Git hosting without AI training
CodeTrunk is EU-hosted Git on Forgejo with one structural guarantee: code not used for AI training, by contract and by design. When teams search for git hosting no ai training, they usually meet two options: a host that trains by default and asks them to opt out, or a host with no AI feature in the repository path. CodeTrunk is the second. You keep pushing with standard Git, and your repositories stay in the European Union, with no model subprocessor to inspect.
Why this question matters now
Developers increasingly ask, in writing, what a Git host does with repository data before they commit code to it. The reason is policy drift: more than one major provider has changed its position on training defaults. Our AI training policy tracker lists the current, source-cited position of each major host, so a team can verify instead of trusting a marketing page.
For procurement and compliance reviews, a no-training claim now has to be checkable, not just promised. A customer or auditor asks whether repository data is used to train models, who the subprocessors are, and what the default is. A host that answers these in writing is in a different category from one that says "trust us".
To verify any no-training claim, ask four questions. Is there an AI feature in the repository path? Does the contract prohibit training, or only promise it? Which subprocessors receive the data? What can you switch off, and what is the default? A provider that answers all four in writing is in a different category from one that asks you to take its word for it.
What CodeTrunk does
CodeTrunk is EU-hosted Git hosting powered by open-source Forgejo. Your repositories, issues, and metadata stay on EU infrastructure: compute from netcup in Germany, backups on a Hetzner Storage Box in Germany, and payments via Mollie in the Netherlands. The full list is documented on the infrastructure page.
We do not train AI models on your data. Repository content, issues, and metadata are never used to train machine learning or AI systems, and private repositories are never accessed or processed for AI training by anyone. This is a contractual and technical prohibition: there are no data-sharing relationships with AI companies and no model ingestion pipeline. The subprocessors in the DPA are netcup, Hetzner, and Mollie, and no model provider appears in the list.
The guarantee also covers public repositories. We never train on or share public repository content for AI purposes, and we never grant third parties access to your content for AI training. We cannot prevent third parties from accessing what you have deliberately made public. That is true of any public repository on any host, and it is why the protection is strongest on private repositories, which are never accessed for training by anyone.
Because CodeTrunk runs standard Git on an open-source forge, your code stays portable. Every repository is a standard Git repository you can clone or mirror out at any time; nothing is stored in a proprietary format, and there is no exit cost. The DPA, which implements Art. 28 GDPR, is public and lists the same subprocessors. Read more about EU hosting.
Who it's for
CodeTrunk fits AI-first development teams of roughly 3 to 50 people, especially in the Netherlands and the DACH region, that want EU data residency and a no-training guarantee without running their own Forgejo or Gitea server. It suits regulated teams that must document where code lives and who processes it, and developers who want their Git host to behave like infrastructure rather than an AI product with a side of Git.
The practical benefit is what is absent: no AI setting to audit, no opt-out deadline to track, and no model provider to add to the DPA. For a regulated team, that removes an item from the security review instead of adding one.
Teams already on another host can move with a standard Git push: add the CodeTrunk remote, push, and you are running in the EU. Nothing in the workflow changes, and the no-training guarantee applies from the first commit.
How CodeTrunk compares
Every statement below about a third party is sourced from the AI training policy tracker (status as of 26 October 2026), which links each provider's official policy.
- GitHub trains on your code. Copilot is trained on public repository content under a license grant in the terms of service with no opt-out, and since April 2026 interaction data from Copilot Free and Pro users (including code from private repositories while it is used) trains models by default unless the user opts out. Business and Enterprise data is never used for training. (GitHub Terms of Service, §D.4 and §J.3.)
- GitLab requires opt-in consent. Its privacy statement says it will not train language models on AI inputs without instruction or prior consent, but GitLab Duo does send code and context to third-party model providers to generate responses. (GitLab privacy statement.)
- Codeberg does not train on your code. Its members formally resolved that the forge will not use project or user data to train AI; the same July 2026 vote also prohibits hosting projects that mostly consist of AI-generated code. (Codeberg: Protecting our FLOSS commons.)
- CodeTrunk does not train on your code, and has no AI feature in the repository path at all. AI-assisted or fully AI-generated code is welcome here. How you write your code is your business, and there is no content policy on how it was produced.
Training is not the same as processing. A model can process your code without training on it: paste a function into a chat window and the provider may send it to a model, log it, and return an answer even if it promises not to train on it. A no-training claim is only meaningful if you also know whether any model touches the code. CodeTrunk answers that structurally: no AI feature and no model endpoint in the repository path, and no training subprocessor.
If you prefer to run the software yourself, Forgejo and Gitea are open-source and have no AI features and no telemetry by default: nothing leaves your server, and there is nothing for a vendor to train on. CodeTrunk is the managed version of that idea: the same forge, operated for you in the EU, with the no-training position written down and subprocessors you can inspect.
Pricing and how to start
CodeTrunk starts free and stays EU-hosted on every plan:
- Free: €0, one private repository and 500 MB storage. No credit card required.
- Pro: €6.61 per month excluding VAT (€8.00 including 21% VAT), unlimited private repositories and 10 GB storage.
- Business: €14.88 per seat per month excluding VAT (€18.00 including 21% VAT), organisations, 30 GB storage, and a DPA included.
- Enterprise: custom, with SSO and audit logs.
Every plan, including Free, gets EU data residency and the same guarantee: we do not train AI on your data. See the pricing page for the full comparison and yearly billing. Switching from another platform is a standard Git import; compare CodeTrunk side by side before you move your repositories.
Frequently asked questions
Is my code used to train AI models?
No. Repository content, issues, and metadata are not used to train AI systems or provided to third parties for that purpose.
Does CodeTrunk offer AI coding features?
No. The service is Git hosting: repositories, issues, pull requests, code review, and account and billing. There is no AI feature in the repository path.
Does "no AI training" mean "no AI features"?
No, they are separate promises. A host can refuse to train on your code while still sending it to a model when you use an AI feature. CodeTrunk does neither.
Can I use GitHub Copilot or another AI assistant with CodeTrunk?
Yes. Those tools run in your editor and ask you to send context to their own service; CodeTrunk does not intercept that traffic or receive the prompt. Your Git host stays a Git host.
Where is the no-training commitment documented?
In the Terms of Service (§5), the Privacy Policy (§12), and the infrastructure documentation (§7). The DPA lists the same subprocessors and includes no model provider.
Keep exploring
Start with the plan that fits.
Create an account with no credit card. Upgrade when your repositories or team need more.
Start Building