Kilo Code 4.21.0 Released

Bringing all the latest features from Roo Code v3.15.5 with performance improvements, UI refinements, and bug fixes

We're excited to announce the release of Kilo Code 4.21.0! This update focuses on synchronizing with the latest improvements from Roo Code v3.15.5, bringing you enhanced performance, refined UI elements, and important bug fixes. As always, we remain committed to providing a comprehensive superset of capabilities that makes your coding workflow smoother and more productive.

These are just a few highlights of the improvements in this release. Read the full release notes for a list of all changes.

Fixes added by the Kilo Code Team

Fixed Slash Commands for Mode Switching

We've resolved an issue where slash commands for changing modes were not functioning correctly after adding the /newtask feature from Cline. This fix ensures you can seamlessly switch between different AI models and modes using simple commands, maintaining your productivity without disrupting your workflow.

Now, slash commands truly can be a feature-merge of Roo Code and Cline!

Below you can find all of the changes we’ve inherited from Roo Code and Cline this release.

\(\begin{array}{|l|l|} \hline \mathbf{\text{Kilo Code 4.21.0}} & \text{Up to Date With} \\ \hline \text{Roo Code} & \text{v3.15.5} \\ \hline \text{Cline} & \text{v3.13.2} \\ \hline \end{array}\)

Features from Roo Code v3.15.5

Improved Chat Performance and Rendering

The chat experience has been significantly enhanced with performance improvements for code blocks and overall chat view responsiveness. These optimizations, contributed by @KJ7LNW via Roo Code, make interactions with the AI assistant feel more fluid and responsive, especially during longer coding sessions with extensive code snippets.

• Faster rendering of code blocks in the chat interface

• Smoother scrolling and navigation through chat history

• Reduced resource consumption during complex interactions

Enhanced Orchestrator Mode

A critical bug that would cause hangs, particularly in orchestrator mode, has been fixed. This improvement ensures more reliable operation when working with complex, multi-step tasks that delegate subtasks to specialized modes.

• More stable performance when managing complex project workflows

• Reliable communication between parent and child tasks

• Removed unnecessary tool groups from the orchestrator mode definition

Optimized Gemini Integration

The Gemini experience has been significantly improved with updates to @google/genai (version 0.12) and enhanced prompt caching efficiency. These updates resolve streaming completion bugs and optimize token usage when working with Gemini models.

• More reliable streaming completions without interruptions

• Improved caching efficiency for both direct Gemini API and OpenRouter connections

• Added prompt caching to the Google Vertex provider (thanks to @ashktn)

Refined Terminal Experience

The terminal integration has received several important improvements, making it more reliable and performant when executing commands and capturing output.

• Fixed empty command bugs that could cause unexpected behavior

• More robust process killing for better resource management

• Added a fallback mechanism for executing terminal commands if VSCode terminal shell integration fails

• Direct controls to stop terminal commands from the chat UI

• Fixed stderr capture in execa-spawned processes

Improved User Interface Elements

Multiple UI refinements have been implemented to enhance the visual experience and usability across different screen sizes and themes.

• Adjusted chat view padding to accommodate small width layouts (thanks to @zhangtony239)

• Fixed padding and height issues when resizing the sidebar (thanks to @zhangtony239)

• Improved UI/UX of code snippets in the chat (thanks to @KJ7LNW)

• Enhanced auto-approve toggle buttons for high-contrast VSCode themes

• Improved support for multi-root workspaces (thanks to @snoyiatk)

Performance Optimization

Several performance enhancements have been implemented to make Kilo Code faster and more responsive, particularly during resource-intensive operations.

• Offloaded expensive count token operations to a web worker (thanks to @samhvw8)

• Improved the performance of mode switching (thanks to @dlab-anton)

• Optimized settings handling and configuration management

• Sound notifications now play only when action is needed from the user

Fun fact - the sound notification fix was contributed to Roo Code by Kilo Code team member @olearycrew. We hope this is one of MANY upstream contributions to come from Kilo to Roo.

Enhanced Provider Configuration

Additional configuration options have been added for various AI providers, giving you more control over your interactions.

• Added customizable headers for OpenAI-compatible providers (thanks to @mark-bradshaw)

• Added config option to overwrite OpenAI's API base (thanks to @GOODBOY008)

• Added reasoning effort setting for the OpenAI Compatible provider (thanks to @mr-ryan-james)

Improved Settings Management

The settings experience has been refined to make configuration more intuitive and reliable.

• Fixed importing and exporting of custom modes (thanks to @julionav)

• Allow Kilo Code settings to be imported from the welcome screen (thanks to @julionav)

• Made retries respect the global auto-approve checkbox

• Simplified and streamlined quick actions for a more intuitive experience

We're continuously working to enhance Kilo Code with each release, making it the most powerful and user-friendly AI coding assistant available. We greatly appreciate feedback from our community as we strive to deliver the features and improvements that matter most to developers.

Happy coding!