0.16.1

Added

• User context to draft management for improved conversation handling

Changed

• Extracted web-fetch to standalone package for reusability

Fixed

• Deduplicate outputs by tool_call_id or name to prevent double rendering during streaming
• Tool output requiring double-click to expand now expands on single click

Full Changelog: v0.16.0...v0.16.1

0.16.0

Added

• Explicit OpenAI provider types: openai-responses for the official OpenAI Responses API and openai-completions for OpenAI-compatible providers like OpenRouter or local LLMs
• Periodic log cleanup for upstream logs in non-test environments

Changed

• Replaced URL-based provider heuristic with specific provider type configuration

Fixed

• Error message no longer persists when changing conversations
• Error message clears without requiring user interaction

Removed

• Unused quality_level parameter
• Unused dependencies and exports

Full Changelog: v0.15.3...v0.16.0

0.15.3

Added

• Reddit content fetching via JSON API for improved reliability

Fixed

• Draft content now persists when switching conversations or starting a new chat
• Stale cache and data leakage issues that occurred when switching users

Full Changelog: v0.15.2...v0.15.3

0.15.2

Fixed

• Fixed tool call ID collision between users in multi-user environments
• Fixed issue where different users could not reuse the same provider name or ID
• Fixed Gemini provider not fetching model list in background
• Fixed hung providers blocking background refresh or batch model list calls
• Fixed code blocks inside thinking blocks causing premature closure of thinking block
• Fixed session index drift with new migration consistency test

Changed

• Cleaned up config.provider design for improved maintainability

Full Changelog: v0.15.1...v0.15.2

0.15.1

Added

• Customizable storage keys for favorites and recent models in ModelSelector and JudgeModal
• Bias-mitigation masking mechanism for the model evaluation (judgment) feature

Changed

• Refactored comparison feature into ModelSelector using contextual actions per row for improved UX
• Decoupled judge model selection from the primary chat model with persistence support
• Judge model selection now orders selected model to the top for easy recognition

Fixed

• Title generation request being called many times in comparison mode
• Cannot change provider in judge model selector
• Cannot use Gemini provider models as judge
• Type errors
• API integration for better maintainability

Full Changelog: v0.15.0...v0.15.1

0.15.0

Added

• N-way judge evaluation for comparing responses from multiple models with configurable judge models
• Real-time conversation title updates for new and existing conversations

Changed

• Improved backend test coverage with additional test cases
• Improved frontend test coverage with additional test cases
• Enhanced useChat hook for better maintainability and reliability
• Refactored MessageList.tsx component for improved code quality
• Optimized mobile UI header row layout
• Judge evaluation now uses actual model names instead of generic 'primary' label
• Removed category parameters from SearXNG search tool for better search result quality

Fixed

• Type errors in codebase
• Ensured messageId is correctly assigned in useChat hook for non-primary conversation cases
• ChatHeader dropdown styling for consistent behavior across breakpoints
• Reduced test output noise with --silent flag

Full Changelog: v0.14.1...v0.15.0

0.14.1

Changed

• Improved visual presentation of judge response display
• Updated delete icon to use consistent Trash component across UI

Full Changelog: v0.14.0...v0.14.1

0.14.0

Added

• Judge/Evaluation System - Compare model responses with automated judge model evaluation, scoring, and reasoning
• Custom Request Parameters - User-defined request parameters with multi-select support for advanced API configuration
• Usage Tracking with Timing Metrics - Comprehensive performance insights including prompt tokens, cached tokens, and timing data
• Judge Response Management - Delete judge responses from evaluation comparisons

Changed

• Message ID Protocol - Unified to use UUIDs consistently for assistant responses across frontend and backend
• OpenAI API Compatibility - Updated response_format parameter handling (moved to text.format for compatibility)
• Judge Response Format - Enhanced judge evaluation response structure for better display and usability

Fixed

• Custom Parameters UI - Improved width consistency in custom request parameter popup items
• Message ID Handling - Resolved issue where frontend mixed sequential (integer-based) and UUID formats in judge requests

Full Changelog: v0.13.2...v0.14.0

0.13.2

Added

• Clear button for custom parameters and tools to quickly reset configurations
• Copy button for custom parameters settings to duplicate existing configurations
• Auto-generated IDs for custom parameter settings for better tracking and management

Full Changelog: v0.13.1...v0.13.2

0.13.1

Added

• Show content of custom parameters on hover for better visibility

Changed

• Improved tooltip visual styling

Fixed

• Tooltips no longer remain visible after clicking buttons that open popups
• Corrected prompt_tokens calculation in timings to properly account for cached tokens

Full Changelog: v0.13.0...v0.13.1