# Read token usage and reconcile it with your dashboard

Capture the usage returned by a chat response and compare it with the actual request and credit records in your own account.

## Understand the two records

A chat response can include a standard `usage` object containing prompt, completion and total token counts. Your account dashboard is the place to inspect the actual request and credit charge. Tokens are a measurement of model usage; they are not themselves a currency balance.

Do not assume that a response includes a monetary `usage.cost` field or that an endpoint from another API platform exists on MandAPI. Read only fields that the actual response returns.

## Capture a small response

Install the OpenAI Python client and set `MANDAPI_API_KEY` and `MANDAPI_MODEL_ID` as described in the chat quickstart. Save this as `usage.py`, then run `python usage.py` when you want to make one real request.

```python
# usage.py
import os
from openai import OpenAI

with OpenAI(
    api_key=os.environ["MANDAPI_API_KEY"],
    base_url="https://api.mandapi.com/v1",
    timeout=30.0,
    max_retries=0,
) as client:
    result = client.chat.completions.create(
        model=os.environ["MANDAPI_MODEL_ID"],
        messages=[{"role": "user", "content": "Reply with one short greeting."}],
    )
    usage = result.usage
    print({
        "response_id": result.id,
        "model": result.model,
        "prompt_tokens": usage.prompt_tokens if usage else None,
        "completion_tokens": usage.completion_tokens if usage else None,
        "total_tokens": usage.total_tokens if usage else None,
    })
```

The script does not print the API key, prompt body or model answer. If usage is absent, it records empty values instead of inventing token counts.

## Compare the same request

Open your [MandAPI dashboard](https://app.mandapi.com/console) and find the corresponding request using the time, model and response identifier when available. Compare the returned token counts with the fields actually shown in your account. A client-generated request ID can help your application's own logs, but it is not automatically a searchable MandAPI field.

If you estimate cost from the [published prices](https://mandapi.com/en/pricing), record the price date and units. A basic estimate is input tokens multiplied by the input rate plus output tokens multiplied by the output rate, each divided by one million. Cache, context tiers and other model-specific rules can change that estimate. Use the recorded credit charge for reconciliation.

## Check the result

You should have a timestamped request, a model ID, returned token counts when available and the actual account record. Investigate missing usage or a material mismatch before relying on a report. Never silently convert missing usage to zero.

Keep raw prompts and customer messages out of routine billing logs. Start with counts, identifiers and status; store additional content only when your application needs it and access is controlled.

## References

- [https://mandapi.com/en/docs/quickstart](https://mandapi.com/en/docs/quickstart)
- [https://mandapi.com/es/docs/quickstart](https://mandapi.com/es/docs/quickstart)
- [https://github.com/openai/openai-python](https://github.com/openai/openai-python)