Coregit Launches Serverless Git API Built for AI Agents, 3.6x Faster Than GitHub

Summary

Two developers from Kazakhstan have unveiled Coregit, a ground-up reimplementation of Git's object model designed specifically for AI coding agents. Unlike GitHub's Content API—which was built for human developers—Coregit dramatically reduces the number of API calls needed for version control operations, committing 1,000 files in a single call compared to GitHub's requirement of 105+ sequential calls. The platform achieves 3.6x faster performance, handles 15,000 requests per hour versus GitHub's 500, and runs entirely on Cloudflare Workers with zero server infrastructure to manage.

Coregit's architecture is built from scratch in TypeScript without relying on git binaries or libgit2, enabling sub-millisecond cold starts and global edge deployment. The service maintains full Git compatibility—users can clone Coregit repos, push to them, and perform standard Git operations because the underlying data format is identical to native Git's object model. By leveraging Durable Objects for rapid acknowledgment and background R2 writes, Coregit delivers commit confirmations in approximately 2 milliseconds while durability operations complete asynchronously, ensuring responsive performance for AI agents that need to persist code changes at scale.

• Uses fire-and-forget architecture with Durable Objects and background R2 writes to deliver 2ms commit confirmations

Editorial Opinion

Coregit represents a thoughtful redesign of Git hosting for the AI-native era. Rather than treating AI agents as power users of human-centric APIs, the team rethought the fundamental architecture—and the results are compelling. For teams building coding agents or AI-driven development tools, Coregit's single-call bulk commits and serverless execution could become essential infrastructure. However, adoption will depend on ecosystem maturity and whether developers trust a new Git implementation with their code repositories.