Five agents. One UI.
Pick the agent for the job. Each session keeps its own transcript.
-
Claude Code
-
Codex
-
Gemini CLI
-
OpenCode
-
Hermes
Web-based. Any provider. Any model.
Works with
Requires Node 20 or newer.
Pick the agent for the job. Each session keeps its own transcript.
20+ built-in presets, or point at any Anthropic- or OpenAI-compatible API. Pick any model per session.
cdesktop holds every session at once. Tile up to four side by side. Switch with no reload.
Split the workspace into halves or quarters. Up to four cells at once.
No reload between threads. Transcripts and diffs stay where you left them.
Pull any session into any cell at any time. The layout follows your attention.
The lead agent calls in teammates that share the same workspace. Each teammate runs on whatever agent and model suits its subtask.
Lead uses Claude Code, teammates use whatever fits the subtask. Per-teammate provider, model, and reasoning.
Every teammate sees the same files and the same transcript. No copy-paste between windows.
The lead agent spawns teammates with npx cdesktop team spawn and routes follow-ups back to itself.
Routines run an agent on a cadence. Each tick spawns its own session you can review, approve, or discard.
Pick a cadence per routine. Disable any time without losing the schedule.
Each tick spawns its own session with diff, transcript, and preview ready for review.
Approve what is useful, discard the rest. Routine runs never touch your branch until you say so.
One branch per session, or work in the folder directly. Non-Git directories work too.
Send feedback to the agent without leaving the diff view.
DevTools, inspect mode, device emulation. Preview your running app side by side with the diff.
Open pull requests with generated summaries. Review on GitHub. Merge from the same window.
No account. No cloud. Your code stays on your machine.