Privacy

• Your account email and a hashed password.
• Metadata about each request: the model, token counts, latency, cost, status, and timestamps.
• Records of operators and their nodes: a hardware summary, reputation, and earnings.

The content of your prompts and the model's responses is never written to our database and never written to our logs. It exists only in memory while a request is being served, and is discarded when the request finishes. Our request records hold counts and timings, not text.

Your prompt is streamed through the orchestrator to an operator's node, processed there in memory by an open-weight model, and streamed back to you. Operators never learn who you are: every request reaches a node with no identity attached.

An operator runs the model on hardware they control, so while your request is being served its content passes through that machine's memory. We protect that today with four things: operators never know your identity, content is never logged, the network (not the operator) controls routing and can ban bad actors, and operators are scored and audited. What that does not give you is a cryptographic guarantee against a determined operator inspecting their own machine.

Two stronger options are on the roadmap. The confidential tier will route your request only to GPUs with hardware confidential-computing attestation, so the host cannot read what it processes. And if you run your own private node, your prompt never leaves your hardware at all, which is the strongest privacy there is.

• We do not sell your data.
• We do not train models on your prompts or responses.
• We do not reveal your identity to operators, or operators' identities to you.

We use a single session cookie to keep you signed in. No third-party tracking or advertising cookies.