Skip to content

v1.2.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 12 Apr 19:02
· 116 commits to main since this release
v1.2.0
81096c8

Smoother Streaming, Custom Endpoints, and Enhanced Stability

We have completely overhauled the core engine in this release to make your AI coding assistant faster, more reliable, and much more intuitive.

User Experience Enhancements

  • Silky Smooth Chat: AI responses now stream seamlessly onto your screen. We implemented character-level throttling that eliminates UI flickering and layout shifts, making it much more pleasant to read as the AI types.
  • Transparent AI Actions: No more guessing what the assistant is doing behind the scenes. The chat status now explicitly shows real-time tool activity (e.g., "Writing file: app.ts", "Searching code", or "Running command").
  • New "Thinking" Controls: Easily adjust the AI's reasoning effort directly from the chat toolbar. Give supported models more time for complex architectural tasks, or speed them up for quick questions.
  • Auto-Accept Mode: Want the AI to just do its job? Use the new shield icon in the toolbar to instantly toggle auto-approval for all file edits and commands for your current chat session.
  • Easier File Attachments: A new attachment button in the chat input lets you quickly browse and attach local files and images without needing to drag and drop.

Custom Models & Endpoints

  • Bring Your Own AI: You can now configure multiple custom OpenAI-compatible proxy endpoints (like Ollama, LM Studio, or your enterprise server). This includes full support for custom HTTP headers, API keys, and specific model selection right from the settings panel.

Under the Hood (Backend & Core Architecture)

For those interested in the technical improvements, we have significantly upgraded how the extension communicates with the local server:

  • OpenCode SDK v2: Upgraded the core communication layer to the latest SDK v2. This brings dramatically better local server discovery, faster process handling, and eliminates "zombie" processes holding up ports.
  • Server-Side State Tracking: Session restoration, checkpoints, and "Undo/Unrevert" states are now tracked server-side rather than in local VS Code workspace storage. This guarantees that your conversation history and revert points remain perfectly consistent even if you completely restart your editor or the backend server.
  • Accurate Diff Calculations: File change statistics (lines added/removed) in the UI are now synchronized directly with the server's Git-level cumulative diffs. This eliminates the "double-counting" bug when the AI edits the same file multiple times in one turn.
  • Windows Path Resolution: Fixed a low-level issue where sessions would sometimes become invisible on Windows due to drive letter casing mismatches between VS Code and the local server storage.
Assets 3
Loading