On this page
- Why many terminal windows stop working
- Start an AI session in a project
- Run sessions in many projects at once
- Send work to an agent from anywhere in the app
- Sessions that do not belong to a project
- Pick up where you left off after a restart
- Tips for running agents in parallel
- Works with every project
- References
It is 3 p.m. and you have six terminal tabs open. One agent fixes a bug in an API. Another writes tests for a mobile app. A third reads a failed CI log. Which one is waiting on you?
The agents are fast. Keeping track of them is the hard part. This guide shows a local setup where each agent runs on your own machine, in its project folder, and one window shows all of them.
In short:
- Start Claude Code, Codex or OpenCode in any project with ⌘+J.
- Every session gets its own headless terminal tab, and the status bar shows what is running.
- The terminal comes to the front when an agent needs input or a command fails.
- Unfinished sessions come back after a restart with one click.
- It works in any folder, whatever the stack.
Why many terminal windows stop working
The usual setup is one terminal tab per agent. That is fine for two sessions. Around five or six, things start to slip:
- You do not know which tab belongs to which project.
- An agent waits for your approval, and you miss it.
- You close a tab by mistake, and the conversation context is gone from view.
- You restart your computer and have to hunt down each session ID again.
Switching between tasks has a cost. In one study, people who were interrupted finished faster but reported more stress, frustration and time pressure (Mark et al., 2008). A clear view of each agent cuts the time you spend looking for the right tab.
Start an AI session in a project
In DevWorkspace Pro, every project has an AI launcher. There are a few ways to open it:
| From | Action |
|---|---|
| Project page | Click New AI session next to the project name |
| Keyboard | Press ⌘+J (Ctrl+J on Windows and Linux) |
| Project switcher | Press ⌘+K, select a project, then press Tab |
| Dashboard | Click the sparkle icon on a project card to start a blank session with your default tool |
Type the task and press Enter. The first line of the prompt becomes the session title, and the tool starts in the project root folder.
Want a different tool for one session? Pick it from the tool menu. You can change the flags for a single session too. For the default tool, the flags field shows the value of AI CLI Tool Arguments from Preferences, for example --dangerously-skip-permissions for claude or --yolo for codex.
Run sessions in many projects at once
Each session runs in a headless terminal, which works in the background. The status bar at the bottom of the window shows what is running:
- A project with more than one running command shows one chip with the number of commands.
- Click the chip to see that project's commands, then click one to open it.
- The project page shows a N running button, for example 2 running.
The headless terminal comes to the front when a command needs input or fails. For agents, this is the part that matters. When Claude Code asks permission to run a command, you see it.
Each session has its own tab in the headless terminal window. The window header shows the project name and a short description, such as Issue #42: Fix login redirect. Double-click a tab name to rename it.
Switch projects without losing the agent
Press ⌘+K (Ctrl+K) to open the project switcher above the headless terminal window. Pick a project that has headless commands, and the window shows its most recent one. The terminal gets keyboard focus, so you can type to the agent right away.
| Action | macOS | Windows / Linux |
|---|---|---|
| Minimize the terminal window | Cmd+Shift+M | Ctrl+Shift+M |
| Previous / next command tab | Cmd+Shift+[ / ] | Ctrl+Shift+[ / ] |
| Close the window | Cmd+Shift+W | Ctrl+Shift+W |
Minimize the window and the agent keeps running.
Send work to an agent from anywhere in the app
No more copying context into a prompt by hand. Several features build the prompt for you and send it to your default tool, in the correct project folder:
| Source | What the prompt contains |
|---|---|
| GitHub issues | Issue title, description and comments |
| GitHub Actions | Failed run, job and step, and the end of the log |
| Logs | The log entry and its stack trace |
| Focus Pad | The task text, in the project that you select |
Email notifications not sent for new orders #137
Send to AI · acme-shop
acme/shop · issue #137
Email notifications not sent for new orders
github.com/acme/shop/issues/137
Description, plus 2 comments with author and date
Each one opens a Send to AI dialog. Add anything else in Additional details before you send.
Sessions that do not belong to a project
Some work has no home project: a quick script, or a question about your shell setup. Click New AI session outside a project in the top bar. The session starts in the AI Session Folder from Preferences, or in your home folder if that setting is empty.
Pick up where you left off after a restart
Quit DevWorkspace Pro during an AI session and the app keeps a record of it. After a restart, the status bar shows a button such as 2 to resume. Click Resume next to a session, or Resume all. The tool opens in the same folder, with the same conversation and the same arguments. The details are in Resume Claude Code sessions after a restart.
Tips for running agents in parallel
- One agent, one task. Give each agent a single clear goal. The first line of the prompt becomes the tab title, so keep it short and specific.
- Keep agents off the same files. Two agents editing the same files in one folder can overwrite each other's changes. Put them on different projects or different parts of the code.
- Check the arguments. A flag such as
--dangerously-skip-permissionslets the agent run commands without asking. Use it only where you accept that risk. - Track what each agent owes you. Put each open item in a checklist. The Focus Pad MCP server lets the agent add tasks for you.
Works with every project
AI sessions work in any folder, with no stack detection needed. A Flutter app, a Rust crate, a Ruby gem and a Laravel app all get the same launcher, the same headless terminal and the same resume list.
References
- Mark, G., Gudith, D., & Klocke, U. (2008). The cost of interrupted work: More speed and stress. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 107–110. https://doi.org/10.1145/1357054.1357072
FAQ
Frequently asked questions
Which AI coding tools does DevWorkspace Pro support?
Claude Code (claude), Codex CLI (codex) and OpenCode (opencode). The tool must be installed and on your PATH. You select a default tool in Preferences, and you can pick a different tool for each session in the launcher.
Can I run more than one AI session in the same project?
Yes. Each session runs in its own headless terminal tab. When a project has more than one running command, the status bar shows one chip with the number of commands. Click it to open a session.
Do the agents run in the cloud?
No. DevWorkspace Pro starts the CLI tool on your computer, in the project folder. The tool sends prompts to its model provider as it always does, but the files, shell and secrets stay on your machine.
Does it work for any stack?
Yes. AI sessions work for every project folder, for example Flutter, Rust, Ruby, Laravel or a docs repository. No stack detection is needed.
What happens to my sessions when I quit the app?
DevWorkspace Pro keeps a record of unfinished sessions. After a restart, a button shows the number of sessions to resume, and the tool opens again with the same conversation.