Skip to content

v0.19.0

Choose a tag to compare

View all tags
@getrapidkit getrapidkit released this 18 Apr 11:39
· 141 commits to main since this release
v0.19.0
cd8a165

Release Notes — v0.19.0 (April 18, 2026)

✦ AI Streaming UX + Model Selector

Summary

This release fixes the core AI streaming experience and gives users control over which model answers their queries.
Responses now appear token-by-token in real time, a thinking indicator covers the context-scan wait, and a compact model selector in the modal header lets users pick Claude, GPT, Gemini, or any other Copilot-registered model on the fly.

Added

  • AI model selector (inline in modal header)
    A compact <select> dropdown appears in the top-right of the AI modal (before the close button) when any Copilot language models are registered in VS Code.

    • Defaults to Auto (respects workspai.preferredModel workspace setting)
    • Premium Copilot users see their full model roster; free users see their available models
    • Selection is remembered per session and sent with every query
    • Disabled (grayed out) while a response is streaming

  • Thinking indicator
    Animated three-dot bouncing indicator shown between query submission and the first token arriving.
    Replaces the empty modal body + blinking cursor that made users think the extension had frozen during the context-scan phase.

  • MarkdownRenderer component (webview-ui/src/components/MarkdownRenderer.tsx)
    Zero-dependency markdown renderer for AI responses supporting:

    • Headings (H1–H3), bold, italic, bold-italic
    • Inline code and fenced code blocks with language label
    • Ordered and unordered lists
    • Horizontal rules, paragraphs

Fixed

  • Real-time streaming — "all at once" delivery resolved
    The extension host previously called postMessage hundreds of times per second inside the same event-loop tick. VS Code's IPC layer batched all of these into a single delivery, so the webview received nothing until the stream finished — then the full text appeared at once.
    Fixed by introducing a 50 ms setInterval flush loop in welcomePanel.ts: tokens accumulate in a string buffer and are delivered as one message per interval, giving the webview ~20 smooth updates per second.

  • Streaming render overhead eliminated
    MarkdownRenderer previously ran parseBlocks() (an O(n) operation) on every requestAnimationFrame during streaming, creating an O(n²) bottleneck that worsened as the response grew.
    Fixed by rendering the raw content prop directly during streaming (no parse, no internal state) and performing a single parseBlocks pass after isStreaming becomes false.

  • Menu item console warning
    workbench.actions.treeView.rapidkitProjects.collapseAll was referenced in contributes.menus but not in contributes.commands, producing a VS Code activation warning.
    Fixed by removing the manual menu entry and enabling "showCollapseAll": true on the rapidkitProjects view definition — VS Code renders the collapse button natively with no registration required.

Changed

  • Quick-prompt suggestion chips are now hidden while the modal is in a streaming or thinking state (they were showing as disabled, adding visual noise)
  • listAvailableModels() added to aiService.ts as a public API, used by the model selector to enumerate registered LM chat models
  • streamAIResponse() accepts an optional preferredModelId parameter; when provided, the exact model is used instead of the workspace preference

Links

Assets 3
Loading