Create Model Integration

post https://api.patronus.ai/v1beta1/model-integrations

Create integration to an LLM Model

Open AI integration

To integrate with OpenAI, you need an API key and to select a model.
Optionally, you can set default parameters for the model.
If these aren't specified, the model will use its default settings.

Example request body:

{
    "name": "my-openai-integration",
    "spec": {
        "open_ai": {
            "api_key": "sk-my-openai-api-key",
            "selected_model": "gpt-4",
            "default_params": {
                "temperature": 0
            }
        }
    }
}

HTTP Integration (webhook style)

Custom HTTP integrations allow connection to any LLM model with an HTTP interface that returns a JSON response.

For security, use secrets to transmit sensitive information like API keys.
Secrets can be included in the method, base URL, path, headers, and body_template
by enclosing the secret key in double braces ({{MY_SECRET}}).

Do not use the following reserved keys for secrets:

prompt
user_prompt
system_prompt

The model integration is expected to return JSON response.
response_jq_path is necessary to extract text model's text output.

{
    "name": "my-custom-integration",
    "spec": {
        "http": {
            "request_method": "POST",
            "request_base_url": "https://my-llm.example.com",
            "request_path": "/v1/completion",
            "request_headers": [
                {
                    "key": "X-API-KEY",
                    "value": "{{API_KEY}}"
                }
            ],
            "request_body_template": "{\"user_prompt\": \"{{prompt}}\"}",
            "response_jq_path": ".choices[0].message.content",
            "secrets": [
                {
                    "key": "API_KEY",
                    "value": "my-secret-api-key"
                }
            ]
        }
    }
}