agentacct

Why we included this project

Coding agents like Claude Code, Codex, OpenCode, and Hermes leave session logs on disk, and agentacct turns those into a per-task work receipt: the commands run, files touched, tools used, time and tokens spent, and which parts of the work carry a passing machine check. The evidence model is the part worth calling out. An agent's own claim is kept separate from hook-observed exit codes and CI results, so a task only reads as verified when independent checks actually pass. Everything runs locally with a loopback-only JSON API, a terminal dashboard, and a macOS app, with no account and no telemetry, which suits teams that want agent observability without shipping logs to a hosted service. Developers and tech leads who need to audit agent work or track token spend per task will find it useful, and it's a decent way to catch agents that report done without real proof.