car-search

External code discovery and indexing for Common Agent Runtime.

What it does

Split out from car-agents to keep the LLM-wrapper agents lean. car-search owns the heavy machinery — HTTP, filesystem walkers, GitHub API clients, AST-driven persistence — while car-agents stays as thin inference shims.

Components

• reference_miner — searches for analogous code across external repos via GitHub, local clones, or composed sources, with optional LLM relevance scoring.
• indexer — persists mined references into the Foreign partition of a memory graph for cross-repo retrieval.

Where it fits

When an agent needs to know "how do other people solve this in OSS?", it asks car-search to mine references and the indexer to make them queryable through the same car-memgine retrieval that handles project-local knowledge.