@yusukeshib/pi-token-counter

A pi extension that appends each tool's own output token cost to its tool-use row, color-coded by size. This is the tool's own contribution to the context (its result content) — not the cumulative total.

What it looks like

Each tool-use row gains a trailing token count, colored by how much it adds:

read src/main.rs                 +312 tok
$ npm test -- --watch=false      +1,840 tok
grep "TODO" src                  +6,204 tok

| Color | Meaning | Threshold | | ----------------- | -------------------------------- | --------------- | | blue (accent) | small output | < 1,000 tok | | yellow (warning) | medium output | < 5,000 tok | | red (error) | large output | >= 5,000 tok |

The token estimate uses the same heuristic as pi's built-in estimator (text characters / 4).

Install

pi install npm:@yusukeshib/pi-token-counter

Via git (no npm publish required):

pi install git:github.com/yusukeshib/pi-token-counter

Try it without installing:

pi -e npm:@yusukeshib/pi-token-counter

How it works

The tool_result event carries each tool's result content, so the extension estimates its tokens there and caches the value by toolCallId. It overrides only renderCall for the built-in tools (read / bash / edit / write / grep / find / ls) to append the colored token count — renderResult is left untouched so pi's built-in result rendering (diffs, syntax highlighting, etc.) is fully preserved. When a result arrives the row re-renders and renderCall reads the cached estimate.

License

MIT