mem（内存）

mem — Agent Memory Store

A CLI tool for storing and retrieving memories with full-text search. Data is stored locally in ~/.mem/mem.db.

When to Use

• Remember user preferences, project decisions, important facts
• Store code snippets, commands, configurations for later recall
• Search your knowledge base before asking the user for information you may have stored
• Attach images (screenshots, diagrams) to memories

Commands

Three operators: (none) = recall, + = remember, - = forget.

Recall (search, list, get)

mem                             # list recent memories
mem "deploy"                    # full-text search
mem "database" --tag db         # search filtered by tag
mem 7sjtNVyZrNIa                # get full content by ID
mem --tag prefs                 # list filtered by tag
mem "api" --limit 5 --json     # limit results, JSON output
mem --full                      # show full content for all

Remember

mem + "user prefers dark mode" --tag prefs
mem + "deploy: bun build --compile" --tag deploy
mem + "chose SQLite for simplicity" --tag architecture
mem + --image ./screenshot.png --title "Current UI" --tag ui
echo "long content" | mem + --tag notes

Forget

mem - <id>                      # delete one memory
mem - id1 id2 id3               # delete multiple

Piping

mem "old" --json | jq -r '.[].id' | xargs -I{} mem - {}
echo "long content" | mem + --tag notes

Best Practices

1. Tag consistently — Use lowercase, descriptive tags like prefs, api, deploy, db
2. Search before asking — Check if you've stored relevant information before asking the user
3. Store decisions — When making architectural or design decisions, store the reasoning
4. Keep memories atomic — One concept per memory for better searchability

Output Formats

• Default: One-line summary per result
• --full: Complete content inline
• --json: Structured JSON for parsing