Release v0.39.0

[wladimiiir]

Highlights

📖 Enhanced file_read Power Tool: Smarter Context Management

The file_read power tool has been significantly enhanced to give your Agent more precise control over file content. It can now read only specific parts of a file and display line numbers, optimizing the context window by providing the AI with only the most relevant information and improving its capabilities when analyzing code.

🛡️ Fine-Grained Control for bash Power Tool: Auto-Approval & Denial Patterns

Take full control over your Agent's shell commands! The bash power tool now includes new settings to define patterns for commands that are automatically allowed or always denied. This allows you to build a trusted execution environment, giving you greater confidence in the commands the Agent runs, even when in auto-approve mode.

✅ Task-Level Auto-Approval for Flexible Workflows

We've moved the auto-approve setting from the Agent profile to the individual task level. This means each task can now have its independent auto-approval configuration. Some tasks can be fully automated for hands-off execution, while others can remain under close watch, offering maximum flexibility for diverse workflows.

🗄️ Archive & Search Tasks: Organize Your Project History

Keep your task list tidy without losing valuable work! You can now archive tasks that you want to keep for future reference but don't need in your active list. Access them anytime, and use the new search functionality in the task sidebar to easily find any task you're looking for, making project history management a breeze.

⚙️ Disable Streaming for Improved Provider Compatibility

Some local or specialized providers don't work optimally with streaming responses. We've added a new option to disable streaming per provider, and even per specific model. This ensures smoother and more reliable interactions with a wider range of AI services.

⚡ New Provider Support: GPUStack & Enhanced OpenAI Compatibility

AiderDesk continues to expand its ecosystem:

• GPUStack Provider: We've added support for GPUStack, allowing you to leverage powerful local GPUs for running models and seamlessly connecting to AiderDesk.
• OpenAI Compatible Reasoning: For those using OpenAI-compatible providers, we've added reasoning effort support, enabling them to utilize this advanced capability for better AI performance.

🚀 Performance Boosts for a Snappier Experience

Expect a faster, more responsive AiderDesk! We've implemented several performance improvements:

• Optimized Message Loading: Faster loading of large numbers of messages.
• Enhanced Autocompletion: Optimized autocompletion, making the prompt field feel quicker and more fluid.
• Faster All Files Loading: Loading of all files is now more efficient.
  These ongoing optimizations contribute to a smoother and snappier overall working experience.

Other Changes

• Improved agentic flow for providers that stop right after a tool call, ensuring smoother operations.
• Fixed occasional white screen errors when using the compacting feature.
• Provider overrides in settings now accurately reflect the current state of provider settings.
• Custom commands now work properly in Agent mode when used within the browser.
• Improved auto-scrolling behavior while responses are being generated.
• Added missing provider parameters when generating commit messages.
• A new task branch is now created automatically when using Git worktree functionality.
• Added a copy button to the subagent prompt for convenience.
• Dropping files now clears them immediately instead of waiting for an event from Aider.
• Fixed provider options handling when multiple OpenAI Compatible providers are configured.

💡 We’d love to hear your thoughts! Your feedback helps us improve AiderDesk - feel free to share any suggestions or issues. Happy coding! 🚀

---

This discussion was created from the release 0.39.0.

[fredkzk]

Regarding the "Enhanced file_read Power Tool", Zed AI has a nice, easy feature where one can just select a code block in the editor, then use the @selection command in the message area to feed the exact code context to the Zed agent.

Are you considering down the road something similar to Claude Skills, specifically the YAML front-matter (name, description, optional allowed-tools) and a body that is only read when relevant?