model-usage

To track Claude and Codex spending from local cost logs, you parse the CodexBar cost JSON to aggregate model-cost data from daily breakdowns. This attributes spending to individual models, exposing results in text or JSON formats with fields like provider, mode, and totalCostUSD.

Yes, you can filter CodexBar cost summaries by date range and input source. This functionality allows you to generate current-model summaries or full model breakdowns across multiple days or providers, enabling targeted cost insights and reporting.

When aggregating model costs from logs, the output is exposed in either text or JSON formats. The JSON format includes specific fields such as provider, mode, model, totalCostUSD, and dailyRowCount to provide structured data for cost reporting.

Processing CodexBar logs does not require any external dependencies. The skill operates locally using built-in scripts and references to parse the cost JSON and aggregate the daily breakdowns without needing additional packages or environments.

The best way to audit spending across different AI models locally is to aggregate per-model costs from CodexBar logs. By parsing the local cost JSON, you can attribute spending to individual models for Codex and Claude, generating a clear breakdown for budget optimization.

CodexBar local cost logs are hard to interpret at the model level because the raw JSON format contains unaggregated daily breakdowns. Aggregating these logs attributes spending to individual models, providing clear, actionable insights for cost analysis.