mnemo

Persistent memory for AI coding agents

A local-first memory layer that gives Claude Code, Codex, Cursor and more a searchable history of decisions across sessions.

Get started View documentation
mnemo init --agent=all

ARCHITECTURE / 01

How mnemo remembers

Agents publish through one controller. Shared project memory comes back to all of them.

supported agents Claude Code Codex Cursor OpenCode Pi hooks · MCP · native tools durable controller event bus · sole SQLite writer mem_save() · lifecycle events carries .mnemo id project memory SQLite · FTS5 observations · tags · topic keys shared across every supported agent mem_search() · mem_context() → agents cross-agent recall

CAPABILITIES / 02

Built for agent workflows

Durable storage, native tools, and one project-scoped store shared by every supported agent.

Durable observations

SQLite + FTS5 stores structured memories with tags, topic keys and review states — searchable across every session and project reset.

MCP tools

Agents call mem_save, mem_search, and mem_context natively — memories survive context resets, compactions, and new sessions automatically.

Multi-agent

Claude Code, Codex, Cursor, OpenCode and Pi share one project-scoped store — a single .mnemo marker activates every integration.

START / 03

Activate a project

Install the binary, wire your agents, and opt in with a local .mnemo marker. Unmarked projects stay untouched.

Get started View documentation
mnemo init --agent=all