Privacy
We never store prompts, and keep answers only when you ask. Here is exactly what we keep.
Your prompts, your images and the model's answers pass through bInference and are never written anywhere, except the files below that you ask us to keep. Prompt logging at our providers is off, except at Audio's voice provider, which keeps a history of the requests it serves.
What we keep for each call
- the model, the endpoint and whether it streamed
- token counts: in, out, cached and reasoning
- the reserve, the cost and the charge
- when it started, how long it took and how it ended
- the provider's id for the answer
That's what your Calls page shows, and what the public numbers on each agent's page are made from.
Files you ask us to keep
Files are stored only when you ask:
- An image you upload (
POST /uploadsorcreate_upload) is kept 24 hours. - Images made as links (
response_format: "url") are kept 7 days. - Files you attach in the AI chat (images, PDFs, Word, Excel, text and code), with the text your browser read from them, are kept 30 days, or until you remove them or delete their chat. Their names stay in your browser.
- Videos you make (at Videos or with
POST /videos), with a picture of their first frame and their settings, are kept 30 days, or until you delete them. Their prompts are not kept: on the site they stay in your browser. - Audio you make (at Audio), with its takes, the recordings a project uses and the voices you keep, is kept until you delete it, then 30 days in Recently deleted. Its exports are kept 7 days, and the record of each sound you ask for, with its words, 90 days. What the voice provider receives, and the consent kept for a cloned voice, are under Audio's privacy.
Then they're deleted. Each one is private to its agent or account, readable only through its own signed link, and nobody at bInference looks at them.
The cockpit
For an agent set up with npx binference agent-os, the hooks in its folder report what it does, never your prompts, its chat or its reasoning. We keep:
- each session's start, its app and version, and how many turns it ran
- each action: its kind, tokens and amounts, dollar value, order id, transaction hash, status, and the limit that blocked it
- the Agentic Wallet's address and settings, read at each session start: its limits and what it used today, its token and Developer Mode settings, and when its sign-in ends
- once you link the wallet: its BNB Chain transactions since the link
- for Telegram alerts: your chat's id and the name it shows
- for each setup code: the city and country it came from, shown when you approve it. Codes are deleted a day after they expire
A track record stays private until you make it public.
Keys
We store a key's first characters, to find it, and a fingerprint of the whole key. The key itself is shown once and never stored, so nobody at bInference can read it.
Your answers belong to the model's terms
Each call runs at the model's provider under its own terms. Some providers may keep data for abuse checks, a video model's provider keeps each video it makes for a short time after making it, and Audio's voice provider keeps a history of the requests it serves. If that matters for your use, pick models from providers whose terms suit you.