Continue v1.2.18 Changelog

This release is all about giving you more options for which AI you use, making the code-editing tools smarter, and squashing a bunch of annoying bugs. It matters most to developers who use Continue in their code editors or command-line tools, especially those who like to run AI models locally on their own computers.

What's new

• New AI providers: You can now connect to MiniMax and Tensorix as AI models.
• Command-line skills: In the command-line tool, you can now import and use custom "skills" (special commands that help the AI do specific tasks better).
• Save and load chats: You can now export your command-line chat sessions to a file and import them later, so you don't lose your history.
• Code review checks: The command-line tool can now look for custom rules in a .continue/checks/ folder when reviewing your code.
• Smarter autocompletion: Added a new template for "Qwen" models that helps the AI look at your whole project to suggest better code.
• Amazon Bedrock support: You can now use an API key (a special passcode) to connect to Amazon's Bedrock AI service.
• Better Gemini connection: Google's Gemini models can now connect through a new software bridge (the AI SDK), making them work more smoothly.

Improvements

• Updated default models: The app now points to newer, smarter models by default, like Claude 4.6, and removed some older Gemini models that are no longer available.
• Better error handling: When the AI connection drops or messes up, the app now recovers and shows errors in a friendlier way.
• Safer terminal commands: The app now adds a time limit to commands run in the terminal, so they can't run forever and freeze your computer.
• Clearer documentation: Added better guides for fixing SSL (security) connection issues and explained how the command-line tool's question feature works.

Fixes

• Python formatting: Fixed a bug where applying AI code suggestions to Python files would mess up the spacing and indentation.
• Terminal links: Fixed an issue where clicking web links in the terminal that had port numbers (like localhost:3000) would open the wrong web address.
• Terminal text: Removed a hardcoded Unix $ symbol that was awkwardly stuck in the terminal interface.
• Startup crash: Fixed a "No chat model selected" error that sometimes happened when you opened the app.
• Local model prompts: Strengthened the instructions given to local AI models so they apply code edits more accurately.
• Ollama tool calling: Fixed an issue where certain local models (like Mistral and Gemma3) couldn't use tools properly when run through Ollama.
• OpenAI errors: Fixed 400 errors (bad requests) that happened when using OpenAI's Responses API for reasoning and tool calls.
• Chat history bug: Fixed a problem where the AI would get confused if your chat history ended with "thinking" or "assistant" messages.
• Proxy settings: Fixed a crash that happened if your computer's NO_PROXY network setting wasn't a normal text string.
• Config files: Fixed the location where the command-line tool saves its config file, so it doesn't save to the wrong folder.

Breaking changes

No breaking changes in this release.