Watch usage live
Follow input, cached input, output, reasoning, tool tokens, and normalized I:O ratios as sessions update.
A live, local monitor for your AI coding sessions. See usage, cache efficiency, subagents, and cost across every tool, without your data leaving the machine.
go install github.com/polera/tokenhawk@latest Tokenhawk indexes the session metadata your tools already keep and turns it into a clear, live picture.

Agent providers, one interface
Default reconciliation interval


Native visibility across the tools you already use
Tokenhawk surfaces the numbers that matter while work is happening, and keeps the raw context where it belongs.
Follow input, cached input, output, reasoning, tool tokens, and normalized I:O ratios as sessions update.
Separate parent and subagent usage, including which agents are running and what each one costs.
A clear low-cache warning flags high-input sessions before unnecessary spend disappears into the scrollback.
Export visible sessions as structured JSON or CSV, with model-level and subagent detail intact.
Tokenhawk runs on your machine and reads only the usage metadata your coding tools already store. No account. No telemetry. No cloud dashboard.
Read the privacy model →Add a compact status line to Claude Code, or run Codex, Gemini, Pi, and OpenCode through a tmux-backed wrapper.
Set up live status →❯ codex
Working ▋
Install Tokenhawk and get eyes on your agent usage in under a minute.
$ go install github.com/polera/tokenhawk@latest