Account
Privacy lanes.
Every model carries a label that says who can read what you send it. Pick the lane that fits the data.
The three lanes
| Lane | Who handles the prompt | Good for |
|---|---|---|
| Free | A provider offering the model at no cost. It may log prompts and use them to improve its service. | Testing, public data, casual use. |
| No training | Open-weight models served only by providers that do not store or train on prompts. Synaps enforces this on every paid request to these models. | Work data, code, drafts you would rather keep private. |
| Provider | The model's maker (Anthropic, OpenAI, Google, xAI) under its own API data policy. | When you need that specific model and accept the maker's policy. |
The lane appears on every model in the chat, on the Models page, and as privacy_lane in GET /models.
What Synaps itself keeps
- Not stored: prompts and replies sent through the API, and chats from signed-out use or with sync turned off. They pass through to the model and are not written to our database. Attached file contents and generated images are never stored.
- Synced (signed in, sync on): the text of your chats, so they follow your wallet across devices. Only your wallet can read them; turn sync off in the chat's Instructions & settings.
- Stored: for each request, the time, model, token counts and amount charged, so billing and your activity list work.
The full details are in the Privacy Policy.
Choosing by data type
- Public or test content: any lane, including free.
- Business documents, source code: No training lane, or a Provider model whose maker's API terms you have read.
- Personal data about others: only if you have the right to process it, in the No training lane, and with as little detail as the task needs.