What it does

SeaGOAT runs a local server that indexes your codebase with vector embeddings, then answers plain-English queries like gt "function calc_.* that deals with taxes". It blends semantic search from ChromaDB with ripgrep’s regex muscle, and falls back to keyword matches while embeddings are still cooking.

The interesting bit

The server architecture is a deliberate trade-off: vector databases can’t yet process files on the fly, so SeaGOAT background-indexes gently (low CPU, long time) and stays queryable from minute one. It’s the rare “AI” tool that proudly does not phone home—no APIs, no ChatGPT, telemetry off by default.

Key highlights

• Local-only by design: ChromaDB + local embedding engine, works offline
• Hybrid results: semantic matches plus ripgrep regex/keyword fallback
• Query during indexing; accuracy warnings tell you how stale the embeddings are
• Supports a hardcoded baker’s dozen of languages (Python, TS/JS, Go, Java, C/C++, etc.)
• Optional remote server hosting, though security is your problem
• Respects .gitignore automatically; extra ignores via .seagoat.yml

Caveats

• macOS “partly tested,” Windows “help needed”—Linux is the happy path
• Language support is hardcoded, not pluggable
• No built-in security for remote server mode; roll your own VPN

Verdict

Worth a spin if you grep constantly but can never remember what your teammates named that helper function. Skip it if you need Windows reliability today or want to search languages outside the supported list.