Cline v4.1.4 Changelog

This release is all about giving you more control and making the AI less clumsy. It matters most if you use Cline for complex coding projects, because you can now safely let the AI investigate your code without worrying about it making changes before you're ready.

What's new

• Chutes support: You can now use Chutes as an AI provider (the service that powers the AI's brain).
• Better command menus: When you use slash commands (like / to do something), you'll see your "skills" and "workflows" listed together. If two commands have the exact same name, the menu will help you tell them apart instead of just hiding one.

Improvements

• You control the plan: Previously, the AI could decide on its own to switch from "Plan mode" (just thinking and researching) to "Act mode" (actually changing code). Now, only *you* decide when it's time to stop planning and start doing.
• Locked-down Plan mode: When the AI is in Plan mode, it's now strictly blocked from editing files. It can still look around and read your code, but it can't run commands that alter files, change Git history, or install new packages.

Fixes

• No more "fake" errors when planning: If the AI tried to do something in Plan mode but got blocked (which is supposed to happen), it used to look like a crash and tell you to "Retry." Now it realizes this isn't a failure, and clicking to Act mode works smoothly.
• Cleaner file paths: The chat used to show long, messy file locations (like C:\Users\Name\...). Now it just shows the path relative to your current project folder.
• No leftover images: If you attached an image to a previous task, it used to accidentally carry over to your next task. Now, starting a new task clears the slate.
• Clearer missing key errors: If you try to use an AI provider but forgot to enter your API key (the password that lets you use the service), you now get a clear message instead of a confusing generic failure.
• Live tool updates: If an external tool server updates its list of available tools, Cline now refreshes automatically instead of just showing a little pop-up notification.
• Real plugin names: Installed plugins used to all just show up named as "index." Now they display their actual, real names.
• Fixed Linux keyboard shortcut: The tooltip for Plan/Act mode had the wrong keyboard shortcut label for Linux users. It's fixed now.
• Smarter memory recovery: When the AI ran out of "context" (its short-term memory), it used to just crash with a raw error. Now, it tries to compress its memory and try again. If it truly can't, it explains why.
• Better empty response retries: If an AI provider sent back an empty message, Cline used to only retry this for one specific provider (Ollama). Now it retries for all of them, fixing crashes on OpenRouter, Cline, and OpenAI.
• Fixed newer Claude models: Claude 4.6 and 5.x models were getting rejected with a weird error about "thinking.type." This is fixed, whether you pick them from a list or type their ID manually.
• Restored Amazon Bedrock caching: A feature that saves time and money by "caching" (remembering parts of your conversation) was broken for Bedrock users. It's fixed now, and Bedrock models are routed more efficiently.
• Smarter model handling: For OpenAI-style setups, the AI now correctly sends a limit for "reasoning" models. Also, if you try to send an image to a model that can't read images, Cline will just remove the image instead of failing the whole request.
• Updated model lists: MiniMax now uses its correct default model. The built-in list of models was also updated to add two new providers: Infomaniak and SCX.ai.
• Less spammy error reports: If something goes wrong repeatedly (like an endless retry loop), the system now reports it once instead of spamming you with double error messages.

Breaking changes

No breaking changes in this release.