CodeBurn maps agent token spend

The useful part here is not the charting, it’s the observability layer: CodeBurn makes AI work measurable without a proxy, API keys, or cloud dependency. That turns token burn from a vague annoyance into something teams can actually optimize.

• –Deterministic classification avoids adding another LLM in the loop, which keeps the tool cheap and predictable
• –Local-only reads are a strong fit for privacy-sensitive teams and solo developers
• –The one-shot and retry views add quality context, not just spend totals
• –Broad provider support makes it more than a single-assistant accessory
• –The deeper signal is category-level spend drift: debugging, planning, and conversation can quietly dominate usage