memvid-core

Memvid Memory Skill

This skill enables natural memory storage and retrieval using a dual-memory architecture.

Memory Scopes

Project Memory (.memvid_data/)

Stores project-specific knowledge:

• Code architecture decisions: "This project uses FastAPI for APIs"
• Bug reports: "Bug in auth.py line 42"
• Technical implementation: "We use MemvidMemory class for storage"
• Project configuration: "Dependencies managed via pyproject.toml"

User Memory (~/memvid_data/)

Stores personal preferences:

• Coding preferences: "I prefer pytest over unittest"
• Style preferences: "I like clean code with type hints"
• Tool preferences: "I use VS Code for development"
• General knowledge: "I always document functions with docstrings"

Automatic Classification

The system uses semantic similarity to auto-classify memories:

• Compares content against project and user examples
• Assigns to scope with higher similarity
• Falls back to user memory if confidence < 0.65

Commands

memvid:store

Store a new memory with automatic scope detection.

Trigger patterns:

• "remember that this project uses FastAPI"
• "note that I prefer pytest"
• "for future reference: bug in line 42"

Action: Call memvid_store tool with:

{
  "content": "<extracted content>",
  "scope": "auto"
}

Response format: "Stored in [project/user] memory: [brief summary]. (ID: abc-123)"

memvid:recall

Search both memories and return merged results.

Trigger patterns:

• "what did I say about testing?"
• "recall memories about authentication"
• "do you remember my coding preferences?"

Action: Call memvid_query tool with:

{
  "query": "<extracted query>",
  "limit": 10
}

Response format: Present results naturally: "Based on your memories:

• [Project] This project uses FastAPI for REST APIs
• [User] You prefer pytest over unittest"

memvid:forget

Delete a memory by ID.

Trigger patterns:

• "forget memory abc-123"
• "delete that memory about authentication"