What is mnemo?

A local-first AI memory layer that runs as a sidecar for LLM applications. It exposes a REST API where you send text, and it uses a configurable language model to pull out named entities and their relationships. Everything gets stored in a SQLite database backed by an in-memory knowledge graph. When you query it, the system blends full-text search, entity matching, and multi-hop graph traversal to return ranked context in milliseconds. The software is distributed as one standalone binary with no cloud dependencies and works with Ollama for fully local operation, plus OpenAI, Anthropic, and any OpenAI-compatible API. A CLI tool and Python SDK are included for integration. The knowledge graph deduplicates entities across sessions, weights relationships based on repeated sightings, and expands query results through configurable breadth-first search.

Should you self-host mnemo?

mnemo scores 43/100 on selfhost.directory (how scoring works) and ranks #91 of 125 in AI. It has 237 GitHub stars.

There was no stable release in the past 12 months, so check the repository's recent activity before you rely on it.

Setup is moderate: expect some configuration beyond a single command. It is MIT-licensed and written mostly in Rust.

Its closest alternative on selfhost.directory is ComfyUI (93/100, 143 752 stars). It scores higher overall, so compare the two before you pick. See mnemo vs ComfyUI side by side.