Privacy and data retention

What we do not retain

• Prompt text, message content, or any model inputs
• Model outputs, completions, or reasoning traces
• Tool-call arguments or results

Request and response bodies exist in memory for the length of the request and nowhere else. Nothing is written to disk, logs or analytics.

What we do retain

Operational metadata only. Per request we keep a request ID, an API-key identifier, the model name, token counts, a timestamp, latency, HTTP status and which worker served it — for billing, abuse prevention and debugging.

None of it contains message content. We keep it 13 months for billing reconciliation, then delete it.

Account data, if you create an account

A key alone is enough to use the API. If you sign in to manage keys and credit, we also store your email address and a one-way hash of your sign-in provider's user id. We never store the raw id, so this database can't be used to look you up there. We store your balance and payment records too; card details go to Stripe and never reach us.

Keys are stored as a SHA-256 hash. Once a key is issued we can't see or recover it — not for you, and not for anyone who compels us.

We don't profile you, don't sell this data, and the only email we send is operational — a low balance, a paused top-up; receipts come from Stripe. To close your account, use Delete account on your account page — it takes effect immediately, no email required — or write to Feedback: hello@jetinfer.com. Either way the account record, your email and your keys go, and any remaining prepaid credit is forfeited with them, so spend it first. Billing records stay as long as accounting law requires. Prompts were never stored in the first place.

Training

We don't train, fine-tune or evaluate models on your data, and we grant no third party the right to.

Where it runs

Inference runs on rented GPU capacity in the EU. Your data is processed there and stays there. The model is open-weights (Qwen3.8-27B, Apache-2.0) quantized to int4, which is declared in /v1/models.

Security

Access is by bearer token over TLS, and keys are stored hashed. Rate limiting and abuse detection read metadata only, never the content of a request.

Data processing agreement

Need a DPA on file before you integrate? Ask at privacy@jetinfer.com and we'll send the current version, sub-processors included.

Contact

Privacy questions: privacy@jetinfer.com. Everything else: hello@jetinfer.com.