v1.1.7

feat: add and optimize several features

• Add support for parsing Mermaid standard syntax output format
• Introduce configuration option for custom Ollama service interface address to enhance flexibility
• Synchronize and maintain the latest Ollama model library list

fix: resolve issues

• Display error messages when model invocation fails to facilitate troubleshooting
• Fix issue where the Ollama model storage directory configuration did not take effect immediately

v1.1.6

1. Fixed the issue where knowledge base document embedding would infinitely wait in certain scenarios.
2. Added a mechanism to detect the availability of embedding models.
3. Improved the PDF parser to enhance usability and better preserve formatting.
4. Added support for selecting the model installation location.

v1.1.5

feat: add new features and fix related issues

• Add intelligent agent functionality module

fix: resolve errors with non-VL models when attachments are images

• Fix issue where non-VL models throw errors when processing image attachments

v1.1.4

chore: optimize certain features and fix related issues

• Automatically use OCR model to extract text from images when uploading to non-vision models, then pass the context to the large model
• Fix issue where images failed to be passed to third-party models
• Fix failure in extracting DOC/PDF/MD documents caused by image extraction errors

fix: resolve document and image processing issues

• Ensure successful extraction of documents even if image extraction fails

v1.1.3

feat: add new features and enhance version stability

• Add conversation attachment functionality
• Add option for conversation without memory
• Introduce automatic context fine-tuning mechanism for ollama (2048-4096/<=4b 8192)
• Allow selection of third-party models and knowledge bases in sharing functionality
• Adjust knowledge base initialization interaction logic

fix: resolve multiple issues

• Fix ollama installation failure for certain installation paths
• Fix issue where close button does not fully exit the application on MacOS
• Fix PDF parsing issue in knowledge base on MacOS
• Update to the latest ollama version and model list
• Resolve other known issues

v1.1.2

• Add third-party API integration functionality
• Optimize default recall parameters for the knowledge base

v1.1.1

• Add knowledge base functionality
• Optimize scrolling overhead in conversation context
• Improve interaction experience
• Add dedicated prompts for the qwq model
• Update ollama model list

v1.0.7

1. Adjusted the behavior of the close button to directly exit the program (reverting the previous change).
2. Fixed the issue where some models installed via ollama pull could not be recognized, such as the latest version.
3. Resolved the issue where the model list could only be retrieved after restarting the process when the Ollama service was stopped.
4. Adjusted the automatic setting of the environment variable OLLAMA_HOST=127.0.0.1 for Ollama installations via AingDesk to prevent potential malicious exploitation.

v1.0.6

1. Optimized exclusive prompts for deepseek-r1.

2. Fixed the issue where the default backend language was English on macOS.

3. Improved the Ollama download mechanism with automatic node switching and resumable downloads.

4. Adjusted the behavior of the close button so that clicking it minimizes the program to the system tray instead of exiting.

5. Added a caching mechanism for web searches.

6. Adjusted the regeneration strategy for user queries during web searches.

7. Fixed the issue where the left and right spacing in the chat window was disproportionately large.

8. Resolved the issue where chat scrolling during the generation phase did not behave as expected.

v1.0.5

1. Optimize the internet search results.
2. Fix the issue of response language when using non - Chinese languages.
3. Fix the problem that the regeneration function doesn't meet expectations.