We're constantly typing commands.
Whether it's for everyday system operations:
- Opening terminals
- Switching directories
- Running Git commands
- Managing files
- ...
Or while working with agent harnesses:
- Using skills
- Running workflows
- Handling interactive sessions
- ...
If any of these problems sound familiar:
- Repetitive operations quietly consume a surprising amount of time.
- Workflow execution is difficult to observe.
- The boundary between automation and human interaction is unclear — agents sometimes act on their own, while at other times they interrupt you too frequently.
- Switching between different agents feels fragmented.
- Building and customizing workflows is often complicated and difficult to put into practice.
Then you might want to try CLILoom.
CLILoom provides:
Visual Workflows
Visually compose commands, conditions, parallel tasks, and more into reusable workflows.
Create Workflows with Natural Language
Tell your AI CLI assistant what you want, and let it create or modify a workflow tailored to your needs.
Parallel Real Terminals
View real-time terminal output from multiple parallel branches on the same screen.
Automation + Human Collaboration
Automatically execute commands, request user input when needed, or switch into a fully interactive terminal whenever human intervention is required.
Works with Any AI CLI
Launch any agent harness you want by defining the command used to start it in the terminal.
Whether you use Claude Code, Codex, OpenCode, Kimi, Pi, or something else, CLILoom doesn't lock you into a specific agent.
Cross-Platform and Local-First
CLILoom supports Windows, Linux, and macOS, with workspace data stored locally on your machine.
Development Status
The first version is now available.
CLILoom has already reached the point where I can develop CLILoom from inside CLILoom itself.
Windows and Linux have both been tested.
If you want to use CLILoom with WSL, I recommend running the Linux GUI version directly inside WSL/WSLg.
I don't currently have a Mac, so the macOS version has not been properly tested yet. Feedback from macOS users would be very welcome!
Screenshots
Project
GitHub:

Top comments (0)