Aimemo

What is Aimemo?

Aimemo is a zero-dependency MCP memory server for AI agents, providing persistent, searchable, local-first memory as a single Go binary. It integrates with Claude Code and OpenClaw, storing memory in a SQLite database inside a .aimemo/ directory next to the project code.

How to use Aimemo?

Install via one-line curl script or Homebrew, then run aimemo init in the project root. Register the MCP server with claude mcp add-json aimemo-memory '{"command":"aimemo","args":["serve"]}' and restart Claude Code – the server loads automatically on each session.

Key features of Aimemo

• Single binary, no Docker, Node.js, cloud, or API keys.
• Memory stored in .aimemo/ next to project code, portable across branches.
• Auto-loads prior context via memory_context tool at session start.
• Full-text search using SQLite FTS5 with BM25 scoring (recency + frequency).
• Concurrent writes supported via SQLite WAL mode.
• Full CLI control: inspect, edit, retract, export Markdown or JSON.

Use cases of Aimemo

• AI coding assistant resumes work on a project after days away.
• OpenClaw skills isolate memory per skill using context names.
• Record decisions, bugs, and progress across multiple sessions.
• Search past observations without repeating instructions.
• Export memory as Markdown or JSON for documentation or backup.

FAQ from Aimemo

What dependencies does Aimemo require?

Zero dependencies. It is a single Go binary with no Docker, Node.js runtime, cloud account, or API keys required.

How is memory isolated per project?

Memory is stored in a .aimemo/ directory within the project root, discovered by walking up directories (similar to Git). This keeps memory project-local and branch-aware.

Can multiple Claude sessions write concurrently without corruption?

Yes, SQLite WAL mode allows multiple Claude windows to write simultaneously without locking each other out.

How does the search ranking work?

Full-text search uses SQLite FTS5 with BM25 scoring weighted by recency and access frequency. Relevant memories surface first; older noise fades naturally.

Is the memory data human-readable?

Yes, you can export all memory to Markdown or JSON using aimemo export --format md or aimemo export --format json.