Antiburn

Compatible with Claude Code, Codex, Cursor, Pi, OpenCode, and other agents, it analyzes session logs to show where tokens are going and which settings or habits may be driving unnecessary usage. Its reports highlight issues such as excessively long contexts, high reasoning levels, overused fast modes, costly subagents, and unused tools, skills, or MCP servers. antiburn also identifies potential model upgrades and estimates the monthly impact of usage patterns using equivalent API pricing.

The app offers practical recommendations to help users adjust configurations and reduce waste, including guidance on auto-compaction and manual session compaction. It tracks usage over time so developers can compare the effects of new models, settings, and working habits. Designed to be lightweight, antiburn aims to use less than 300 MB of memory and 5% CPU. All analysis runs locally on the user’s device, without requiring a server or cloud service.