Mnemos is a local-first memory framework designed specifically for coding agents, with a focus on reliable, scoped memory management. The tool addresses the challenge of agent memory drift and scope confusion by keeping project, workspace, and global memory partitions separate, ensuring that knowledge relevant to each context remains distinct and does not accumulate contradictions. Mnemos is intended for solo coding-agent workflows, where maintaining accurate and adaptive memory across projects and sessions is critical.
The platform operates using a single local SQLite file for persistence, allowing memory to survive restarts while remaining compact and efficient. It features a guided user interface (mnemos ui) for setup and host configuration, making operational readiness and host integration more accessible. Mnemos integrates with Claude Code, Claude Desktop, generic Model Context Protocol (MCP) hosts, and provides documentation for Codex setups. Its architecture is inspired by neuroscience, employing modules such as a Surprisal Gate for predictive coding, Mutable RAG for dynamic memory reconsolidation, an Affective Router for state-dependent retrieval, a Sleep Daemon for episodic memory consolidation and pruning, and Spreading Activation for associative memory priming.
These modules enable Mnemos to selectively encode only surprising or salient information, adapt stored knowledge as new context emerges, blend semantic and affective cues for retrieval, and keep memory stores clean by consolidating and pruning episodic traces. The system is designed to avoid the pitfalls of standard append-only memory layers, which can lead to bloated, contradictory, and operationally opaque memory pools.
Mnemos is open source and can be installed via pip. It does not require extra services, as all retrieval and consolidation processes are handled locally. The framework includes built-in tools such as a graph edges doctor and a health check utility (mnemos_health) to help users inspect and maintain operational readiness. Its tier 1 host support includes Claude Code, Claude Desktop, and generic MCP hosts, with additional documentation for integrating with Codex.
In the Frameworks & SDKs space, mnemos takes a focused approach. It focuses on providing reliable, partitioned memory for coding agents with local storage and agent protocol support. mnemos is an open-source project aimed at AI agent developers. mnemos is open source under the MIT license. It ships for the web and the command line, and it can be self-hosted.
Behind mnemos is making-minds.ai, and it first shipped in 2026. The project is developed in the open on GitHub with 24 stars and 2 commits in the last 90 days. Among its 5 catalogued features are scoped memory, SQLite backend, and MCP support. It exposes integrations via an MCP server.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do