OpenAI

You can only have one OpenAI key per organization. This grants you access to set of whitelisted models. For a list of these models, see Supported base models.

If you don't already have one, you can create an OpenAI account here.

You can find your OpenAI API key on the API key page.

Gemini

You can only have one Gemini key per organization. This grants you access to set of whitelisted models. For a list of these models, see Supported base models.

For information on getting a Gemini API key, see Get a Gemini API key.

Vertex AI

You can only have one Vertex AI key per organization. This grants you access to set of whitelisted models. For a list of these models, see Supported base models.

Follow the instructions here to generate a credentials file in JSON format: Authenticate to Vertex AI Agent Builder - Client libraries or third-party tools

The JSON credentials are required. You can also optionally provide the project ID and location associated with your Google Cloud Platform environment.

Anthropic

You can only have one Anthropic key per organization. This grants you access to set of whitelisted models. For a list of these models, see Supported base models.

For information on getting an Anthropic API key, see Anthropic - Accessing the API.

Azure OpenAI

Each Azure OpenAI key is tied to a specific deployment, and each deployment comprises a single OpenAI model. So if you want to use multiple models through Azure, you will need to create a deployment for each model and then add each key to Label Studio.

For a list of the Azure OpenAI models we support, see Supported base models.

To use Azure OpenAI, you must first create the Azure OpenAI resource and then a model deployment:

1. From the Azure portal, create an Azure OpenAI resource.

1. From Azure OpenAI Studio, create a deployment. This is a base model endpoint.

When adding the key to Label Studio, you are asked for the following information:

You can find all this information in the Details section of the deployment in Azure OpenAI Studio.

Azure AI Foundry

Use the Azure AI Foundry model catalog to deploy a model: AI Foundry docs.

Once deployed, navigate to the Details page of the deployed model. The information you will need to set up the connection to Label Studio is under Endpoint:

When adding the key to Label Studio, you are asked for the following information:

Custom LLM

You can use your own self-hosted and fine-tuned model as long as it meets the following criteria:

• Your server must provide JSON mode for the LLM, specifically, the API must accepts response_format with type: json_object and schema with a valid JSON schema: {"response_format": {"type": "json_object", "schema": <schema>}}
• The server API must follow OpenAI format.

Examples of compatible LLMs include Ollama and sglang.

To add a custom model, enter the following:

• A name for the model.
• The endpoint URL for the model. For example, https://my.openai.endpoint.com/v1
• An API key to access the model. An API key is tied to a specific account, but the access is shared within the org if added. (Optional)
• An auth token to access the model API. An auth token provides API access at the server level. (Optional)

Example with Ollama

1. Setup Ollama, e.g. ollama run llama3.2
2. Verify your local OpenAI-compatible API is working, e.g. http://localhost:11434/v1
3. Create an externally facing endpoint, e.g. https://my.openai.endpoint.com/v1 -> http://localhost:11434/v1
4. Add connection to Label Studio:
   • Name: llama3.2 (must match the model name in Ollama)
   • Endpoint: https://my.openai.endpoint.com/v1 (note v1 suffix is required)
   • API key: ollama (default)
   • Auth token: empty

Example with Hugging Face Inference Endpoints

1. Use DeepSeek model
2. In API Keys, add to Custom provider:
   • Name: deepseek-ai/DeepSeek-R1
   • Endpoint: https://router.huggingface.co/together/v1
   • API key: <your-hf-api-key>
   • Auth token: empty