Bound the suite. Keep the provider key.
A model eval, benchmark, or vendor bake-off can burn through a standing provider secret. Issue a Till key for that suite with a required request ceiling so remaining capacity stays readable and the run can stop.
Isolate the eval traffic, not the provider account.
-
01
Issue a disposable Till key for the suite
Set the request ceiling from the eval plan, then add optional token, estimated-spend, expiry, and IP bounds when the harness needs a tighter envelope.
-
02
The provider credential stays behind Till
The harness uses a Till key, not an unconstrained provider secret in the eval runner. When the suite is retired, revoke or let expiry stop that key without rotating every connection.
-
03
This is not a model scorer or procurement tool
Till does not grade quality, certify vendors, or replace your eval harness. It is a hosted control layer in front of 12 AI providers: a required request ceiling and optional token, spend, expiry, and IP controls.
-
04
Pro: $59/month, 25 scoped keys, 25,000 monthly activations
Those are plan limits. Public signup is closed; request Pro beta access. Provider inference is billed by the provider. Till charges a platform subscription.
Connect provider accounts once. Each evaluation run gets a disposable Till key with a required request ceiling and optional token, estimated-spend, expiry, and IP controls. Till is not a zero-knowledge vault, a model-quality judge, or a replacement for provider billing or IAM.
These are plan limits, not usage claims.
Pro
Scale
Public anonymous signup is closed. New accounts are approved and provisioned manually. Existing customers can open the dashboard. Compare the full ledger on pricing.
Run the eval. End the key.
Request Pro beta access. Till is a controlled beta; accounts are onboarded manually. Do not start from self-serve checkout.