Farol
2026Shows AI coding agents how microservices connect, by linking the strings that join them.
MoreLess
The link between two services is usually a string: a queue task type, a route, an OpenAPI path. An index that reads one repository at a time can't see it. Farol extracts those strings, links them across repositories with provenance and confidence, and serves the result to coding agents over MCP.
Answers with the 5 to 15 symbols to change instead of whole files. On four public repositories (Gin, Dispatch, NestJS, Django) it uses 12% to 42% of the incumbent tool's tokens, with higher recall on two and slightly lower on the other two.
Picks its ranking strategy per repository by turning the repo's own git history into evaluation questions, with no human labelling.
No LLM calls in the pipeline.
Python · SQLite · MCP · Embeddings









