To log your OpenAI Assistant chats, you can log the messages using the API logging method.

When you log to Athina, each chat message should be logged using a separate POST request.

You can group messages into a single conversation by using the session_id field. This is equivalent to the thread ID from OpenAI.

  • Method: POST

  • Endpoint: https://log.athina.ai/api/v1/log/inference

  • Headers:

    • athina-api-key: YOUR_ATHINA_API_KEY
    • Content-Type: application/json

​
Request Body

Here’s a simple logging request:

{
    // Required fields for logging
    "language_model_id": "gpt-4",
    "prompt": [
        { "role": "system", "content": "System Instructions for the assistant" },
        { "role": "user", "content": "Individual message level instructions" },
    ],
    "response": "Response message from the assistant (as a string)",
    "session_id": "session-123",

    // Required fields for most evaluations
    "user_query": "The message from the user you are responding to",
    "context": "Your retrieved documents as a string or object", // Can be a string or object

    // Token usage logs (optional, but recommended). These can be found in the "Run" object from OpenAI Assistants API.
    "prompt_tokens": 123,
    "completion": 100,
    "total_tokens": 223,
    "cost": 0.12314, // if you provide tokens, we will calculate the cost for you automatically.

    // Optional fields for better segmentation and analytics (recommended)
    "prompt_slug": "customer-support/v1", // any string
    "environment": "production", // any string
    "customer_id": "disney-usa-2314832", // any string, this is your internal customer identifier
    "customer_user_id": "michaelmouse@gmail.com", // any string, this is the end user identifier,
    "external_reference_id": "239ej92-r24-4838-45663", // any string, this is your internal reference ID for the message,

    // Tool calls
    "tool_call": tool_calls, // can be any JSON object
    "tools": tools, // can be any JSON object

    // Ground truth data - used for some evaluations
    "expected_response": "If you have a ground truth response, you can send this here",

}

Note: OpenAI Assistant’s API doesn’t provide the actual prompt that was sent. So, the prompt you are logging might not be the complete prompt that was actually sent to the model.

However, the prompt field is not currently used for evaluations so it will not affect the ability to run evals

OpenAI Chat (0.x)Tracing