Overview
This notice explains how OpsRabbit prepares model context, uses configured model endpoints and protects customer information during AI-assisted operations. It also states the limits of model output and the controls required before an action is permitted.
The notice covers permitted data use, model selection, inference flow, retention, human oversight and material changes.
No customer-data training
OpsRabbit provider does not use customer prompts, outputs, tickets, logs, documents or operational data to train or fine-tune shared foundation models. Any materially different use requires an explicit written agreement, purpose definition and applicable privacy and security review.
Model selection and provenance
- Approved deployments record the model provider, model or version, hosting arrangement, intended purpose and material limitations.
- OpsRabbit does not represent that it knows the complete provenance of a third-party foundation model’s original training corpus.
- The customer approves the endpoint and applicable provider terms when the endpoint is customer-selected.
Inference data flow
OpsRabbit retrieves authorized operational signals, assembles relevant evidence and instructions, sends only the context required for the approved task to the configured model endpoint, and returns evidence-backed output. Raw operational data is not intended to be persistently copied outside the configured deployment solely for model training.
Human authority and limitations
- Model outputs are recommendations or analysis unless a separately authorized workflow permits an action.
- High-impact, write or destructive actions require explicit authorization and appropriate approval gates.
- Users must validate outputs against source evidence; AI output may be incomplete or incorrect.
- Material model, prompt, tool or data-flow changes require risk review and regression testing.
AI system description
OpsRabbit uses generative and analytical models to correlate operational signals, explain evidence, propose likely causes and recommend investigation or remediation steps. The approved deployment record identifies the model provider, model/version where available, endpoint location, enabled tools, prompt or skill version, retention setting and accountable owner. Model output is probabilistic and must be evaluated in context.
Permitted and prohibited data use
Customer data may be processed only to deliver, secure and support the authorized OpsRabbit service. OpsRabbit provider does not sell customer data, use it for targeted advertising, combine it into a cross-customer training corpus, or train/fine-tune shared foundation models on it. Product analytics must be aggregated or de-identified and must not reconstruct customer content. Any customer-specific fine-tuning requires a separate written scope, lawful basis, data specification, evaluation plan and deletion terms.
Model input and output controls
- Retrieve information using customer-approved identities and scopes.
- Separate trusted product instructions from untrusted tickets, logs, documents and tool output.
- Minimize context and exclude credentials, secrets and unrelated personal data.
- Apply provider retention and training settings approved for the deployment.
- Cite or link supporting operational evidence where the product supports it.
- Label limitations and require human validation for material conclusions.
Model and provider changes
A material change includes a new provider, model family, hosting region, retention behavior, training use, high-impact tool, or significant prompt/agent change. Such changes require impact assessment, security and privacy review, representative regression tests, approval, release records and customer notice where the agreement requires it.
Transparency evidence
Available evidence may include the approved model register, data-flow description, shared-responsibility matrix, evaluation results, change records and incident records. Provider proprietary training-corpus details may be unavailable; OpsRabbit will not imply knowledge or rights it does not possess.