Bifrost provides deterministic static analysis for AI agents

Brock AI is developing Bifrost, a static analysis tool and MCP server designed to provide deterministic code analysis to AI agents. Unlike legacy tools like Coverity, which are slow and difficult to extend, Bifrost supports 13 languages and returns results in milliseconds via tool calls. The tool uses Git blob IDs as primary keys to enable instant re-indexing across multiple work trees. This allows agents to check for concurrency crimes or use-after-free errors during coding sessions rather than in long CI cycles. Brock AI also built Mjolnir, a meta-harness that drives multiple agents, including Codex, CloudCode, and KimiCode, over the ACP protocol. Mjolnir features a virtual workspace for sub-agents to optimize subscription usage. Additionally, the company is developing a benchmarking effort using live Bifrost tickets to evaluate models like DeepSeek Flash, GLM 5.3, and MuseSpark. Ellis notes that existing benchmarks like DeepSwee are saturated and no longer discriminative. Ellis positions Bifrost as a necessary infrastructure layer to bridge the gap between fast AI code generation and the need for rigorous, deterministic quality assurance. He emphasizes that agents should be able to run checks and receive answers in milliseconds rather than minutes or hours.