yeaboi

Agent Usage Beta

What your AI coding agents actually cost, broken down by model, by project, by telemetry source, and across a daily trend.

Beta. The Agents family is new, and every number it shows is computed from telemetry your agents happen to leave behind. Treat it as an indicator, not an audit.

How it works

Usage is a deterministic gather followed by a single LLM call. yeaboi refreshes the collector's ingest, aggregates the stored session rollups over the window, and prices every (model, session) pair from the shared rate table in pricing.py — the same table every cost estimate in the product goes through, cache-aware so cache reads and cache writes are not billed as ordinary input.

The LLM writes only the prose. Insights and recommendations are generated over the finished aggregates; every number on the page is computed, never written by a model. With no LLM configured the report still builds — you get the figures and a warning instead of the commentary.

The window, and two deliberate approximations

The default window is the last 30 days, set with --window-days. Two rounding decisions in that windowing both err toward showing work rather than hiding it:

  • A session is placed by when it ended, so a session that ran across the window boundary has its whole cost attributed to the day it finished.
  • The window has no upper bound, so a session whose end timestamp is in the future because of clock skew stays visible instead of vanishing into a gap.

Both are the same judgement call: a cost dashboard that silently omits spend is the worse failure.

Filters

--project is a substring match on the session's project directory name, so --project api catches every checkout with api in the path. --source is an exact match on the telemetry source, currently only claude_code.

From the command line

yeaboi agents cost
yeaboi agents cost --window-days 7 --project checkout
yeaboi agents cost --format json
yeaboi agents cost --strict         # exit 3 if the run was degraded

Over MCP the same engine is agents_usage, with agents_usage_history for previously stored reports.