1. Summary
ArtRuby does not use prompts or completions for model training. Prompt text, completion text, and submitted image content are not retained in persistent storage.
2. Inference content
ArtRuby processes text and, for supported multimodal models, image input to generate text output. Inference content exists only for the request lifecycle and the transient execution state needed to serve it. ArtRuby does not persistently log or store:
- prompt or message text;
- completion or reasoning text;
- submitted image content;
- system instructions or tool-call content;
- authorization credentials.
3. KV and prefix cache
To reduce repeated computation and improve latency, ArtRuby may maintain derived KV or prefix-cache state in volatile RAM or VRAM. Cache entries:
- are not written to persistent storage;
- are not used to train models;
- are not exposed for routine human review;
- are evicted according to memory and resource pressure;
- are cleared by process restart, resource reallocation, or normal lifecycle operations.
4. Operational metadata
ArtRuby retains limited, non-content operational metadata for up to 30 days. This supports metering, reliability, security, capacity planning, incident investigation, and reconciliation. It may include request IDs, timestamps, model identifiers, token counts, cache usage, performance measurements, routing information, status codes, and error categories.
Aggregated billing and accounting records that do not contain inference content may be retained for longer where required by law or commercial settlement obligations.
5. Training and human access
ArtRuby does not use customer prompts, completions, submitted images, or derived cache state to train, fine-tune, or evaluate models. Inference content is not available for routine human access. Operational access is restricted according to role and service need.
6. Processing regions
ArtRuby operates inference infrastructure across multiple regions worldwide. Request placement depends on model availability, endpoint configuration, capacity, node health, and applicable routing requirements. New regions may be introduced only after the relevant infrastructure and policies are operationally ready.
7. Contact
Questions about this Data Policy may be sent to support@artruby.co.