Memanto is an open-source memory agent for AI systems: a dedicated companion agent that manages what your other agents should remember, what conflicts, what expires and who needs to know. Released under the MIT license by Moorcheh AI with more than 2,300 GitHub stars and active development in 2026, it gives coding assistants and agent frameworks a long-term memory that survives across sessions.
Agents forget everything when a session ends, and naive vector stores just accumulate stale, contradictory facts. Memanto treats memory as a managed system instead: memories are ingested instantly, categorized autonomously, scored with confidence and information-theoretic measures, and checked for conflicts when new facts arrive. Recall is semantic and temporal โ ask “how do we deploy?” and get the current answer, with verifiable sources. It runs locally in Docker with local embedding models so nothing leaves your machine, and connects to Claude Code, Cursor, Codex, Gemini CLI, Copilot and dozens more tools through simple connectors and an MCP interface.
Memanto is free and open source under the MIT license for self-hosting. An optional Moorcheh Cloud backend offers instant hosted memory with an API key for teams that do not want to run the Docker deployment themselves.
Memanto is for developers and agent builders whose assistants keep re-learning the same project facts โ deployment details, conventions, decisions โ and who want that knowledge to persist, stay consistent and stay private.