agenttop
A local, cross-tool usage dashboard for AI coding assistants that reads the logs and local databases already kept by Claude Code, Cursor, Kiro, Copilot and Codex to show cost and workflow patterns.
agenttop computes cost and token numbers from each tool's own local session data rather than by guessing from keywords, since each assistant already stores usage locally but shows only its own slice of the picture. A terminal dashboard shows seven live panels covering cost by project and by model, a daily cost histogram, hourly activity, a breakdown of coding, debugging, testing and exploration time, per-tool session and token counts, and the share of edits that pass without a retry.
A companion web dashboard adds a force-directed knowledge graph of activity, model usage split by input, output and cache tokens, and a searchable, paginated session browser. Everything runs locally with no Docker and no required API keys, installed with a one-line script that also offers a demo mode with sample data safe for screen recordings.