Use agentacct when an operator needs a local audit trail for what coding agents did, what evidence they recorded, and what the run likely cost. The workflow is to install agentacct, run onboarding once per machine or project, let it detect local Claude Code and Codex session logs, and then use the local dashboard plus MCP recording tools to inspect tasks, sections, machine checks, token counts, cost estimates, and attribution confidence. This is useful after multi-agent work, client-billable coding sessions, or team reviews where a transcript alone is too noisy and a token counter alone is too thin. Invoke it instead of reading raw client logs when the question is which task consumed resources, which checks passed, and how strongly each usage record maps to the work. The scope boundary is local-first coding-agent observability and evidence review. It is not a cloud billing system, provider invoice source, generic analytics platform, or broad LLM observability SDK.