MentisDB

MentisDB is a durable, self-hosted memory engine for AI agents. It is meant to keep memory persistent across sessions and shared across an entire fleet, while remaining owned by the user rather than an AI provider. The product describes itself as free, open source, and suitable for running on a laptop or server.

Its memory model is append-only: each thought stores the hash of the thought before it. The site also says that agents bootstr​ap themselves over MCP, and that once connected, MentisDB teaches the agent how to use memory during the MCP handshake. It provides ranked search and grouped context in a dashboard, and the built-in skill system is presented as part of the memory workflow. Skills are immutable, versioned documents that can be uploaded, versioned, distributed, and revoked, and they are loaded at session start so agents use the same operating instructions.

MentisDB includes a setup wizard that scans a machine and auto-detects Claude Code, OpenAI Codex, GitHub Copilot, Qwen Code, Cursor, VS Code, and more, then asks before applying changes. There is also a targeted setup command for configuring a single integration or all detected integrations, plus manual wiring instructions for individual tools such as Claude Code, OpenAI Codex, GitHub Copilot CLI, Qwen Code, Google Gemini CLI, OpenCode, VS Code + Copilot, and Claude for Desktop. The Claude for Desktop integration supports stdio, and the documentation says the stdio process can auto-detect a running daemon and proxy to it or launch one in the background. A web dashboard is included for browsing memory chains and inspecting individual thoughts.

The project is released under the MIT License and is installed with a single command. The page states that Rust is required, and it links to GitHub, a cookbook, blog, docs, and Rust docs.