Memory Recall

Search and apply stored memories from the autopilot system to inform current work. Memories exist at two scopes (user and project) and are searchable via the autopilot-memory MCP tools.

When to Recall

Explicit Recall

The user directly asks about stored memories or knowledge:

•"What do you remember about my preferences?"
•"What conventions does this project follow?"
•"Do you remember how I like commit messages?"

Implicit Recall

Proactively search memories when:

•Starting work on a task that might have relevant conventions
•Making a decision where user preferences would apply
•Encountering a pattern that seems familiar

Recall Process

1. Search Memories

Use the autopilot-memory MCP tool memory_search:

code

memory_search({
  query: "descriptive search query",
  scope: "all",    // or "user" / "project"
  limit: 10,
  category: null   // or specific category
})

Search strategies:

•For preferences: search by topic ("commit messages", "indentation", "testing")
•For conventions: search by technology ("TypeScript", "SwiftUI", "API")
•For patterns: search by workflow ("deployment", "review", "branching")
•For broad recall: use memory_stats first to see what's available

2. Present Results

For explicit recall requests, present memories clearly:

code

Here's what I remember about [topic]:

**Preferences:**
- Use Bun over NPM (importance: 0.9)
- Conventional commit style (importance: 0.9)

**Conventions (this project):**
- Zod validation on all API routes (importance: 0.8)
- Supabase with RLS policies (importance: 0.8)

**Patterns:**
- Run tests before committing (importance: 0.7)

For implicit recall, apply memories without announcing their source. Do not say "I found a memory that says..." — simply follow the stored preference as if it were built-in knowledge.

3. Apply Memories

When memories are relevant to current work:

•Follow preferences without asking (they've been explicitly stored)
•Apply conventions to code being written
•Follow workflow patterns naturally
•If a memory contradicts current instructions, follow the most recent instruction

Memory Analytics

Use memory_stats to provide overview:

code

memory_stats({ scope: "all" })

Present as:

code

Autopilot Memory Stats:
- Total memories: 47
- User scope: 32 | Project scope: 15
- By category: preferences (18), conventions (12), patterns (9), corrections (5), workflows (3)
- Most recent: 2026-01-30

Forgetting

When the user wants to remove a memory:

code

memory_forget({
  query: "the preference to forget",
  scope: "user",
  confirm: false  // first search to show what would be deleted
})

Show matches first, then confirm before deleting:

code

memory_forget({
  query: "the preference to forget",
  scope: "user",
  confirm: true  // actually delete
})

Additional Resources

Reference Files

For advanced recall patterns and troubleshooting:

•references/recall-strategies.md — Search optimization and result ranking