Continue v1.2.14 Changelog

This release is all about making the AI assistant smoother and safer for developers. It matters most to people using the CLI tool or building automated agents, as it fixes several bugs where the AI might accidentally lose track of a conversation or try to change the wrong files. It also rolls out a ton of new guides to help users understand how to configure and use these new AI agents.

What's new

• Smarter AI safety rules: The AI is now guided to use safe editing tools rather than running raw terminal (Bash) commands, and it will actively avoid touching sensitive files when suggesting edits.
• Better context tracking: In the CLI, you can now always see a percentage showing how much of your "context" (the memory limit the AI can hold in a single chat) is being used.
• Cleaner "thinking" display: When the AI shows its "thinking" process before answering, that box now takes up less space on your screen.
• Failure reporting: There is a new tool that lets the AI report when a task fails, helping you figure out what went wrong.
• Grok upgrade: The integration for the "Grok" AI model has been upgraded to a faster version.

Improvements

• Less memory usage in JetBrains: The app now uses a programming trick called a "singleton" (a single shared instance) to handle data in JetBrains IDEs, which cuts down on how much computer memory it eats up.
• Better MCP server connections: MCP (Model Context Protocol) is a way for the AI to talk to outside tools. The app now passes all your computer's environment variables to these tools, instead of just a few, so they run much more smoothly.
• Cleaner chat history: The AI now removes old, already-applied rules from your chat history. This keeps the conversation uncluttered and saves context space.
• Helpful documentation: The guides now include YouTube videos, better icons, and a whole new section explaining how AI agents work.

Fixes

• CLI memory bleed: Fixed a bug where different CLI chat sessions would accidentally share information with each other. Now, each session is private.
• Lost code comments: Fixed an issue where applying an AI prompt would accidentally delete the helpful comments written in your code.
• VS Code rejection: Fixed a bug so you can instantly "reject all" AI changes in VS Code without a hitch.
• CLI infinite loop: Fixed a glitch where the CLI would get stuck in an endless loop trying to compact (shrink) messages. It now smartly prunes old messages until the chat is valid again.
• Mercury Coder glitch: Removed a weird text tag that was accidentally being attached to the front of code generated by the Mercury Coder model.
• Headless loading bug: Fixed an issue where the app would crash or fail to load when running in "headless" mode (without a graphical user interface).

Breaking changes

No breaking changes in this release.