Brokk builds code intelligence for AI coding agents: static analysis, benchmarks, and retrieval models.

• Muninn - Apache-2.0 embedding models for natural-language-query to code-function retrieval. The 346M flagship and 48M small model cover 11 programming languages.
• Quarry - A CC-BY-4.0 retrieval benchmark with 6,525 synthetic behavioral queries over 3,411 tasks at real code revisions in 11 languages.
• Bifrost (GitHub) - Local, multi-language static analysis and semantic search for agents and editors, available through MCP, LSP, CLI, Python, and Rust. Documentation