Skip to content

CHANGELOG.md

Latest commit

 

History

History
221 lines (176 loc) · 7.77 KB

CHANGELOG.md

File metadata and controls

221 lines (176 loc) · 7.77 KB
Raw

Changelog

All notable changes to the Dorian AI Agent Builder project.

[Unreleased] - 2026-02-09

🔧 Hotfix - Model Accuracy & Decommissioned Model Handling

Fixed

  • Decommissioned Model Error: Removed deprecated mixtral-8x7b-32768 from fallback models
  • Default Model Updated: Changed default from llama-3.1-70b-versatile to llama-3.3-70b-versatile (latest)
  • Auto-Cache Refresh: Model cache is automatically cleared when decommissioned model errors are detected
  • Better Error Detection: Added detection for model_decommissioned and invalid_request_error codes

Updated Fallback Models

Current supported models (as of Feb 2026):

  • llama-3.3-70b-versatile (default) ⭐
  • llama-3.1-70b-versatile
  • llama-3.1-8b-instant
  • llama3-70b-8192
  • llama3-8b-8192
  • gemma2-9b-it

Improved Model Filtering

  • Backend now filters out decommissioned models from API response
  • Models sorted with newest versions first (llama-3.3 → llama-3.1 → others)
  • Only shows LLaMA and Gemma models (excluded deprecated Mixtral variants)
  • Better error messages for unsupported/decommissioned models

[Version 1.0.0] - 2026-02-09

🧹 Cleanup - dorian-ui

Removed

  • Dependencies
    • Removed unused dependency: react-blockly@8.0.0
    • Removed unused devDependency: @types/react-dom@18.2.17

Added

  • devDependencies
    • Added eslint-plugin-react@7.33.2
    • Added eslint-plugin-react-hooks@4.6.0
    • Added eslint-plugin-react-refresh@0.4.5

Deleted Files

  • src/components/builder/BlockPalette.jsx and .css (unused component)
  • src/components/builder/WorkflowBlock.jsx and .css (unused component)
  • src/components/onboarding/index.js (unused barrel export)

🧹 Cleanup - Server

Removed

  • Dependencies
    • Removed unused dependency: uuid@13.0.0

Exports

  • Removed unused export SCOPES from src/config/gmail-config.js

🔄 K2 Think → Groq Migration

Removed

  • Completely removed all K2 Think AI provider code and references
  • Removed K2_THINK_API environment variable from .env
  • Deleted callK2ThinkAPI() function from workflowEngine.js

Added

  • Groq SDK Integration
    • Installed groq-sdk@0.37.0 (already present in package.json, now actively used)
    • Created callGroqAPI() function with proper error handling
    • Added support for dynamic model selection
    • Default model: llama-3.1-70b-versatile

Modified

  • src/services/workflowEngine.js

    • Replaced K2 Think API calls with Groq SDK
    • Added selectedModel parameter to workflow execution
    • Enhanced error messages to show selected model in logs
    • Added quota/capacity error detection and user-friendly messages
    • Added input validation for messages, temperature, and maxTokens

  • src/routes/workflows.js

    • Added selectedModel parameter to /execute endpoint
    • Added input validation for workflow execution requests
    • Created new /models GET endpoint to fetch available Groq models
    • Added fallback model list for offline/error scenarios

✨ New Feature - Groq Model Selection

Added Components

  • ModelSelector.jsx - Dropdown component for model selection

    • Fetches available models from Groq API
    • Caches model list for 1 hour (localStorage)
    • Displays current selected model in toolbar
    • Matches existing UI/UX design (maroon theme, rounded corners, shadows)
    • Supports keyboard navigation and accessibility

  • ModelSelector.css - Styling matching existing design system

    • Uses CSS variables from global.css
    • Supports light/dark themes
    • Responsive design for mobile/tablet
    • Smooth animations with framer-motion

Modified Files

  • BuilderPage.jsx

    • Added ModelSelector component to toolbar (before Load button)
    • Imported and integrated into builder interface

  • builderStore.js

    • Updated runWorkflow() to read selected model from localStorage
    • Added error handling for model quota errors
    • Displays helpful message to select different model on capacity errors

  • api.js

    • Added get() helper function for GET requests
    • Exported get in API object for use by components

Features

  • Model List Caching: 1-hour TTL to reduce API calls
  • Fallback Models: Shows common models if API fetch fails
    • llama-3.1-70b-versatile
    • llama-3.1-8b-instant
    • mixtral-8x7b-32768
    • gemma2-9b-it
  • Error Handling: Graceful degradation when API is unavailable
  • Quota Detection: Detects rate limit/capacity errors and prompts model switch
  • Persistent Selection: Stores selected model in localStorage (dorian_selected_model)
  • Visual Feedback: Shows loading, error, and fallback states

🛡️ Code Quality Improvements

Error Handling

  • React Error Boundary

    • Created ErrorBoundary.jsx component
    • Wrapped entire App with ErrorBoundary
    • Shows user-friendly error page on crashes
    • Displays error details in development mode
    • Provides "Reload Page" button for recovery

  • Server-side Validation

    • Added input validation to /execute endpoint

      • Validates code is non-empty string
      • Validates selectedModel is string (if provided)
      • Returns structured error responses with success: false

    • Added input validation to callGroqAPI()

      • Validates messages is non-empty array
      • Validates temperature is between 0 and 2
      • Validates maxTokens is positive number
      • Validates API response structure

    • Added null/undefined checks

      • req.session?.googleTokens (optional chaining)
      • error?.message checks throughout
      • response?.data validation in model fetching
      • completion?.choices?.[0]?.message?.content validation

Improved Error Messages

  • Model capacity errors: "Model X is out of capacity. Please select another model."
  • Invalid responses: Specific error for malformed API responses
  • Input validation: Clear messages for invalid parameters
  • Fallback behavior: Gracefully handles missing environment variables

📊 Test Results

Build Status

  • ✅ dorian-ui: Build successful

    • Output: dist/index.html (0.63 kB)
    • CSS: 102.23 kB (16.39 kB gzipped)
    • JS: 1,096.35 kB (304.67 kB gzipped)
    • No errors

  • ✅ server: Syntax validation passed

    • server.js: Valid
    • routes/workflows.js: Valid
    • services/workflowEngine.js: Valid
    • No syntax errors

Dependencies

  • ✅ dorian-ui: 329 packages audited

    • 2 moderate vulnerabilities (acceptable for dev environment)
    • All new dependencies installed successfully

  • ✅ server: 188 packages audited

    • 0 vulnerabilities
    • All dependencies installed successfully

📝 Notes

Design Consistency

All new UI components follow the existing design system:

  • Colors: Maroon primary (#6a041d), warm neutrals
  • Typography: DM Sans (body), Fraunces (headings)
  • Spacing: Consistent with --space-* variables
  • Shadows: Subtle shadows matching existing components
  • Borders: Rounded corners (--radius-lg, --radius-md)
  • Animations: Smooth transitions with framer-motion
  • Themes: Full support for light/dark modes

Breaking Changes

None. All changes are backwards compatible. Existing workflows will continue to work with the default Groq model.

Migration Notes

  • No action required for existing users
  • Default model (llama-3.1-70b-versatile) is automatically selected
  • Users can change model via toolbar dropdown at any time
  • Model selection persists across sessions (localStorage)

🔮 Future Improvements

  • Add model performance metrics in UI
  • Add cost estimation per model
  • Add user preferences for default model selection
  • Add A/B testing for model recommendations
  • Add analytics tracking for model usage
  • Implement backend database for model selection per user account

Generated: 2026-02-09 Contributors: Claude Sonnet 4.5