memory

Memory

When to Use

• User asks to remember something for later
• User wants to recall previously stored information
• User needs semantic search across stored knowledge
• User says "remember that...", "what do you know about...", "save this for later"

Instructions

Storing Memories

Use memory_store to save information with optional metadata:

• Store key decisions, preferences, or facts
• Include relevant metadata (category, source, date) for better retrieval
• Keep content focused — one concept per memory for better search results

Searching Memories

Use memory_search for semantic similarity search:

• Queries are matched by meaning, not exact keywords
• Default returns top 5 results — adjust limit for more or fewer
• Results include similarity scores — higher is more relevant
• Falls back to hash-based pseudo-embeddings if no embedding API is configured

Best Practices

• Store information when the user explicitly asks or when context is important for future conversations
• Use descriptive content — "User prefers Python for backend development" is better than "Python"
• Add metadata tags to categorize memories (e.g., { "category": "preference" })

Tools

• memory_store — Save content with optional metadata to long-term memory
• memory_search — Search memories using semantic similarity (returns ranked results with scores)