Changelog
Stay up to date with the latest features and improvements in Donut AI.
This update focuses on enhanced security, improved transaction reliability, new TradingView auto-switch feature, and various performance and stability optimizations.
New Features
TradingView Chart Auto-Switch
When AI analyzes a token, the TradingView chart automatically switches to display the corresponding token's candlestick chart. No manual searching required - view technical charts instantly while chatting with AI.
Confidence-Based Hallucination Detection
The Critic Agent has been upgraded with a confidence scoring system that more precisely evaluates AI output reliability. The system assesses confidence levels for each key piece of information, helping you identify content that may need secondary confirmation.
History Search Pagination
Chat history search now supports paginated loading. Even with extensive history, search results appear quickly without lag.
Improvements (8)
- Transaction execution now waits for Solana on-chain confirmation, ensuring trades are truly successful before showing completion status
- Limit order amount calculations now use high-precision BigNumber to avoid precision loss on large trades
- Scenario prompts optimized to auto-check SOL gas balance when creating limit orders, with Gasless Swap guidance when balance is insufficient
- LP and DeFi operations now support smart balance-aware sequencing for logical operation order
- Added publisher confirms and timeout protection to RabbitMQ message queue for improved message delivery reliability
- Session search switched to cursor-based pagination for significantly improved query performance
- Added Langfuse debugging tools for better AI call chain tracing and analysis
- Feedback dialog data now integrated with PostHog analytics for better understanding of user feedback
Security Fixes (3)
- Fixed message API role parameter vulnerability preventing malicious users from spoofing system messages
- Fixed potential SQL injection risks - all database queries now use parameterized queries
- Removed unnecessary role parameter from AddMessage API to eliminate role spoofing attacks at the source
Bug Fixes (8)
- Fixed token marker position offset in report rendering - token information now displays correctly at corresponding positions
- Fixed terminal token info retrieval incompatibility with new backend API
- Fixed null value error caused by list-type content handling during message compression
- Fixed overly strict Critic Agent - now allows token names in strategies instead of requiring contract addresses
- Fixed error caused by sending empty content blocks to Anthropic API
- Fixed negative balance display caused by duplicate SOL balance updates
- Fixed TypeScript type checking control flow narrowing compilation errors
- Fixed risk metrics dashboard endpoint caching issue - now uses cache for faster responses
This update further enhances AI hallucination detection with contract address verification and tool name validation, while improving system observability.
New Features
Contract Address Authenticity Detection
The Critic Agent now includes FABRICATED_CA detection to identify AI-fabricated fake contract addresses. When an invalid contract address is detected, the system automatically flags it as a FATAL error and triggers regeneration, ensuring you always receive verified token addresses.
Strategy Tool Name Validation
The system now dynamically fetches valid tool lists from MCP servers to validate strategy tool names used in AI reports. Invalid tool names are immediately flagged and corrected, preventing execution of non-existent operations.
Data Collector
New data collection module added to support future data analysis and model optimization.
External API Request Logging
Backend now includes unified external API call logging for easier tracking and debugging of third-party service calls, improving system observability.
Improvements (3)
- Critic Agent regeneration strategy optimized - Round 2 regeneration now only triggers when invalid tool names are detected, reducing unnecessary wait time
- Added database search optimization indexes for improved query performance
- Improved DeFi position table layout with unified hover interaction styles
Bug Fixes (6)
- Fixed MCP server URL configuration and dynamic strategy checking issues
- Improved logging when fetching action tools for easier troubleshooting
- Clarified decimal format validation logic - human-readable formats (e.g., 0.1 SOL) are now correctly recognized
- Improved MCP tool error handling with more detailed HTTP error information
- Improved error log output format for easier debugging
- Added bun PATH configuration in pre-push hook for GUI applications
This update introduces AI hallucination detection, DeFi position management, and backtest history features to make your trading experience smarter and more efficient.
New Features
AI Hallucination Detection & Auto-Correction
We've introduced a new Critic Agent system that automatically detects inaccurate information in AI-generated content. When the system identifies false data, incorrect amounts, wrong prices, or invalid token addresses, it automatically flags and regenerates accurate responses. This enhances the reliability of AI-generated market analysis and trading recommendations.
DeFi Position Management
You can now view all your DeFi protocol positions directly in your wallet. The new DeFi tab supports expand/collapse functionality, giving you a clear overview of assets across protocols like Drift, Kamino, and Meteora. Locked amounts and wrapped token (wSOL) balances are also clearly displayed.
Backtest History
View and manage all your backtest results. Features paginated loading and detailed performance analysis to help you review strategies and continuously optimize trading decisions.
Improvements (17)
- You can now select your preferred AI model directly from the chat interface for better control over response quality and speed
- AI responses now display verification badges indicating information reliability
- Added average price line toggle in chart overlay menu for quick display of your position's average cost
- TradingView chart timeline now shows your local timezone, eliminating manual time conversion
- Friendly warning displayed when limit order price would result in immediate execution
- Auto wrap/unwrap SOL/WSOL support in Swap and limit orders, simplifying the trading flow
- Quote polling now displays estimated gas fees so you know the full cost before confirming trades
- Login status check before creating limit orders on charts to prevent issues from unauthenticated operations
- Feedback dialog appears when clicking the thumbs down button on AI responses to let you tell us what went wrong
- Trading panel height settings are now auto-saved, maintaining your preferred size on next visit
- Links in AI reports now open in new tabs without interrupting your current workflow
- Added support for "claimed" status for clearer subscription status feedback
- Improved wallet balance loading speed during page navigation
- DeFi position table layout optimized with unified hover interaction styles
- Improved analytics tracking for USD amounts
- Improved tracking for unlock button interactions
- Improved number display precision in trading components
Bug Fixes (14)
- Fixed occasional timing issue in chart price line rendering - price lines now display reliably
- Fixed AI drawings not re-displaying from chat history after being hidden
- Fixed rare bug where volume bars displayed at identical heights in certain cases
- Fixed technical indicator studyInputs configuration parameters
- Fixed incorrect HTML code block fence insertion position in Markdown
- Fixed unclosed HTML code blocks in AI reports - now auto-completed
- Cleaned up redundant HTML tags in report content
- Fixed inconsistent scrollbar styling in embedded iframes
- Removed auto-reserve SOL for gas logic to avoid user confusion
- Fixed insufficient balance error caused by AI rounding up initial amounts
- Fixed verified users still repeatedly seeing invitation dialog
- Fixed strategy/action tags in multi-line JSON format not parsing in list sub-items
- Fixed child element array filtering logic in tool renderer