1. What the AI does
CodeGene's work is done by AI agents: they read your site and instructions, plan and write pages and code, check the result, host it, watch it, answer your visitors' questions where you turn that on, and write reports. You see each change before it goes live unless you choose to let small changes publish on their own.
2. Which models
The agents use AI models from providers such as Anthropic, OpenAI and Google, reached through OpenRouter. The list changes as better models appear; the subprocessors page names the current ones.
3. What goes to the models
Only what a task needs: the pages, code, brand materials and instructions for that task, and, for a visitor-facing assistant you turn on, what the visitor types. Secrets and connection tokens are never sent to a model.
4. No training on your data
We do not use your content, your visitors' data or your conversations with CodeGene to train AI models. We send requests only to model providers that keep no copy of them, so they cannot train on what we send.
5. Mistakes
AI output can be wrong, incomplete or similar to work done for others. Our agents check their own work, and you review it before it goes live. You are responsible for what you publish, especially prices, legal text, health or financial claims and anything regulated. Every change can be undone.
6. Telling your visitors
When your site uses an AI assistant that talks to visitors, it must say that it is AI. Do not remove that label or present the assistant as a person.
7. What we will not do
We do not make decisions about people (credit, jobs, housing, insurance) with AI on your behalf, and you may not use CodeGene to do so.