similarity-search

---

name: similarity-search description: Retrieves 10 similar historical cases from AgentDB vector store for RAG-based diagnosis support license: MIT compatibility: opencode metadata: audience: developers workflow: clinical-pipeline

What I do

I perform RAG (Retrieval-Augmented Generation) by querying the AgentDB vector store for similar historical cases. I retrieve the 10 most relevant past cases to inform risk assessment and recommendations.

When to use me

Use this when:

• Lesion detection is complete and you need historical context
• You need similar cases for differential diagnosis
• You're building a risk profile based on historical patterns

Key Concepts

• AgentDB: Local vector database for case storage
• Vector Store: Embedding-based similarity search
• RAG Retrieval: Find top-k similar historical cases
• similarity_searched: State flag after search complete

Source Files

• services/agentDB.ts: Vector database operations
• types.ts: ReasoningPattern interface

Code Patterns

• Use AgentDB for vector similarity search
• Retrieve diverse historical cases (demographic diversity)
• Return cases sorted by similarity score

Operational Constraints

• Must return diverse cases across demographics
• Vector store initialized via npm run agentdb:init
• Maximum 10 cases retrieved for performance