Monitoring for the agents that write your code

For each session, AgentPeek shows what a working developer actually checks: the state (executing, thinking, waiting, or idle, plus an attention flag), the current activity, files touched, commands run, diff counts, the agent's own todo checklist, and a timeline of tool calls, prompts, and answers. Running subagents show up under their parent. It reads the same for all twelve agents, from Claude Code to Pi, so one habit covers your whole stack.

The primary surface is the Mac notch: always present, glanceable, expanding on demand. The Agent Board floats a kanban of every session for the wide view, widgets pin todos, transcripts, or usage anywhere on any display, and menu bar mode carries usage gauges to notchless setups. Same live state, whichever surface fits your desk.

A dashboard you can only read still makes you switch to a terminal to act. From AgentPeek you allow or deny permissions, answer questions, approve plans, and send the next prompt into a resumable session, with replies streaming into the session card. Monitoring and response in one place is what actually kills the alt-tab habit.

AgentPeek is a desktop app, not a hosted observability service. It reads what your agents already keep on your Mac; transcripts, diffs, prompts, and usage numbers never leave your machine, and there are no accounts, no analytics, and no telemetry. Monitoring your agents shouldn't mean streaming your codebase to someone's cloud.