-based memory storage.

How to get started

No credentials are required. Setup involves running "icm init" with various modes, such as "--mode all", "--mode mcp", or "--mode cli", to configure integration with detected AI tools.

Use cases

• Providing permanent, cross-tool memory for AI agents, allowing context to persist across different AI tools and sessions.
• Enabling automated extraction of facts from tool outputs and injection of recalled context into user prompts, reducing the need for agents to re-read files or re-discuss decisions.
• Supporting multilingual memory storage and recall without additional configuration.

Icm provides a local-first, Rust-based memory solution for AI coding agents, storing data in SQLite. It offers persistent context across various development tools and AI clients, reducing repetitive information processing for agents. The system supports multiple installation methods and integrates through CLI commands and hooks, enhancing agent efficiency. Its design focuses on local data management without requiring external authentication. Try configuring Icm to provide permanent memory for your Claude Code or Gemini CLI agent.