AgentMeter: AI Agent Cost Tracking for GitHub Actions

I built AgentMeter because AI agent costs are a complete black box — and I got tired of guessing.

Read the full blog post for the full story.

Every agent run costs money. A single feature can spawn 3–5 runs: implement → review → test → fix → retry. But Anthropic's console shows only aggregate daily spend, GitHub Actions shows only runner minutes, and the agent itself surfaces nothing. You can't answer "How much did issue #47 cost?" without AgentMeter.

AgentMeter is two things: a GitHub App that gives you a dashboard of every agent run across your repos — token counts, cost, duration, status, and which issue or PR triggered it — and a companion GitHub Action you drop into any existing workflow in three lines of YAML. It works with Claude Code, Codex, and GitHub Agentic Workflows. Cost summary comments are posted directly on every PR or issue automatically.