Cline v4.1.9 Changelog

This update is mostly about fixing annoying bugs and making the app more reliable, especially if you use alternative AI models or leave the app open for a long time. It matters most to developers who use Cline heavily and need it to work without crashing or silently losing their login.

What's new

• The app now uses two new fonts: "Inter" for regular text and "Geist Mono" for code. This gives the whole interface a cleaner, more modern look.

Improvements

• If you use a custom color theme, the text shown when comparing code changes (a "diff") is much easier to read. Before, the color was hard-coded and could look washed out and invisible against certain backgrounds.

Fixes

• The "idle timeout" bug: If you left Cline alone for a long time, it would forget your login token, causing your next request to fail—even though your login was totally valid. It now correctly remembers your refreshed login.
• Silent logouts: If you ever got logged out unexpectedly (like if your login token was rejected), the app used to just erase your info and fail silently. Now, it actually tells you what happened.
• Fake success reports: When moving old tasks to a new format, the system used to spam its error trackers. It also used to report a "success" even if the move actually failed. It now waits to make sure the move fully works before reporting success.
• Claude Code model fixes (SDK bundle only): If you use the Claude Code AI model, it was basically broken for doing actual work. It now uses its own built-in tools correctly, starts in your correct project folder, and respects your permission settings.
• Broken AI commands: Sometimes, the AI tries to send a command (like "read this file") but the text gets cut off. The app used to try and guess the missing parts, which led to it doing the wrong thing. Now, it just rejects the broken command instead of guessing.
• Empty message crashes: Some strict AI models would crash if you sent them a blank message. This is now handled correctly.
• Mid-task crashes: Fixed a crash that happened when the AI tried to use tools out of order while using LiteLLM (a tool for connecting to different AI providers).
• Inflated token counts: The app was accidentally counting the same parts of your conversation multiple times, making it look like you used about five times more "tokens" (chunks of text the AI reads) than you actually did. The math is now correct.

Breaking changes

No breaking changes in this release.