Free for individual repos

Know what your AI agents actually did.

Every Claude Code, Cursor, and Codex session — plus your CI agent runs — with the tokens burned, who ran it, what triggered it, how long it took, and whether it worked. Per engineer, per project, per task.

Install free

By signing in you agree to our Terms and Privacy Policy.

AgentMeter runs feed showing local and GitHub Actions sessions with tokens, engine, source, status, and quality score

AgentMeter run detail page showing run metadata, trigger, status, tokens, and quality rating

AgentMeter run detail token breakdown showing input, output, and cache tokens

AgentMeter trends dashboard showing token usage over time broken down by source, project, engine, model, and contributor

AgentMeter contributors dashboard showing per-engineer token usage, run counts, and connection status

AgentMeter contributor detail page showing token usage over time and run history for a single engineer

Every session and CI run in one feed — local or GitHub, tokens, engine, status, and quality score at a glance.

Drill into any run: what triggered it, what it did, how long it took, and how good the result was.

Exact token counts per run — input, output, cache writes, and cache reads — plus your cache hit rate.

See token usage by source, project, engine, model, and engineer. Know that local sessions are 31% of your usage — or that one teammate is driving a third of it.

Score run quality against token usage. Find out which models and workflows actually deliver.

See how your team uses AI agents — per-engineer tokens, run counts, and live connection status in one view.

Drill into any contributor: their token usage over time, run history, and how they compare to the team.

Two ways to track. One dashboard.

Local sessions

Track Claude Code and Cursor sessions on every engineer's machine. A lightweight CLI scans local data — no API keys shared, no proxying.

npx @agentmeter/cli init

Most teams start here

CI agent runs

Track GitHub Actions agent runs — gh-aw, Claude Code Action, Codex. One step in your workflow records tokens, trigger, and outcome, and posts a summary right on the PR.

- uses: agentmeter/agentmeter-action@v1
  with:
    api_key: ${{ secrets.AGENTMETER_API_KEY }}

See every session

Local Claude Code, Cursor, and Codex sessions plus GitHub Actions agent runs in one feed. Input, output, and cache tokens for every run.

Know who and what

Usage attributed to engineers, projects, workflows, and tasks. Know that issue #47 took 2.1M tokens across four runs, or that one teammate drives a third of the team’s usage.

Measure outcomes, not just usage

Status, duration, and 0–100 quality ratings for every run. Find out whether your heaviest runs are actually your best ones.

The problem isn't usage. It's visibility.

No one could see the usage piling up until the bill arrived.

Uber burned through its entire 2026 AI budget in four months. Microsoft revoked Claude Code licenses across its engineering org. The FinOps Foundation calls it “existential crises.”

The common thread isn't heavy use — it's that no one could see where the tokens were going until the bill arrived.

AgentMeter is the visibility layer. Token usage per engineer, per project, per task — as it happens, not after the invoice.

Read the full story →