Manage several AI coding agents in parallel

One workspace per project

Add each project folder as a workspace. Every workspace keeps its own layout, and switching workspaces does not stop anything: background tabs keep running. The sidebar shows how many shells are running in each workspace, and a summary of its agents ("1 agent needs you · 2 working").

Name what each tab is for

Split a pane so an agent sits beside a plain shell for tests or logs, and rename the tabs (Backend, Tests, Logs). Two agents in one folder are fine; each tab tracks its own agent. Drag a tab to another pane to rearrange, and its shell keeps running.

See state without switching tabs

Each agent tab shows a logo, a name and a state. Working is mint and pulsing, Needs you is amber, then Done and Error. The window title carries a count, such as (1) car-rental — SplitNest, so you can see it in the taskbar while SplitNest is behind another window. A state the agent reported is solid; a guess is dashed with a "?".

Go straight to the one that is waiting

The bell in the title bar opens the agent inbox (Ctrl+Shift+A), which lists every waiting agent across all workspaces. Agents that need you come first, then errors, then finished turns, and within each group the longest wait comes first. Ctrl+Shift+U jumps to the next waiting agent from anywhere.

Which agents

Claude Code, Codex, Gemini CLI, GitHub Copilot CLI, Cursor Agent, Aider, OpenCode, Qwen Code and Amp are recognised. Agents that report their own state (the Program Status Protocol, OSC 7501) give the most accurate result; for others SplitNest falls back to process detection in PowerShell tabs, and to clearly marked guesses.

Good to know

SplitNest is an MVP prototype tested on Windows 11 x64. Scrollback is per pane and lives only in memory. See the known limitations for the full list.