Model access controls
Choose which models each API key can call, so a key only reaches the models it is meant for.
What is Model access controls?
Model access controls decide which models an API key is allowed to call. A key can be set to all models, or restricted to an allowlist of model slugs. On Spro the allowlist is part of the key, so a key for a cheap support bot cannot call the most expensive models, and a request for a model outside the list is refused.
Not every app needs every model. Restricting keys limits cost and risk: a key that only needs a small model cannot run up the bill on a large one, and a compliance review can see which models each workload can reach.
The allowlist is set on the key, so it follows the workload wherever the key is used.
Key features
What this feature does for your requests.
- Each key can allow all models or a chosen list
- Requests for models outside the list are refused
- The list is stored on the key, not in your app
- Changing the list takes effect on the key's next request
How it works
4 steps, from setup to the first request.
- 01
Choose the key
Pick the key for one app or workload.
- 02
Set the allowlist
Select the models that key may call, or leave it open.
- 03
Requests are checked
A request for a model outside the list is refused.
- 04
Review as needs change
Update the list when the app's needs change.
Is it the right fit?
Specific situations where this feature helps, and where it does not.
Good fit if
- A cheap support bot should never reach the largest, most expensive model
- A compliance review needs to show which models each workload can use
- Experimental keys should be limited to the models under test
Not a good fit if
- An app that legitimately needs every model, where the open setting is right
- Restricting access per person. Model access is set per key, and roles cover people
- You need the model chosen automatically per request. Use automatic model selection for that
Model access controls FAQs
Common questions about Model access controls.
Route every AI request through SPRO.
One OpenAI-compatible API for 350+ models, with intelligent routing, 29 built-in guardrails, observability, and automatic failover.
