Millions of git repos.
One S3 bucket.
Open source git server built for scale — one repo per agent, per branch, per user. Fork in milliseconds, clone with standard git, nothing to provision.
A git server, reimagined on object storage
Everything your tooling expects, none of the infrastructure you don't.
Standard git
Clone, fetch, push and branch with the git CLI, libgit2, or any CI runner. No proprietary client, no migration tax.
Fork in milliseconds
Copy-on-write forks from any repo or remote. Branch a million times without copying a byte.
Backed by S3
Repos live directly on object storage — durable, cheap, and infinitely scalable. Nothing to provision.
Built for scale
Millions of repos: one per agent, per branch, per user. No rate limits, no capacity to plan ahead.
Why manifester
Object native
No disks, no stateful nodes. Storage is just your S3, GCS or R2 bucket — stateless servers in front.
Standard tooling
Speaks the native git protocol over HTTPS, so the tools your agents and pipelines already use just work.
Open source
MIT licensed. Self-host on your own bucket, or hand off a URL to any git client.
Self-host in minutes
Run the server on your own object storage, then point any git client at your instance — clone, push, fork. That's it.
# run manifester against your own S3 bucket
$ docker run -p 8080:8080 \
-e S3_BUCKET=my-repos ghcr.io/manifesterio/manifester
# clone, push and fork against your instance
$ git clone http://localhost:8080/acme/agent-42.git