Memo

Memo is a tiny local memory tool for AI agents.

It stores notes in one SQLite database, searches them with FTS5, and uses local Ollama embeddings with sqlite-vec when available. If Ollama is not running, Memo still works with lexical search.

Principles

• local-first
• one Go binary
• one SQLite database at ~/.memo/memo.db
• small CLI
• small MCP server
• no hosted service required

Install

task build
sudo install -m 0755 bin/memo /usr/bin/memo

Optional semantic search:

ollama pull mxbai-embed-large

CLI

memo add --title "Preference" --body "Use concise answers."
memo search "concise answers"
memo remove 1
memo stats
memo doctor
memo embed
memo mcp
memo agent install --agent codex

Memo creates the database automatically. Override paths with MEMO_HOME, MEMO_DB_PATH, --home, or --db.

MCP

{
  "mcpServers": {
    "memo": {
      "command": "memo",
      "args": ["mcp"]
    }
  }
}

Tools:

• memo_add_memory
• memo_remove_memory
• memo_search_memory
• memo_memory_stats

Agent Install

Memo can install minimal agent memory instructions and MCP config:

memo agent install --agent codex
memo agent install --agent claude
memo agent uninstall --agent codex
memo agent uninstall --agent claude

Manual SQLite Writes

You can insert memories directly into the SQLite memories table. Full-text search is updated by SQLite triggers immediately, but semantic search needs embeddings to be backfilled:

memo embed
memo embed --limit 100