finds.dev

mikehasa/agentacct

See what your coding agents did and what it cost. Breaks each task down into work steps — tools used, files changed, tests run, time and tokens spent. Local-first dashboard for Claude Code, Codex, OpenCode, and more. No login, no telemetry.

View on GitHubHomepage ↗

★ 762PythonMITupdated Oct 2026

// the take

agentacct is a local-first CLI/TUI (plus a native macOS app) that parses the session logs of coding agents like Claude Code, Codex, and OpenCode to show what each agent task actually did and what it cost. It's for developers running multiple AI coding agents across projects who want a real work log and spend tracker instead of trusting the agent's own "done" summary.

It only reads local session files, binds via hooks/MCP at session start, and the local API listens solely on 127.0.0.1 - the privacy claims actually match the architecture rather than being marketing. It separates an agent's self-reported "Reported" status from an independently checked "Verified" one, and tags every usage-to-work join with a confidence level (exact/high/medium/low) instead of silently merging uncertain data. The per-agent coverage matrix is honest about depth: Claude Code and Codex get full receipts via hook bridges or log matching, while agents like Hermes only join on IDs the agent explicitly passes, and that's documented rather than glossed over.

No native Windows support - WSL only - which cuts out a large slice of the dev audience for a developer tool. Cost figures are explicitly pricing-table estimates (≈$), not invoice data, and daily token totals are bucketed by activity date rather than true per-day consumption for sessions spanning multiple days, so it's unusable for actual billing reconciliation. Data quality is uneven across agents by design: less-integrated clients (Hermes, DeepSeek Harness, Kimi Code) will produce thinner, lower-confidence receipts, which could frustrate users whose primary agent isn't Claude Code or Codex. The project is self-described early alpha but already spans a Python CLI, a separate Swift macOS app, and a large snapshot/visual-regression test suite - a lot of surface area to keep in sync as upstream agent log formats change.

We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →

// at a glance

Stars
762
Forks
82
Language
Python
License
MIT
Last commit
1 Oct 2026

// nearby

If agentacct isn't quite it.

All repos →