Agent-scale Git Infrastructure.
Where agents work on code.

Agents will write orders of magnitude more code than humans. Where should that code live?

Agents should be able to provision a repository for each project, create a branch for every possibility, and read and write code at speed of chain-of-thought.

We aren't building a new version control system. Corigin is git, with a storage and compute stack rebuilt for agent scale.

Scalability

Your agents cannot start until they clone the code. They cannot share their work until they push it. In our latest benchmark, Corigin is was the fastest Git server we tested for agent workloads:

• •27.8M clones/hour
• •4.7M pushes/hour across 1,408 branches
• •220 ms median clone latency

Agent Context

Today, git stores the what, but not the why. Corigin captures each completed agent turn. The transcript and workspace changes are stored in the repository.

Inspect changes by agent turn or session, and correlate commits with the agent context that produced them. Push and fetch transcripts so agents can continue work across machines. Undo and redo agent actions with git commands.

Repository Sandboxes

Sandboxes can boot fast, but agents can't start until there's code. Spin up an agent environment for a branch with a single command. Don't lock your code into a sandbox provider's volume. Your code should live in a repository and be accessible anywhere.

Our Vision

Code will be the primary way agents reason about the world. Our goal is to make Corigin the best code storage and retrieval system for agents. Agents will use code to verify mathematical proofs or to synthesize new molecular compounds.

Code storage should also be able to answer which context was used, or why that code exists. Code is collaborative: we chat with our agents; we review work as a team. Agents should be able to tie those artifacts together in a verifiable way.