Working Around VS Code APIs to Render LLM Suggestions (docs.getpochi.com)

Pochi's NES (Next Edit Suggestion) model has achieved a significant milestone by predicting timely edits in the VS Code editor. The challenge lies in how to present these suggestions to developers without interrupting their workflow. Built as a native VS Code feature, NES must leverage the editor's public APIs while addressing the unique demands of a continuously changing programming environment. This complexity requires a careful balance to ensure that AI-generated suggestions are actionable yet unobtrusive. To tackle this, NES employs a Dynamic Rendering Strategy that adapts to various editing scenarios. Suggestions that align with the current cursor position are rendered inline to blend seamlessly with the user's typing. For off-cursor suggestions or substantial multi-line edits, NES uses inline diff decorations or floating previews, respectively, ensuring that the developer's focus remains intact. This approach necessitates an advanced rendering pipeline involving theme matching and syntax highlighting, ultimately generating visually consistent previews without shifting the cursor or disrupting the user's context. By honing the presentation of AI suggestions, NES enhances the coding experience, ensuring that developers can leverage AI assistance effectively while maintaining their concentration.