Skip to content

Futurefeature.md

Latest commit

 

History

History
36 lines (26 loc) · 3.16 KB

Futurefeature.md

File metadata and controls

36 lines (26 loc) · 3.16 KB
Raw

CodeCollab - Future Features & Upgrades

This document outlines the planned upgrades for CodeCollab. These features will improve the user experience, optimize API costs, and ensure a highly resilient AI and code execution infrastructure.

1. Implement Resizable Sidebar (Room Settings)

Goal: Allow users to dynamically adjust the width of the workspace sidebar to accommodate different screen sizes and preferences.

Implementation Details:

  • Avoid Manual Logic: Instead of writing custom CSS/JS event listeners for mouse drag (which can be janky and break layouts), we will use the react-resizable-panels npm package.
  • Why this package: This is a highly robust, buttery-smooth library used by Vercel and Shadcn UI. It natively handles drag-to-resize, minimum/maximum width constraints, and snap-to-collapse functionality without layout shifting.
  • Impact: It will give the editor a much more premium, IDE-like feel (similar to VS Code).

2. Rate Limiting & AI Suggestion UX Improvements

Goal: Prevent abuse of the Code Execution engine and AI Suggestion API, while improving the user experience so they understand why a suggestion might take time.

Implementation Details:

  • Code Execution Rate Limits: Enforce strict token-bucket or sliding-window rate limits (e.g., max 5 executions per minute per user) to prevent server overload or malicious infinite loops.
  • AI Suggestion Debouncing & UX:
    • Currently, users might think their internet is slow or the browser is caching if AI suggestions take too long. We need to add visual feedback (like a subtle loading spinner or "AI is thinking..." ghost text in Monaco).
    • Add debouncing so it doesn't trigger an API call on every single keystroke (e.g., wait for the user to pause typing for 800ms before fetching a suggestion).
    • Remove Default Output: The system currently returns a default boilerplate return result when it shouldn't. This needs to be stripped out so the AI only returns genuine, context-aware code completion.

3. Highly Resilient AI API Fallback System (API Key & Model Rotation)

Goal: Ensure the AI features never go down due to rate limits by implementing a smart fallback system for both API keys and AI models.

Implementation Details:

  • Intra-Model API Key Fallback: Support comma-separated API keys in the .env file for a single provider (e.g., GEMINI_API_KEYS=key1,key2,key3). If key1 hits a 429 Too Many Requests error, the backend will catch it and automatically retry the same request with key2.
  • Cross-Model Fallback: If all API keys for the primary model are exhausted, the system will automatically cascade to the next available AI provider in the fallback chain.
    • Proposed Chain: Gemini ➜ Grok ➜ Claude ➜ DeepSeek.
  • Optimize Gemini Model for Suggestions: For real-time autocomplete, sending massive contexts to heavy models (like Gemini 1.5 Pro or 3.0) is too expensive and slow. We will configure the suggestion engine to use an older/cheaper and much faster token model (e.g., gemini-1.5-flash or the upcoming gemini-2.0-flash) which is specifically designed for high-frequency, low-latency tasks like code autocomplete.