Kilo Code v7.4.23 Changelog

What's new

• Custom tool views: You can now choose in your display settings whether blocks for MCP (Model Context Protocol—tools that let the AI interact with outside apps) and other tools start open or collapsed by default.
• Better PR review comments: Reviewing GitHub pull requests got a major makeover. Instead of just dimming resolved comments, they now collapse into neat one-line summaries. You can easily see replies, send comments straight to your AI agent to fix, and even send all unresolved comments to the agent with a single button.
• Easier multitasking: When the AI delegates a task to a "subagent" (a smaller helper AI), you can now open and track that session in its own tab right next to your terminal windows.

Improvements

• Smoother project loading: When the Agent Manager opens, all your projects now start loading their data right away instead of just showing a spinning loading wheel until you click on them. If a project is still loading, it shows a clean "skeleton" outline instead of a single spinner, so multiple projects loading at once look organized.
• Better terminal capture: When you use the @terminal command to show the AI what's on your screen, it now grabs the text from the exact terminal you're looking at in the Agent Manager, instead of randomly pulling from an unrelated terminal.
• Smarter chat memory: The default speech-to-text model (the thing that turns your voice into text) has been upgraded to NVIDIA's Parakeet TDT 0.6B v3 for better accuracy. It still works offline, too.
• Longer-lasting terminals: Background terminals and nested AI sessions won't randomly close on you anymore. They now survive configuration reloads and your computer going to sleep. They only close when you actually want them to, or if a project gets deleted.
• Cleaner menus: The dropdown menu for picking an AI model is less cluttered. It now hides the provider's name (like OpenAI or Anthropic) in the compact view to save space, but keeps it visible when you fully expand the menu.
• Faster performance: We sped up how fast large chat sessions load, optimized how the app switches tabs, and reduced lag when the AI calculates timeline progress bars.
• Cleaner code diffs: Pull request comment diffs (the side-by-side comparisons showing what changed in the code) now use the Pierre-backed diff viewer and proper syntax highlighting, so the code is much easier to read.

Fixes

• Stable diff scrolling: When scrolling up through a large code review, the inline diff (the view showing code changes) will now stay in the right place instead of jumping around.
• GitHub branch fixes: If your local Git data is outdated and still points to an old main branch, the app will now smartly check the remote server to find the correct current default branch to use.
• Agent switching: When you switch from "Ask" mode (where the AI just answers questions) to "Code" mode (where it actually writes code), the app now keeps your selected AI agent and reminds it that it is allowed to write code again.
• Safe text deletion: If you typed extra text after a /memory show command (like /memory show my notes), the app used to accidentally delete your extra words. Now it keeps them safely in the chat box.
• Visible buttons: If an AI asks for permission to run a command and shows a huge amount of code or text, the "Allow" and "Deny" buttons will now stay on screen and scroll with the text, so you don't have to hunt for them.
• Accurate PR timers: Fixed an issue where wrong timestamps from GitHub checks would make pull requests look like they took way longer than they actually did.
• Better file suggestions: When you use @ to mention a file, the suggestions now show the most up-to-date files for your current workspace.
• Restored text streaming: Fixed an issue where the AI's text wasn't smoothly typing out across different chat surfaces.
• Git fetch compatibility: Fixed a bug where Agent Manager worktrees broke if your repository used a strict, restrictive Git fetch setup.
• Safe session imports: You can now safely import shared sessions using secure JWT (JSON Web Token) share links.
• Cleaner agent deletion: When you delete an AI agent, the app now makes sure it's completely removed from every possible configuration file, and it will tell you if something fails to delete.
• Smart worktree bases: When creating a new worktree (a separate working folder for your code), the app now correctly uses the branch you are currently looking at as the starting point.

Breaking changes

No breaking changes in this release.