Skip to main content

Example Integration (REST API)

Sending LLM Logs to LogSpend REST API​

In cases where you can’t use the LogSpend SDK, you can directly send your LLM logs to the LogSpend REST API.

Endpoint​

POST https://api.logspend.com/llm/v1/log

Headers​

Replace <LOGSPEND_API_KEY> and <LOGSPEND_PROJECT_ID> with your LogSpend API key and the relevant LogSpend project ID accordingly.

curl -X POST https://api.logspend.com/llm/v1/log \
     -H "Authorization: Bearer <LOGSPEND_API_KEY>" \
     -H "Content-Type: application/json" \
     -H "LogSpend-Project-ID: <LOGSPEND_PROJECT_ID>" \
     -d "<PAYLOAD>"

Body​

The body of the request should follow the structure outlined below:

{
  "input": dict,  # Input data passed during the LLM API call
  "output": dict,  # Response generated by the LLM API call,
  "identity": dict,  # Identity object to identify the user interacting with the AI Assistant. It must contain a session_id and can include a user_id
  "custom_properties": dict,  # Any additional custom datapoints you would like to add like task_name, customer_id, etc.
  "start_time_ms": int,  # Timestamp in milliseconds just before making the LLM call
  "end_time_ms": int  # Timestamp in milliseconds after receiving the response of the LLM call
}
  1. Prepare your input data, custom properties and record the start_time before calling the LLM API.
import time

input_data = {
    "provider": "openai",
    "model": "gpt-3.5-turbo-instruct",
    "prompt": "Say this is a test",
    "messages": [{
        "role": "assistant",
        "content": "",
        "function_call": {},
    }],
    "functions": [{}],
    "max_tokens": 7,
    "temperature": 0
}

identity_data = {
    "session_id": "session-123",
    "user_id": "123455",
}

custom_properties_data = {
    "task_name": "chatbot-qa",
    "customer_id": "chatbot-qa",
}

start_time_ms = int(time.time() * 1000)
  1. Make a call to OpenAI (or whichever provider you're using) to generate the output_data and record the end_time.
def call_openai(input_data):
    # Placeholder for actual OpenAI call
    output_data = {
        "id": "chatcmpl-123",
        "choices": [{
				    "index": 0,
				    "message": {
				      "role": "assistant",
				      "content": "\n\nHello there, how may I assist you today?",
				    },
				    "finish_reason": "stop"
				 }],
		    "usage": {
		        "prompt_tokens": 9,
		        "completion_tokens": 12,
		        "total_tokens": 21
		    },
		    "http_status_code": 200,
		    "http_error_message": "",
    }
    return output_data

output_data = call_openai(input_data)
end_time_ms = int(time.time() * 1000)
  1. Build the json payload as described in the body section above, only after the API call to the LLM
json_payload = {
  "input": input_data,
  "output": output_data,
  "identity": identity_data,
  "custom_properties": custom_properties_data,
  "start_time_ms": start_time_ms,
  "end_time_ms": end_time_ms,
}
  1. Send the log to LogSpend.
curl -X POST https://api.logspend.com/llm/v1/log \
     -H "Authorization: Bearer <LOGSPEND_API_KEY>" \
     -H "Content-Type: application/json" \
     -H "LogSpend-Project-ID: <LOGSPEND_PROJECT_ID>" \
     -d "<json_payload>"