Changelog

The following log records the version update history and change details of FlowDown.

1.20 (200/201/202)

Chat Template System 📋

• Brand new reusable chat templates for workflow efficiency
• AI-generated templates from existing conversations
• Intuitive template editor with creation/deletion
• Quick preset selection when starting new chats

General Improvements

• Fixed missing entitlements for running efficiently on iOS/iPadOS
• Fixed on device visual language model support
• Added Gemma3 model support
• Added HTML preview for generated codes

1.19 (190)

• Added new menu items
• Added welcome log for first-time app launch
• Improved OpenRouter integration
• Fixed issue where input box couldn't auto-focus when creating new chat on Mac
• Improved app's background processing capabilities
• Fixed issue where reasoning content wasn't updating in some cases
• Fixed logical errors related to model tools

Thanks to the following contributors:

• @at-wr
• @Zach677

1.13 (130)

• Fixed issue where keyboard shortcuts for copying text in code view were ineffective
• Fixed incorrect cursor display when selecting text
• Fixed unresponsive drag gestures on first and last lines of text
• Fixed code syntax highlighting not being applied
• Fixed reasoning requests being added to auxiliary tasks
• Added option to use developer role when needed
• App can now be opened via flowdown:// protocol
• App can now open model files in place

1.11

• Fixed incorrect model identifier when copying models
• Fixed mathematical content not rendering correctly
• Fixed menu disappearing when right-clicking conversation list
• Fixed documents not being indexed after web search

1.10 (101/102)

• Improved visual task scheduler with added option to skip auxiliary tasks for vision-native models
• Fixed editor lag when processing large images
• Fixed editor inability to restore editable content
• Improved auxiliary task structure and enhanced execution speed
• Improved data flow for web search when tool calling is enabled
• Fixed missing reasoning content in image recognition tasks
• Added support for OpenRouter model inference tokens

1.9 (90)

• Added preliminary support for mathematical content
• Support for LaTeX format mathematical formula display

1.7 (70)

• Fixed Qwen3 local model support issues
• Updated MLX model manifest structure to fix compatibility issues
• Local vision functionality temporarily unavailable, awaiting upstream fixes
• FlowDown is now fully open source, users can verify privacy protection commitments themselves

1.6 (60)

• Improved local model support
• Fixed multiple known issues

1.5 (50)

• Fixed misbehave on pasting content
• Fixed misbehave on web search phase indicator
• Fixed wrong return key title on input box
• Fixed a bug where multiple tool call was received at the same time
• Made support for connecting to local hosted LLMs via OLLAMA and LM Studio

1.4 (49)

• Fixed a crash at using location tool when location service is disabled

1.4 (47/48)

• Fixed issue where table border colors weren't updating correctly
• Fixed a crash when quickly deleting messages
• Allow sending empty messages when attachments are present
• Added ability to directly select text

1.3 (44)

• Fixed crash issues with certain reasoning models
• Fixed the issue where content were unexpectedly shown as reasoning
• Fixed calculation errors in chat list element positioning
• Fixed the issue where some model endpoints don't support usage statistics, causing unavailability
• Fixed the issue where crashes caused content not to be saved
• Fixed the issue where code blocks couldn't open details page in Hugging Face model download screen
• Fixed the issue where code blocks and tables couldn't use correct information in special cases
• Fixed logical errors that may occur when interrupting inference
• Fixed the issue of deleting error messages during retry process
• Disabled local model functionality for devices that don't support MLX
• Adjusted the order and logic of several menus
• Adjusted the logic for deleting conversation content
• Adjusted the logic for editing conversation content
• Added the option to send messages with Command + Enter
• Added the option for long press to create a new line
• Added the option to save reply content as images
• Removed the temporary chat option

1.2 (36/37/42)

• Added image export feature that allows exporting chat content as images
• Added ability to terminate ongoing conversations at any time
• Improved web search with deduplication algorithm for search results
• Fixed inconsistent text size rendering in supplement views
• Refactored code and resolved several known issues and bugs

1.1 (35)

• Fixed a bug that extracted image information may not being stored correctly.
• Fixed a bug that system prompt is not updated correctly.
• Fixed a bug where tool call may result server error.

1.0 (32/33)

First release to the App Store.

• Changed the software icon
• Fixed the issue where the DeepSeek model did not support out-of-order system prompts
• Fixed the issue where retries did not use the correct model
• Fixed the tool call failure issue, removed the minLength parameter that some models do not support
• Fixed the issue where the model did not use the latest time
• Adjusted the prompts used by the tool
• Added an option for 64k context length

1.0 (28)

• Supported cloud models can now use tools
• Added a right-click/long-press shortcut menu to the model selector
• Improved the status display of tool calls, some tools support viewing call details
• Added an option for 64K context
• Resolved the issue of application crashes caused by undo operations
• Optimized exception handling when the model is unavailable
• Fixed permission-related issues and improved error messages
• Optimized data export logic and logging system
• Improved the pre-dialogue check mechanism to improve stability

1.0 (20)

• Fixed menu event response issues
• Fixed code view height mismatch issues
• Fixed the issue where the model name was not updated as expected
• Optimized the long-press interaction experience of the conversation list
• Optimized the efficiency of the inference engine
• Improved the URL opening mechanism and added security warning prompts
• Significantly reduced network crawler resource consumption
• Replaced widget icons