Burn Bar for Omarchy visualizes Claude/Codex token burn, quotas and GPU load locally

DanWahlin · x · 2026-09-05

A new Burn Bar for Omarchy shows every model your machine is burning in real time: token usage bucketed in 30-minute windows, split into input, cache write, output and cache reads; per-plan quota bars with countdowns and reset times that pulse above 90%; per-model Claude spend; plus GPU load, watts, temperature, VRAM share and which model is resident, with buttons to warm or unload models.

The implementation is deliberately minimal: a Python 3 collector with zero third-party dependencies, no API key and no telemetry — it just reads the transcripts Claude Code and Codex already write to disk. The only network call is to your own Ollama.

Original post →

More from coding & agent

coding & agent channel →