Terms of service

Last updated 2026-10-19

These terms govern your use of the hosted Atoma service at atoma.run: the website, the web app, the MCP server at https://atoma.run/mcp and the Atoma plugins for Claude and Codex, which connect to that server. By signing in or using the service, you agree to them. If you use it on behalf of an organisation, you confirm that you may accept them for it.

1. Who provides the service

The service is provided by Matthieu Foillard, a sole proprietor established in Greece, Pavlou Mela, 1, 15233 Athens, Greece ("we", "us"). Contact: mgf@atoma.run.

2. The service

Atoma takes a request, plans the work, has AI agents carry it out in isolated workspaces, checks the result against the request, and delivers it. A run may deliver, stop incomplete or fail. The service is in beta: features, limits and models change, and it can be interrupted for maintenance or deployment.

Atoma is under active development and has not been independently audited. Use it to evaluate the product and for work that contains no confidential, personal or otherwise sensitive data.

3. Accounts and organisations

• You sign in with a GitHub, Google or ChatGPT account and must be at least 18 years old.
• Your first sign-in creates an organisation that you own. Owners and admins invite members and assign roles, and they are responsible for who they admit and for their organisation's use of the service.
• Keep your API tokens and other credentials secret. You are responsible for activity carried out with them until you revoke them.

4. Shared learning

Atoma is a shared-learning service, and this is part of what you agree to:

• Every run can use the platform's catalogue of agents, learned recipes and trust history. What a run learns can update that catalogue and history for later runs, including runs by other organisations. Contributions are reused without an individual approval step, and learned recipes may include executable code.
• By submitting material to the service, you authorise us to process it to perform your runs, and to extract, store, adapt and reuse learned recipes within the service, including running them for other organisations. This does not transfer ownership of your material, and it does not give other organisations access to your projects.
• Projects, workspaces, run traces and retrieval corpora stay within their organisation. The catalogue, recipes and trust history are shared across the service. Learning does not guarantee that every detail of its inputs is removed, so do not submit material that must never influence a shared recipe.
• Platform administrators can read across organisation boundaries for operation and support. These reads are recorded, and organisation owners are notified.

5. Decisions taken by TypeSafe

Every run uses Decision, TypeSafe's model (typesafe.ai), at our cost, for bounded decisions: which agent or recipe takes a step, whether a plan or result can be approved without a model review, and whether a newly learned recipe duplicates an existing one. For each decision we send TypeSafe its content: the step's task text and constraints, the candidates' descriptions, and the plan or result under review with excerpts of its evidence. TypeSafe processes it under its own terms and may retain it. A decision that TypeSafe does not take is taken by the run's model, and final acceptance of a delivery never rests with TypeSafe.

6. Your content and the results

• Submit only material that you have the rights and permissions to have processed and reused as described here.
• As between you and us, you keep your rights in the material you submit and in the results delivered to you, subject to the rights of third parties and to the licences of any material the results include. Using Atoma does not by itself place results under Atoma's software licence.
• Results and learned recipes are generated by AI and can be wrong, incomplete, insecure or unsuitable. Neither the checks a run performs nor a recipe's trust history guarantee correctness. Review and test results before you rely on, distribute or deploy them.
• A result is published to GitHub only after you accept it. Hosted previews and applications are provided for evaluation.

7. Acceptable use

You must not:

• try to access another organisation's data, or bypass the sandbox, container isolation, network restrictions, rate limits or other execution boundaries;
• misuse credentials, or use the service with another person's account or credentials without their permission;
• use the service to break the law, infringe others' rights, create malware, attack or scan systems you are not authorised to test, send spam, or produce content that exploits or harms children;
• submit personal data about other people without a lawful basis, or special categories of personal data;
• overload or disrupt the service, or resell access to it without our written agreement.

Report vulnerabilities privately, as described on the support page, instead of exploiting them.

8. Models, credentials and costs

• The service is free of charge during the beta. We do not bill you, and we do not meter or rebill model providers' charges. If paid plans are introduced, they will be offered under separate terms and never applied without your agreement.
• Runs use the models configured for your organisation. When you supply a provider API key or connect a personal subscription, that use is subject to the provider's own terms and is charged by the provider to you. You are responsible for being entitled to use it this way. Some processing, such as run titles and TypeSafe's decisions, is at our cost.
• Limits apply, such as concurrent runs, run duration and the assistant's daily allowance. These are admission limits, not spending caps.

9. Retention

A finished run's trace and workspace become eligible for deletion 90 days after it ends, unless they are still needed (for example, as the base of later work or a publication that is not finished). Eligibility is not an exact deletion date. Run records, payer attribution and the history of what was learned are kept separately, and removing a run's files does not automatically withdraw what was learned from it. The privacy policy gives the details, and your rights over personal data are not affected.

10. Suspension and ending

You can stop using the service at any time, and ask us to delete your account or organisation by writing to mgf@atoma.run. We may suspend or end access that breaches these terms or puts the service or other users at risk, and we may change or discontinue the service. We will give reasonable notice where we can, so that you can retrieve your results.

11. No warranty

The service is provided free of charge, "as is" and "as available", without warranties of any kind, to the extent the law permits. We do not guarantee that it will be uninterrupted, secure or free of errors, or that any result will be correct or fit for a particular purpose.

12. Liability

To the extent the law permits, we are not liable for indirect or consequential loss, for loss of data, profit or business, or for the consequences of relying on, distributing or deploying results. This does not limit liability for intent or gross negligence, for death or personal injury, or any other liability that cannot be limited by law, and it does not affect the mandatory rights of consumers.

13. The open-source software

Atoma's software is published under the GNU Affero General Public License v3.0 (github.com/atoma-run/atoma). These terms cover the hosted service only. They do not restrict the rights that licence grants you, or replace the licences of third-party material.

14. Law and disputes

These terms are governed by Greek law, and the courts of Athens have jurisdiction. If you are a consumer, you keep the protection of the mandatory rules of the country where you live, and you may also bring proceedings there.

15. Changes

We may update these terms. The date at the top shows the current version, and we will announce significant changes in the app before they take effect. If you keep using the service after a change takes effect, you accept the updated terms.