SCOPE
Document scope
OpenAI-compatible providers, OpenRouter-routed providers, Groq/Cerebras-style inference providers, customer-configured provider keys, evaluators, and model routing metadata.
Status: first operational draft, pending legal review. It must be reviewed by counsel, the privacy owner, and the security owner before it is published as a binding policy or attached to a signed enterprise agreement.
§ 01
Data that may be transmitted
Prompt and task content: customer instructions, task descriptions, selected context, retrieval snippets, system/developer messages required for orchestration, and answer-format instructions.
Operational metadata: model name, provider name, routing profile, timeout, retry settings, temperature/top-p, token budget, and benchmark or workflow label when needed for execution.
Evaluation content: candidate outputs and verifier prompts may be transmitted to judge or evaluator models if the customer enables evaluation flows.
§ 02
Data minimization
AdaptOrch sends only the content needed for the configured route and avoids sending secrets, API keys, billing data, or unrelated tenant records to LLM providers.
Provider routing should prefer redaction, truncation, summarization, and retrieval-scoped context when full raw data is not required.
§ 03
Customer controls
Customers can choose supported providers, disable optional evaluation flows where available, configure provider keys, and request enterprise routing restrictions in a signed agreement.
Provider privacy, retention, abuse-monitoring, and training policies are controlled by the provider and should be reviewed before transmitting sensitive data.
CONTACT
Questions and updates
For support questions, contact ict03@rfems.com. For security reports, contact ict03@rfems.com. For privacy requests, contact ict03@rfems.com.