recall
Semantic search for memory. Use to find solutions, patterns, or context from Chroma Cloud.
When & Why to Use This Skill
The Recall skill provides a powerful semantic search interface for an agent's long-term memory system. By integrating with Chroma Cloud, it enables the retrieval of specific solutions, recurring patterns, and historical context using natural language queries. This tool is essential for maintaining project continuity and ensuring that past decisions and procedural knowledge are instantly accessible to improve decision-making accuracy.
Use Cases
- Technical Troubleshooting: Quickly locate previously documented solutions to complex technical issues, such as specific error fixes or configuration patterns, reducing redundant debugging time.
- Architectural Context Retrieval: Access historical design decisions and best practices stored in the memory system to ensure new implementations align with established project standards.
- Knowledge Base Augmentation: Enhance agent performance by retrieving relevant facts and procedural 'how-tos' from semantic memory to provide more accurate and context-aware responses.
| name | recall |
|---|---|
| description | Semantic search for memory. Use to find solutions, patterns, or context from Chroma Cloud. |
Recall Memory
This skill allows you to search your memory system using semantic queries.
Workflow
Formulate Your Query: Think about what you're trying to find:
- A solution to a specific problem (e.g., "How do I fix CORS errors?")
- A pattern or best practice (e.g., "Python async patterns")
- Historical context (e.g., "What did we decide about routing?")
Run the Search: Execute the memory manager recall command:
uv run python .fleet/context/scripts/memory_manager.py recall "<your query>"Example:
uv run python .fleet/context/scripts/memory_manager.py recall "memory system implementation"Review Results: The system will return:
- Top matches from semantic memory (facts, decisions)
- Relevant skills from procedural memory (how-tos)
- Similarity scores to gauge relevance
- Source metadata (file paths, timestamps)
Refine if Needed: If results aren't relevant, try:
- More specific queries (add context/domain)
- Different terminology (synonyms)
- Breaking complex queries into simpler parts
Tips
- Use natural language - the system uses semantic search, not keyword matching
- Be specific - "fix DSPy routing errors" is better than "errors"
- Combine with other commands: recall → apply solution → learn new variation
- Check episodic memory separately if you need conversation history
Output Format
Results include:
- Matched text snippets
- Source file paths
- Relevance scores (0-1, higher = better match)
- Metadata (creation date, tags, etc.)