Agent Trace — coding agent usage audit

See how many tokens each coding agent burns and which steps cost the most

Problem

Developers juggling Claude Code, Codex, Cursor, OpenCode, and Grok have little idea how much each task consumes, which step burns budget, or which agent is cheaper. Official billing shows only totals, with no attribution by task, repo, or step — and the community is openly asking how to cut agent usage.

Solution

A local-first trace collector: parse token and call records from each agent's logs/sessions, aggregate usage and cost by task, repo, and tool, flag expensive steps, and suggest optimizations like trimming repeated context or switching to cheaper models.

Users

Indie developers heavily using multiple coding agents, tech leads at small teams, engineering managers watching AI spend

MVP

Parse local session logs of major agents; aggregate tokens and cost by task/repo; highlight expensive steps; daily/weekly usage reports; local storage, no code upload

Revenue

Free for a single agent, subscription for multi-agent merged stats, team dashboards, and history retention

Why now

Coding agents are fragmented and priced differently, creating fresh demand for usage and cost transparency; logs are local, so a solo dev can ship a cross-tool unified view that big vendors won't build across competitors.