Every security review we go through opens with the same three questions. Rather than answer them one procurement form at a time, here are the answers in public.
What happens to your data
Your content is not used to train AI models. Not by us, and not by the model providers we route requests to. Chat and agent requests are pinned to zero-retention endpoints, and we verify that commitment per model before we make the model available in the platform.
That last part is worth spelling out. "The provider offers a no-retention option" is not the same as "the model you selected runs on it". We check at the level of the individual model endpoint, and a model that cannot meet the bar does not get added to the chat catalogue.
Two paths sit outside that pin, and we would rather name them than let you find out later. Text embeddings for data-store search run on a vendor endpoint without the zero-retention pin, and live voice calls go straight to the model vendor rather than through our router, because the extra hop costs audible delay on a phone call. Both are covered by the no-training commitment above. Our privacy policy lists every sub-processor.
How you verify it
Transparency is the reason the platform exists, so it applies to us too:
- Every run produces a full execution log. Inputs, model responses, condition results, tool calls, state changes and outcomes, timestamped.
- The model behind each step is recorded. You always know what processed a given piece of data.
- Logs can be exported. Your compliance team can review them outside the platform.
Deletion
You can delete workflows, sessions, data stores and records at any time. Account-level deletion requests are processed within 30 days. Details on retention and your rights as a data subject are in our privacy policy.
If your security review needs something more specific than this, write to us and we will answer it directly rather than pointing you at a document.