Sub-processor disclosure
Last updated 24 July 2026 · effective on that date
How routing works
A plan is a sequence of steps, and each step goes to one approved provider. A provider receives the prompt and the assets that step needs, and nothing else: not your thread, not your other assets, not your spend. Judgement models never generate media, and media models never see the run record.
Pinning and objection
If a provider is unacceptable to you, pin an engine per chat or per role in settings and KIT ENGINE will not route around it. Where a pin makes a request impossible, KIT ENGINE says so before you spend rather than substituting silently.