AgentSight — Cockpit for Parallel Coding Agents

Bring scattered AI coding agents from ten terminals into one observable, interruptible dashboard

Problem

Developers routinely run 3+ AI coding agents exploring the same idea in separate terminal windows, and have to switch back and forth to see which one is stuck or drifting. One builder says they were tired of having 10 terminal windows open, and another notes that what they really want is not a window tiling tool but something that understands agent state and keeps multiple agents on the same thread.

Solution

A locally running agent cockpit that auto-discovers common CLI coding agents (Claude Code, Codex, etc.), aggregates each session's status, recent output, and waiting-for-input events into a single window, lets you assign a shared goal and constraints to multiple agents, and offers a one-click pause when any agent is about to run a risky command.

Users

Heavy developers, indie hackers, and small engineering teams running multiple AI coding agents at once

MVP

Auto-discover processes and sessions; unified status/event stream view; desktop notifications when an agent waits for input or errors; pause/resume a single agent; export a short summary when a session ends

Revenue

One-time license at $29–$49, or $8/month subscription; a per-seat team collaboration view as an upsell

Why now

AI coding is shifting from a single session to many parallel sessions, and terminal fragmentation is the most immediate friction; indie developers are themselves the target users, so validation is cheap