---

name: recall-memory description: Search and retrieve from memory. Use when user asks about past conversations, "what did we discuss", "find my notes about", "remember when", or needs historical context. allowed-tools: mcp__aiii__recall_memories, mcp__aiii__list_spaces

Recall from Memory

Search and retrieve previously saved memories from aiii's local store.

When to Use

• User asks about previous conversations
• User needs context from past sessions
• User wants to find specific notes or decisions
• Before making a decision that might contradict past context

Instructions

1. Parse the user's query for:

   • Keywords and topics
   • Time references ("yesterday", "last week", "when we discussed X")
   • Space hints ("my work notes", "health stuff")

2. Formulate the search:

   • Extract primary search terms
   • Identify space filter if mentioned
   • Determine result limit (default 5-10)

3. Call recall_memories tool:

   Copy{
     "query": "search terms",
     "space_id": "space_xxx", // optional
     "limit": 10
   }
   

4. Present results clearly:

   • Show title and date for each result
   • Include brief summary preview
   • Indicate the source (claude-code, claude-desktop, etc.)
   • Group by space if multiple spaces returned

5. Offer to expand:

   • Ask if user wants full content of any specific memory
   • Offer to narrow search if too many results
   • Suggest related searches if no results found

Result Format

Copy## Found 3 memories

### Work Space
1. **Auth refactoring: JWT to session-based** (Jan 7, 2026)
   > Refactored authentication from JWT to session-based approach...
   Source: claude-code | Tags: auth, jwt

2. **Rate limiting implementation** (Jan 6, 2026)
   > Added sliding window rate limiter to API endpoints...
   Source: claude-code | Tags: api, security

### Health Space
3. **Vitamin D discussion** (Jan 5, 2026)
   > Discussed optimal vitamin D levels and supplementation...
   Source: claude-desktop | Tags: vitamins, supplements

No Results Handling

If no memories match:

1. Acknowledge the search found nothing
2. Suggest broadening search terms
3. List available spaces that might contain relevant info
4. Offer to help reformulate the query

Example Interaction

User: "What did we discuss about authentication last week?"

Claude:

1. Parses: topic="authentication", time="last week"
2. Searches: query: "authentication", limit: 10
3. Finds 2 results in Work space
4. Presents formatted results
5. Offers: "Want me to show the full content of any of these?"