Atomic Agent is a local-first, open-source AI agent runtime. By design, the agent loop, the model, and your data all run on your own machine. We built it this way because the most private architecture is the one where your data never has to leave your computer in the first place. This policy explains what that means in practice, and the narrow cases where data does travel.

1. Our Privacy Principles

Local by default. Your prompts, files, browser activity, shell commands, and agent memory stay on your device unless you deliberately configure otherwise.
No account required. You can download, install, and run Atomic Agent without registering, signing in, or giving us an email.
No hidden telemetry. The agent runtime does not phone home with your task content, prompts, or results.
Open to audit. The full source is public under the MIT license, so you can verify every claim in this policy in the code itself.

2. Data Processed Locally on Your Device

When you run Atomic Agent, it processes data entirely on your own hardware. This includes the instructions you give it, the contents of files it reads or writes, the pages it visits in your browser, the shell and git commands it runs, and the persistent memory it stores in a local SQLite database between sessions.

None of this is transmitted to AtomicBot. It lives in your local application data and working directories, under your control, and you can delete it at any time.

3. The Local Model

Atomic Agent serves an open-weight model locally through llama.cpp. Inference happens on your machine, so your prompts and the model's responses are not sent to any external inference provider. There is no token bill because there is no remote model call.

4. When Data Leaves Your Device

Data only leaves your machine when you direct it to. The main cases:

Tools you invoke. If you ask the agent to browse a website, call an API, or use a Model Context Protocol (MCP) tool, it will contact that destination — exactly as a browser or script would. Those third parties receive whatever the task requires and are governed by their own privacy policies.
Optional cloud models. If you deliberately configure a cloud model provider instead of the local one, your prompts are sent to that provider under their terms. This is off by default.
Remote drivers you enable. Optional integrations (for example, driving the agent from Telegram) route your instructions through that service if you turn them on.

5. The Website

This website (atomicagent.io) is separate from the agent software. When you visit it, we use privacy-respecting analytics (Google Analytics via Tag Manager, and Cloudflare Web Analytics) to understand aggregate traffic — pages viewed, referrers, approximate region. We do not sell this data or use it to build advertising profiles. The website does not require an account.

6. What We Do Not Collect

The content of your agent tasks, prompts, or results.
The files, code, or credentials the agent touches on your machine.
Your local agent memory.
Any personal account data — because there is no account.

7. Data Retention and Deletion

Because your agent data is stored locally, you control its lifecycle entirely. Uninstalling Atomic Agent and clearing its application data and any working files removes it. For aggregate website analytics, we retain only what our analytics providers store by default and hold no records that identify you personally.

8. Children's Privacy

Atomic Agent is a developer tool intended for adults and is not directed at children under 13. We do not knowingly collect personal information from children.

9. Changes to This Policy

We may update this policy as the software evolves. Material changes will be reflected on this page with a revised "Last updated" date. Because the runtime is open source, changes to how data is handled are also visible in the project's public history.

10. Contact Us

Questions about privacy? Email us at [email protected], or open an issue on GitHub.