Prerequisites
- An organization Owner for an organization key, or a workspace Admin for a workspace key. See BYOK prerequisites.
- Provider credentials with access to the model you want to use.
- For a hosted organization or workspace key, an HTTPS endpoint with a public hostname. Private addresses and localhost are refused.
Provider fields
Use OpenAI API credentials, an Azure resource with a model deployment, a Google service account JSON key, or a LiteLLM virtual key.
Azure uses its v1 API. You do not enter a dated API version. A deployment name identifies your Azure deployment, which can differ from the model’s published name.
Keep the Google JSON key unchanged. CloudThinker refuses a key that names a non-Google token service. A location must support your selected model; check Google’s model locations.
Connect your provider
A saved provider key connects the model list to the jobs you choose.
- Discover models using your key.
- Assign a model to a job.
- Run that job on your provider.
1
Choose the scope and provider
For the organization, open Admin Settings → BYOK. For one workspace, open Workspace settings → Workspace BYOK and choose Workspace.Click Add provider, choose the provider, and click Next. Enter an optional Name, then click Next.
2
Enter credentials and pick models
Fill in the provider fields above, then click Next. In Models, choose which jobs use your key.Select a model for at least one row. For Azure, enter the Deployment name. Other endpoint providers offer Or type a model name during setup.
3
Verify and activate
Click Verify and activate. CloudThinker checks your selected models before saving the key.Success state: the provider appears as Active. Open it to inspect its models and use Check key to test its saved credentials.
Choose an endpoint model
The saved key’s Models list shows what the provider exposes. Search for a model, then assign it in What runs on which credential and click Save. A listed model can run even when it is absent from CloudThinker’s built-in catalog. Such models use a default context of 128,000 tokens and a maximum output of 16,000 tokens. LiteLLM’s reported limits replace those defaults when available. Endpoint models are selected through BYOK assignments. In chat, choose the assigned mode, such as Pro; an endpoint model does not become a separate composer choice.Troubleshooting
The endpoint URL is refused
The endpoint URL is refused
Use HTTPS and a public hostname for hosted organization and workspace keys. Enter the endpoint’s final URL; redirects are not followed.
An Azure model cannot be found
An Azure model cannot be found
Check that the deployment exists on the same resource as your endpoint and key. Use its deployment name rather than assuming the public model name matches it.
No models appear
No models appear
Use Check key and read the displayed error. Confirm that the endpoint exposes a model list and that the key can use those models. For Azure, you can type a deployment name during setup.
Next steps
BYOK scopes and routing
Choose organization or workspace keys and assign each kind of work.
Usage and credits
See usage and charges when work uses platform models.