How Sage AI works
Most vendors are vague about their AI plumbing. We publish ours: the provider, the data flow, the controls, and the audit trail. If you're going to trust AI with compliance work, you should be able to govern it.
Generation across all 13 AI agents runs on Anthropic Claude; semantic search embeddings and voice run on OpenAI. Both are disclosed below.
Where your data goes — exactly
Four hops, each one governed. This is the whole journey; there is no fifth step.
Your tenant
Your data lives in your RLS-isolated tenant. When you invoke an AI feature, only the minimum context needed for that task is assembled.
Row-level security · fail-closedCompliance Enablers platform
The platform enforces your AI settings, strips what doesn't need to leave, logs the action, and sends the request over TLS.
Policy enforcement · audit loggingModel APIs — disclosed per feature
Generation requests go to Anthropic Claude; search-embedding and voice requests go to OpenAI. Under both providers' commercial API terms, your data is not used to train models.
Two named subprocessors · no training on your dataBack to a human
The output lands in your workspace as a draft awaiting review. Accept, edit, or discard — the decision and the action are both on the audit trail.
Human approval · loggedFive commitments, in writing
Disclosed providers, per feature
Generation runs on Anthropic Claude; semantic search embeddings and voice run on OpenAI. Both are named subprocessors in our DPA, mapped to the exact features they power — not a vague "proprietary AI" you can't evaluate.
Never trained on your data
Your compliance data is not used to train AI models. This is a contractual commitment backed by the commercial API terms of both Anthropic and OpenAI.
Humans approve everything
AI output always lands as a draft for your review — a policy draft, a suggested answer, a proposed mapping. Never a silent write to your records.
Every action logged
Each AI action is recorded on the platform's hash-chained audit trail with the model used, the prompt class, and the approving human.
You control the switch
AI can be disabled per organization and per module. Want questionnaire AI but not policy drafting? That's a setting, not a support ticket.
13 agents, embedded in your workflows
Not a chatbot bolted on the side — AI inside the work itself, always drafting, never deciding.
Frequently Asked Questions
See the audit trail for yourself.
Ask us to show the AI action log, the per-module toggles, and a draft-approval flow in your demo. Transparency you can verify beats claims you have to trust.