feat(bolt terminal): added dedicated bolt terminal, and attached to workbench

[thecodacus]

Add Dedicated Terminal for AI Operations

Overview

This PR introduces a dedicated terminal ("Bolt Terminal") for AI operations, replacing the previous approach of using throwaway shell processes. This architectural change improves reliability, provides better process management, and enhances the user experience when executing AI-driven commands.

Key Changes

1. Terminal Architecture

• Added a new BoltShell class to manage a persistent terminal session
• Implemented command execution queue management
• Added proper terminal initialization and cleanup
• Introduced command execution state tracking
• Added support for command interruption and process management

2. Action System Updates

• Added new start action type for application startup operations
• Modified ActionRunner to use the dedicated Bolt terminal
• Updated action parsing to support the new action type
• Improved error handling and logging for shell operations

3. UI Enhancements

• Added dedicated "Bolt Terminal" tab in the terminal panel
• Updated terminal UI to distinguish between AI and user terminals
• Modified terminal icons and status indicators
• Added proper terminal initialization states

4. Process Management

• Implemented proper command execution lifecycle
• Added support for command interruption
• Improved error handling for failed commands
• Added command execution state tracking

Technical Details

BoltShell Class

The new BoltShell class provides:

• Persistent terminal session management
• Command execution queuing
• Process lifecycle management
• OSC code handling for terminal states
• Improved error handling

Action System Changes

• New start action type specifically for development server operations
• Modified action runner to use BoltShell for command execution
• Updated action parser to support the new action types
• Improved error handling and status reporting

Testing

• Tested terminal initialization
• Verified command execution and interruption
• Tested multiple command scenarios
• Verified proper cleanup and state management
• Tested error scenarios and recovery

Migration Impact

• No breaking changes for existing functionality
• AI operations now use the dedicated terminal automatically
• Existing terminal operations remain unchanged
• No changes required for existing action implementations

Future Improvements

• Add support for multiple AI terminals if needed
• Add more sophisticated command queuing
• detect terminal error and feedback to AI for auto correction

Preview

https://github.com/user-attachments/assets/632514d9-537c-4b75-a69e-cf7f47fc605d

[coleam00]

This is incredible, thank you so much @thecodacus! We are looking into merging this this weekend!

[wonderwhy-er]

Ouh, that was on my list of things, great addition!

[thecodacus]

thanks @wonderwhy-er , i can see there are lots of things that we have in bolt.new website is missing on the repo. These are very needed things in terms of experience. @coleam00 I have also raised another pr for code streaming please check that out as well

[chrismahoney]

Taking this on to review & merge between tonight and tomorrow; already seen some vast improvements locally, and simplicity of not having to use dev tools for the console output.

I’ll reach out if there are edge cases or non-local scenarios I’d like to confirm are working. Thanks so much for giving this the time it deserves 🙌

[chrismahoney]

@thecodacus The repo name for this PR changed to dash-ai, could you update the PR? I'm reviewing based off of this same branch there for now.

[thecodacus]

@chrismahoney, I have set the name dash-ai from start when I forked it. I don't know if it will cause any issue in here. Shall rename it?

[chrismahoney]

I do see this occasionally, but I believe it's a function of the general prompt not playing well with my local qwen2.5-coder model. Despite the visual it performs the codegen/artifact steps. More an FYI so I can add an issue for this and the Bolt modifications syntax later.

[thecodacus]

thats weird, maybe addition of one extra command sometimes confusing the llm

[chrismahoney]

Oh okay, probably disregard for now. I'll review and if there are merge issues that may just be local for me vs. upstream.

[veerababumanyam]

can someone please suggest me on preview. I tried all possible ways mentioned but not able to get preview. Windows, docker, multiple LLMs like OpenAI, Anthropic, Ollama with wen, phi, deepseek, llama. Did max tokens setting as well but not able to get preview. tried reinstalling but no help. It would be great if someone suggest how to get a preview. Thanks in advance

[chrismahoney]

If you’re running a small model, try to verify that preview is intact by running a prompt such as ‘Create a basic Hello World page using HTML5’. I know it’s simple, but if you’re having issues related to the context length of your model, that prompt fits within the context window and should result in a functional preview.