How we handle your data.
This policy covers the Semurg hosted inference API, whether you call it directly or through a partner network such as OpenRouter. Software you run on your own hardware is covered by our privacy page.
Status: the hosted API is not open to customers. This policy takes effect when it opens and describes how the service handles your data from the first request.
No training, no stored memory
We do not train models on your prompts or completions, and we do not add them to any stored knowledge or memory feature.
What we log
- We do not write the text of prompts or completions to our logs.
- For each request we record its id, the model, the status, token counts, timings and the charge, to run and bill the service.
- For API requests, our web server log keeps the request path, status and time, without your IP address or request headers, for up to 14 days.
Prompt caching
To make repeated prompts faster and cheaper, a computed cache of the opening of a prompt is kept on the serving machines' disks for every customer, including traffic from partner networks such as OpenRouter, separately for each customer account, and encrypted with a key held only in memory. A partner network is one account: its calls are kept separate per end user when the network sends one, and otherwise share one scope within that network. Nothing is shared across networks or direct customers. It becomes unreadable after 10 minutes unused and after 60 minutes at most, when its key is destroyed.
Account and billing records
We keep your e-mail address, balance and per-request usage records for as long as your account exists. If you close your account and ask us to delete your data, we delete it by hand within 30 days, except billing records, which we keep for 7 years for tax purposes. Card payments are handled by Stripe; we never see your card number.
Where inference runs
Inference runs in the United States once the hosted API opens.
Contact
Questions about this policy: semurg@semurg.org
Last updated 17 November 2026.