Coregit — serverless Git API for AI agents. Atomic multi-file commits, instant forks, time-travel snapshots. 3.6× faster than GitHub.
New3.6x faster than GitHub — ROI benchmarks published

Store & version
your AI agent's
code

Coregit gives your AI agents a versioned filesystem. Atomic commits, instant forks, time-travel snapshots — one REST API, built on git. No lock-in, no infra to manage.

Get your API keyRead the docs

No lock-in. git clone your data anytime, run anywhere.

Pay-per-useStandard Git protocol3.6x faster than GitHub (100 files)REST + Git HTTP15K req/hr (GitHub: 500)

Trusted by startups

“Never felt this easy to use versioning at scale. Fast setup and easy to use.”

Sameer

Prettiflow

“While working with Prettiflow, we needed infrastructure that could handle agent-driven commits at scale. Coregit fit perfectly. Managing multiple changes in a single flow became seamless, making our entire workflow faster and more reliable.”

Shyam

Prettiflow

“The quality and speed is peak. Really polished to every detail.”

Amit Dey

CEO, Prettiflow

Prettiflow logo — Coregit customerPrettiflow

Product

One API for everything
your agent needs.

Response200 OK
response.json

Benchmarks

Built for fleets of agent repositories.

Per-repository throughput, per-repository isolation, per-repository storage. Tested against major Git hosting providers on 100-file atomic commits — and engineered to scale across thousands of repos per organization.

3.6x

Faster

100-file commit: 19.8s vs 72.1s

394x

More write throughput

15,000 commits/hr vs 38

1 call

vs 105 API calls

Atomic multi-file commit

GitHub API (100 files)72,064 ms / 105 calls
Coregit (100 files)19,769 ms / 1 call
View full benchmark methodology

Numbers

Coregit by the numbers.

1000
files per atomic commitOne API call. GitHub: one file per call.
15,000
API requests per hourGitHub authenticated: 5,000. GitHub search: 500.
3.6×
faster than GitHubMedian throughput on April 2026 benchmarks.
19.8s
for a 100-file commitGitHub takes 72 seconds and 105 sequential API calls.
0
infrastructure to runFully managed. No servers, no DevOps, no infrastructure to run.

Features

The most advanced filesystem
for AI at scale.

Infrastructure

Stop managing servers.

Running Gitea or GitLab yourself means Docker, backups, SSL, disk monitoring, and zero AI features. Coregit gives you all of that as API calls.

No servers to manage.

Serverless at the edge. No Docker, no Kubernetes, no VMs, no disk provisioning. We handle deployments, scaling, and uptime — you call the API.

Semantic search built-in.

Embeddings, vector retrieval, and a structural code graph — first-class. Search by meaning, not just keywords. No Elasticsearch, no plugins — it's part of the API.

$0 when idle.

A self-managed Gitea instance on a $50/mo VPS costs $50 even with zero repos. Coregit costs $0 until your agents start committing. Scale to thousands without capacity planning.

Encrypted, isolated, audited.

AES-256 at rest, TLS 1.3 in transit. Per-org isolation — separate storage paths, scoped keys, SHA-256 hashed credentials. Immutable audit log for every write.

Security

Secure, auditable,
and ridiculously fast

AES-256 encryption. Per-org isolation. Immutable audit logging.

Built for trust

Purpose-built Git engine with AES-256 encryption, per-org data isolation, and immutable audit logging on every write.

Security docs

Blazingly fast. 3.6x faster than GitHub at 100 files. 5-layer edge caching. 2ms writes with Sessions.

See benchmarks

Pricing

Top up once. Unlock for life.

Lifetime top-ups unlock higher caps. No subscriptions, no seat licenses. Metered usage runs on top of any tier, deducted from your balance.

Free

Prototyping and side projects

$0

Forever free

Get started
3 repositories
50 MB per repo
60 requests / minute
100 API calls / day
$5 signup credit
LLM Wiki not included

Starter

Solo builders shipping their first AI product

$50

Total top-up to unlock

Top up to start
100 repositories
500 MB per repo
300 requests / minute
Unlimited API calls
LLM Wiki — 20 runs / day, 1M tokens / month
All features unlocked

Growth

Teams running real AI workloads

$200

Total top-up to unlock

Top up to unlock
1,000 repositories
2 GB per repo
1,000 requests / minute
Unlimited API calls
LLM Wiki — 200 runs / day, 10M tokens / month
Priority email support

Scale

Platforms with thousands of agents

$1,000

Total top-up to unlock

Top up to unlock
Unlimited repositories
10 GB per repo
3,000 requests / minute
Unlimited API calls
LLM Wiki — unlimited
Direct founder support

Pay-per-use, on any tier

Deducted from your wallet balance. Same rates whether you're Free or Scale.

API calls$0.10 / 1,000 calls
Storage$0.10 / GB / month
Git transfer$0.10 / GB
Lazy-edit & agentic search$1.32 / M output tokens
LLM Wiki ingest run$0.01 / run
LLM Wiki tokens$1.00 / M tokens

Need SOC 2, SSO, custom rate limits, or spending caps? Email us.

FAQ

Common questions

What is the best Git API for AI agents?
Coregit is purpose-built for AI agents: a serverless Git API with atomic multi-file commits up to 1000 files in a single call, 15,000 requests per hour, and 3.6× higher throughput than GitHub on the same workload. Standard Git Smart HTTP works for clone and push, while the REST API exposes commits, snapshots, branching and semantic search.
What is a serverless Git API?
A serverless Git API is a hosted Git backend you call over HTTPS without running git, ssh keys, runners or self-hosted servers. Coregit accepts atomic multi-file commits, branches, diffs and clones via REST and Git Smart HTTP. There is nothing to deploy — agents send JSON, Coregit persists immutable Git objects in distributed, encrypted storage.
How is Coregit different from GitHub for AI agents?
GitHub caps multi-file commits at ~38 per hour. Coregit handles 15,000 through the same endpoint. A 100-file commit on GitHub takes 105 sequential API calls and 72 seconds; on Coregit it is 1 call and 19.8 seconds. GitHub's Terms of Service also explicitly warn against reselling programmatic access — Coregit is licensed for agent-driven workloads.
Why are GitHub rate limits a problem for AI agents?
Authenticated GitHub clients get 5,000 requests per hour and 500 search calls per hour. A single coding agent producing a 100-file change can exhaust 100+ contents API calls in seconds. Once throttled, the agent stalls for the full reset window. Coregit's 15,000 req/hr ceiling and 1000-file atomic commit collapse that into one request.
Can ChatGPT, Claude or Cursor write commits via the Coregit API?
Yes. Any HTTP-capable client works: ChatGPT with custom GPT actions, Claude with tool use, Cursor with MCP, OpenAI Assistants, LangChain, Mastra, Vercel AI SDK. Use the @coregit/sdk TypeScript package or plain fetch with an API key. A commit is one POST with a list of {path, content} entries.
Do I need this if I'm already using a sandbox?
Yes. A sandbox gives your agent compute — Coregit gives it memory. When the sandbox dies, Coregit keeps every commit. You can fork a repo per session so parallel agents never conflict, roll back any run in one API call, and give humans a diff to review before changes go live. Compute is ephemeral. Your files shouldn't be.
Can Coregit store agent memory and knowledge bases?
Yes. Coregit's LLM Wiki API turns repositories into versioned, AI-synthesised knowledge bases: upload documents, agents extract entities, concepts and decisions into a knowledge graph, every edit is a Git commit. Use it for long-term agent memory, research notes, project wikis or compliance-grade audit trails.
Is Coregit only for code?
No. Code, markdown docs, JSON configs, images, PDFs, research notes, knowledge bases, spreadsheets — any file type your agent produces. Coregit stores binary and text files identically. If your agent writes it, Coregit versions it.
Can't I just write agent output to S3?
S3 has no branching, no commit history, and no protection against concurrent write conflicts. Two agents writing to the same key at the same time will silently overwrite each other. Coregit gives you atomic commits, instant forks, full diff and merge, and a queryable history — built on git, so you can clone your data out anytime.
Can I use standard git clone, push and pull?
Yes. Every Coregit repo is accessible via Git Smart HTTP. Use any standard Git client to clone, push, and pull. API keys authenticate via HTTP Basic auth. Your data is portable — git clone any repo and run anywhere.
Do I need the git CLI to use Coregit?
No. The entire API is HTTP-based. Use curl, fetch or any HTTP client. We also publish an official TypeScript SDK, @coregit/sdk, for typed access to every endpoint. Git CLI compatibility is provided as a fallback, not a requirement.
How does Coregit protect my data?
All data is encrypted at rest (AES-256) and in transit (TLS 1.3). API keys are stored as SHA-256 hashes with timing-safe comparison. Every organization's data is fully isolated — separate storage paths, scoped database queries, org-bound API keys. All write operations are recorded in an immutable audit log. Read our security documentation →
Can I run Coregit on my own infrastructure?
Coregit is a fully managed, proprietary platform. For enterprise customers with specific deployment requirements, we offer dedicated instances with custom SLAs, data residency guarantees, and private networking. Contact us at team@coregit.dev to discuss your needs.
How is Coregit priced?
Free forever for prototypes: 5 repos, 1 GB storage, 10K API calls per month and 5 GB Git transfer. Beyond that, pay-as-you-go: $0.10 per GB stored per month, $0.50 per 1,000 API calls, $0.10 per GB Git transfer. No seat licenses, no commitments. Enterprise plans add SSO, dedicated support and volume discounts.

Start shipping with Coregit.

Free tier. No credit card. Get your API key in 30 seconds.

Get your API keyRead the docs