JETINFERPrivacy and data retention
Effective 2026-08-21 · applies to the inference API at api.jetinfer.com
We never store your prompts or the model's outputs. Inference is zero-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 only in memory for the duration of the request. They are never written to disk, logs, or analytics systems.
What we do retain
Operational metadata only. For billing, abuse prevention, and debugging we keep, per request: request ID, API-key identifier, model name, token counts (prompt and completion), timestamp, latency measurements, HTTP status, and the identifier of the worker that served it.
This metadata contains no message content. It is retained for 13 months for billing reconciliation, then deleted.
Training
We do not train, fine-tune, or evaluate models on your data. Ever.
Infrastructure
Inference runs on a GPU server rented from RunPod, in their EU-CZ-1 region
(Prague, Czech Republic). Model weights are public
open-weights (Qwen3.8-27B, Apache-2.0) quantized to int4; the quantization level is
disclosed in our /v1/models metadata rather than left for you to
infer. Because request content is never persisted, no customer content resides on
these hosts after a request completes.
Security
API access requires bearer-token authentication over TLS. Keys are stored hashed. Rate limits and abuse detection operate on metadata only — never on the content of your requests.
Data processing agreement
A DPA is available on request for customers who need one on file before integrating. Ask at [email protected] and we will send the current version.
Contact
Privacy questions: [email protected]. Everything else: [email protected].