mikehasa/agentacct
agentacct turns local coding-agent logs into audit-style receipts for what an agent did, what it cost, and what evidence supports it.
The project says it works across Claude Code, Codex, OpenCode, and Hermes by reading session logs already on disk. It separates an agent’s “done” claim from verification, with evidence tiers ranging from self-reported checks to hook-observed exits and CI. It offers the same receipts in a macOS app, terminal dashboard, and local JSON API, with no hosted account or telemetry. Costs and usage are labeled by their basis, including estimates where exact provider billing is not available. Source: GitHub · LLM repos' note.
The project says it works across Claude Code, Codex, OpenCode, and Hermes by reading session logs already on disk. It separates an agent’s “done” claim from verification, with evidence tiers ranging from self-reported checks to hook-observed exits and CI. It offers the same receipts in a macOS app, terminal dashboard, and local JSON API, with no hosted account or telemetry. Costs and usage are labeled by their basis, including estimates where exact provider billing is not available. Source: GitHub · LLM repos' note.
score 4