Relace builds AI infrastructure for autonomous coding workflows. The company develops specialized language models and tools optimized for code-related tasks - including semantic search, code generation, file merging, and codebase navigation. These models are co-optimized with Relace's infrastructure to achieve high throughput: code edits at 10,000 tokens per second and codebase searches completed in under two seconds. The models power products at Lovable, Magic Patterns, Figma, and Vercel.
The company's approach addresses limitations of general-purpose large language models for autonomous coding. Rather than rely on models designed for broad tasks, Relace builds smaller, faster models trained specifically for code workflows and the infrastructure to support them at scale. This co-optimization strategy prioritizes performance and reliability for production environments.
Relace recently launched Relace Repos, a source control system designed for AI agents. The system features lightweight push and pull operations, no rate limits, and automatic codebase indexing to enable rapid retrieval. The product reflects the company's focus on building complete infrastructure for agent-driven development workflows. Relace is based in San Francisco and is a Y Combinator alum that raised $23m in Series A funding in late 2025, led by Andreessen Horowitz with participation from Matrix Partners.