AI Credits: Track and Analyze Consumption in Automation Logs

TimeTonic automation logs are much more than a simple error journal. They are your reference tool for tracking your AI credit consumption in detail.

What do the AI consumption logs contain?

For each execution, the logs display:

• the AI model used (GPT-4o, Mistral, Claude, etc.)
• the number of input tokens
• the number of output tokens
• the corresponding AI credit cost

Compare the cost of different AI models on the same task

The logs let you concretely compare the consumption of several models on the same automation. A valuable tool for choosing the most cost-efficient model without sacrificing the quality of your results.

Anticipate, optimize, and diagnose

Using the AI credit logs, you can:

• optimize your model choices based on the quality-to-cost ratio
• anticipate your consumption and avoid unexpected overruns
• diagnose unexpected behavior in your AI automations

Accessing automation logs: two entry points

Automation logs can be accessed from several places in TimeTonic. This section presents two ways to reach them: from the automation's execution logs, and from the field history of a record.

Each entry corresponds to an execution on a data row, with its status, date, and details. You can open an execution to analyze the result and understand the automation's behavior.

From an automation

Open your automation, then access the execution logs. Each row corresponds to an execution on a specific data row, with its status, date, and exact time. Click the arrow on a successful execution to expand its details and analyze the result.

From a record's field history

It is also possible to view the changes applied to a record directly from its row. This view details the changes made to each field, execution by execution.

Reading an execution's details

By clicking on an execution, you access the full details of the AI call. You will find:

Comparing two models on the same task

The logs are particularly useful for comparing the real cost of 2 models on an identical task, before deciding which one to use in production.

Here is a concrete example from real logs, on the same expense receipt extraction task:

💡 Tip: Both models perform the same image document extraction task with a correct result. Mistral-Small-3.2-24B consumes 8 times fewer credits than Qwen2.5-VL-72B here. The logs let you verify this yourself and choose the model best suited to your budget and the complexity of your documents.

To go further and anticipate your consumption before launching your automations → Estimate your AI credit consumption