AI Use Policy

No training on your data

We do not train any model on your prompts, code, files, or model output. Period. This applies to every plan and every model in the catalog.

Where the underlying model provider has its own data-use terms (for example, the major US labs in our managed catalog), we use their zero-retention / no-training endpoints wherever the provider offers them. Provider-side terms remain governed by the provider; we link the most relevant provider policies in the Sub-processors page.

No server-side chat retention

Chat content is not retained on the server. Requests stream through the backend and the response stream is the only thing the server emits — nothing is written to a database. Chat history that you see in the Desktop app is stored locally on your device.

What we do log (for the run, not for you)

• Aggregate token counts and approximate cost per run, so we can bill correctly and surface usage in your dashboard.
• Operational metrics: latency, error rate per provider, rate-limit headroom.
• Audit entries when an admin acts on your account (impersonation is blocked from the MCP surface; every admin write is logged).

These logs do not include the content of your prompts or model output.

Output ownership

You own the output that S.A.M.I. produces from your prompts, to the maximum extent the underlying model provider permits in its terms. You are responsible for using the output: review it, test it, and do not ship code, claims, or content that you have not validated.

Hallucination and accuracy disclaimer

Large language models can produce plausible-looking but incorrect content. The multi-agent consensus design reduces — but does not eliminate — that risk. Treat every output as a draft that needs review, especially security-critical, legal, medical, or financial output. Do not rely on S.A.M.I. as the sole source of truth for decisions that matter.

Prohibited use

Some uses are prohibited regardless of provider terms — e.g. generating malware, CSAM, deepfakes targeting real people, mass-scale spam or harassment content. The full list is in the Acceptable Use Policy.

High-risk domains

S.A.M.I. is built for software work. It is not certified for medical diagnosis, legal advice, financial advice, autonomous safety-critical control, or any use regulated by the EU AI Act's high-risk categories. If you build a product that targets one of those categories, the responsibility for compliance, certification, and human oversight remains with you.

Questions

Email contact@sami-agent.com for anything related to AI handling, model selection, or data flow.