Relace: Relace Search

Relace Search is an agentic codebase search model purpose-built for navigating large repositories and returning highly relevant files based on user queries. Unlike traditional retrieval approaches, it deploys 4 to 12 parallel tool calls—including view file and grep operations—to explore codebases with multi-step reasoning. The model is engineered to work as a specialized subagent within larger coding workflows, passing its findings to an orchestrating "oracle" agent that handles subsequent coding tasks. This design positions Relace Search as a precision component in agentic pipelines rather than a standalone assistant, with its architecture optimized for situations where pinpointing the right files quickly matters more than generating code directly.

The model delivers its results with claimed speed advantages—sources indicate it produces precise outcomes four times faster than frontier models on search tasks. Its large context window supports processing extensive repositories and maintaining conversation history across complex exploration sessions. Relace Search excels when integrated into agent harnesses that parse its responses and route findings to downstream coding agents, making it particularly valuable for development platforms, IDE extensions, and automated code review systems. The emphasis on agentic multi-step reasoning and tool use reflects a design philosophy centered on complementing, rather than replacing, human developers and their orchestrating AI counterparts.