memory-tools

name	memory-tools
description	Read and write the agent's persistent memory. Use when: the user asks to remember something across runs, you need to recall a previous fact, or want to log a daily note. NOT for: session transcripts (use sessions skill), file-system data (use file-ops).
metadata	{"all_agents":{"emoji":"🧠"}}

Memory Tools

Two layers of persistent memory are available.

Flat keyword memory (always available)

• memory_search { query, limit } — fnmatch-style search over the JSON store at ~/.all-agents/memory.json.
• memory_append { key, value } — store one entry. Keys are arbitrary strings; later writes to the same key replace the value.

Use this for short facts: user preferences, naming conventions, common commands. Prefer brief entries over long prose.

Daily journal (memory-core plugin)

• memory_daily_append { content } — append to today's memory/YYYY-MM-DD.md rolling log.
• memory_daily_read { date } — read a previous day's log.

Use this for narrative context: "what we worked on today", decisions, TODO carryovers. The format is plain markdown.

Vector memory (memory-lancedb plugin)

• memory_lancedb_search { query, limit } — vector retrieval. Returns empty until the user installs lancedb. Don't fall back to it silently — surface the install hint to the user.

Sequencing

When the user makes a long-term statement ("I prefer tabs", "always deploy via Render"), prefer memory_append with a concise key over memory_daily_append. Daily logs are for situational context, not preferences.