Local by default
Tokens, cost, per-model breakdown, session blocks and burn rate — all read from the transcripts Claude Code already stores. 100% offline, no API key.
claudgetA lightweight, always-on-top desktop widget that reads the transcripts Claude Code already stores — and turns them into a live ledger of tokens, cost, burn rate, plan limits and recent sessions. Zero setup. Fully local.
Tokens, cost, per-model breakdown, session blocks and burn rate — all read from the transcripts Claude Code already stores. 100% offline, no API key.
Switch on plan tracking to see your 5-hour and weekly windows, percent remaining and a live reset countdown from Anthropic's usage endpoint. Off by default.
Strictly read-only to ~/.claude — it never writes there, never logs your OAuth token, and sends nothing anywhere except api.anthropic.com.
Frameless, translucent and always-on-top, with a compact mode, click-through, tray hiding and global hotkeys. Dark, light or system theme, updating in real time.
~/.claude/projects/**/*.jsonl is parsed and aggregated into token counts, cost estimates, a per-model breakdown, ~5-hour blocks, burn rate and an hourly activity series. This is the source of truth for spend — and it works entirely offline.
Optionally, api.anthropic.comis read with the OAuth token Claude Code already stored — the source of truth for plan limits (percent used, percent remaining, reset time). Polled no more than every 180s with backoff. Off until you enable it.
xattr -dr com.apple.quarantine /Applications/claudget.appI built claudgetbecause I wanted a calmer, always-on way to keep an eye on Claude Code spend without leaving the editor or pasting an API key anywhere. It reads only what the CLI already stores on your machine, stays out of the way, and is free and open source under the MIT license.