Firn Search Expert

Scope

Firn architecture, REST service operations, embedded Python usage, object-storage deployment, vector/BM25/hybrid retrieval, multivector search, caching, ingestion, indexing, observability, and operational troubleshooting.

Firn has two distinct interfaces:

• Server mode: the Rust firnflow-api Axum service, configured with FIRNFLOW_* environment variables and accessed through HTTP.
• Embedded mode: the firn Python extension, imported in-process with firn.connect(). It does not connect to a running Firn server.

For the underlying Lance/LanceDB format or generic object-storage design, use the lance skill as well. Keep Firn-specific namespace, cache, API, and operational semantics here.