
Agent Monitor
A live, fully local mission-control dashboard for AI coding agents.
// The problem
When Claude Code runs several sessions and subagents in parallel, it is hard to see what each one is doing, which one is waiting for your input and what the work costs. The terminal shows one session at a time, and streaming prompts and code to a cloud service is not an option for private projects.
// The solution
Agent Monitor hooks into every Claude Code event and streams it to a local dashboard. A Bun server stores the events in SQLite and broadcasts them over Server-Sent Events, while a React and Redux Toolkit interface shows every session, subagent and tool call live. Everything binds to 127.0.0.1, so nothing ever leaves the machine.
// Highlights
- Session list with live status (working, needs you, idle) and system-wide alerts when an agent waits for input
- Per-session timeline and trace waterfall, plus a tool-call inspector with real diffs
- Token usage and cost estimates from local transcripts: today, 7 days, 30 days, year
- Agent-agnostic event API: Cursor and other agents report through the same contract
- Replay of past sessions from the local history
// Tech stack
- React 19
- TypeScript
- Redux Toolkit
- Bun
- SQLite
- Server-Sent Events
- Vite
- Tailwind CSS



