SETUP GUIDES · AI MODELS
Azure OpenAI endpoint and key
Use OpenAI models that run inside your own Microsoft Azure subscription.
Some companies must keep AI traffic inside their Azure account for compliance. Azure OpenAI gives you OpenAI models there. You create a resource, deploy a model under a name you choose, and give Claspine the resource’s address and key.
Words you’ll see
- API key
- A long secret string that lets software use your account, like a password made for programs. Anyone who has it can spend on your account, so treat it like a password.
- Endpoint
- The web address Claspine uses to reach your storage provider, such as https://s3.us-east-1.amazonaws.com.
- Model
- The specific AI you call (for example a fast, cheap one or a larger, smarter one). Claspine’s AI Gateway lets you choose models and fall back to another if one fails.
Before you start
- An Azure subscription with access to Azure OpenAI.
Steps
- In the Azure portal, create (or open) an Azure OpenAI resource.
- Open the resource in Azure AI Foundry and deploy a model. Write down the deployment name you choose; this is the name Claspine will use as the model.
- Back in the portal, open the resource’s Keys and Endpoint page.
- Copy the Endpoint (it looks like
https://your-resource.openai.azure.com) and KEY 1. - In Claspine, open AI Gateway → Your AI provider keys → Azure OpenAI → Add key, paste both values and save.
- In AI Gateway, add a step with this provider and type your deployment name as the model.
What goes where in Claspine
| Console field | Value |
|---|---|
| ENDPOINT | The resource address, https://<resource>.openai.azure.com. Claspine only accepts Azure OpenAI addresses. |
| API KEY | KEY 1 (or KEY 2) from the Keys and Endpoint page. |
Keep it safe
- Azure gives you two keys so you can change them without downtime: switch Claspine to KEY 2, then regenerate KEY 1.
- Set a budget with alerts in Azure Cost Management.
Claspine stores the values encrypted, never shows them again in full, and never sends them to your app or SDK.
If something goes wrong
- Error that the deployment was not found.
- Use the deployment name from step 2, not the model’s own name.