Feature Updates​

1. New Server Bot, supporting integration with third-party web apps, with a built-in sample app Whisper-WebUI for quick speech recognition and meeting transcription.
2. New Agent Bot, enabling Agentic Workflow to integrate different bots.
3. Bots can now connect to other bots, supporting prompt batching and Welcome messages.
4. Integrated n8n workflow automation platform.
5. Enhanced Function Calling, now supporting integration with external tools via MCP.
6. Added NIM Executor, enabling connection to models provided by NVIDIA.
7. Supports model inference acceleration via Qualcomm NPU.
8. New chart rendering tools: added support for Mermaid and Diagram Generator Executors.
9. Added a file management interface to manage all user-uploaded files (browse, delete, upload, etc.).
10. Updated message input box UI, now supports voice recording and sending.
11. New chatroom and bot search feature.
12. UI improvements: online model update and download; added “select all” button for group permission model selection; updated system settings interface with categorized views; improved bot categorization display in the Store.
13. Audio file links in chat now display as audio tag players.
14. Improved chatroom sorting options, allowing quick switch between sorting by time or model group.
15. New Construct Vector DB Bot, supports creating knowledge bases via web interface.
16. The Windows version now provides a more user-friendly offline/online installer and model downloader, with built-in Gemma 3 multilingual small model.

Other Updates​

1. UI redesigned for full-width layout, allowing up to 12 bots to be displayed simultaneously.
2. Added php artisan workers:start and php artisan workers:stop commands to control the task manager via CLI.
3. Improved the way php artisan commands update web config parameters.
4. When creating a new chat within an existing chatroom, current chat members are automatically selected.
5. Website translation fixes and refactored translation file structure.
6. Improved performance of the translation tool used for website language files, with auto-validation of output files to prevent format errors.
7. Editing a bot within a chatroom no longer redirects to the Store homepage.
8. Added RAG (Retrieval-Augmented Generation) library.
9. Painter no longer preloads models, reducing memory usage.
10. Improved stability of the Whisper executor.
11. SysInfo displays more useful system information.
12. API support for downloading and deleting Hugging Face models.
13. API support for managing Executors.
14. Kuwa API now supports passing Botfiles.
15. Unified naming conventions and parameter settings for models, tools, and bots.
16. The Windows version now supports CUDA 11.8 and above.

Bug Fixes​

1. Redis reset button not functioning
2. Terms of Service not displaying
3. Creation of default user API token and user upload directory
4. Email verification required even when SMTP is not configured
5. User sessions do not expire on multiple logins
6. Botfile auto-corrects comments to uppercase
7. Error importing initial bots in Docker environment
8. Error when importing bots with non-English names
9. Missing TAIDE model download link error
10. "Composer not found" error in Windows version — auto-generation of composer.bat

Feature Updates​

1. Consolidated offline English documentation #39
2. Support for direct integration with third-party APIs. The default model now includes Groq Llama3.1 70B.
3. A Pre-built Docker Image is provided for faster installation.
4. The Docker version directly supports CUDA, eliminating the need for repeated container installation.
5. Added Kuwa Javascript Library for calling Multi-chat API.
6. Ability to directly edit Bot settings by clicking the Bot image within the chat room.
7. RAG supports caching Embedding models and Vector DB.
8. SearchQA facilitates integration with third-party search engines.
9. DocQA/WebQA offers a Failback mode, allowing direct interaction with an LLM.
10. Pipes can now specify parameters from the prompt.
11. Added Media Converter for editing and splicing videos or audio.
12. The default Gemini model has been updated to Gemini 1.5 Flash.

Bug Fixes​

1. Docker WARN message "FromAsCasing: 'as' and 'FROM' keywords' casing do not match" #38
2. The Docker Version Executor diagram disappears
3. Chinese display garbled on the Windows version
4. The Chinese file path is not supported when creating Vector DB on the Windows version
5. The file name is too long and exceeds the path length limit on the Windows version
6. ChatGPT executor does not follow custom context_window

Feature Updates​

1. Added Pipe executor, which can execute programs (tools) within a specified directory, such as directly executing Python programs output by models via Python interpreter
2. Provided Calculator, Iconv and Python example tools that can be called via Pipe executor
3. Added Uploader executor to allow users to upload files to a specified directory, including tools, RAG knowledge bases, or website components
4. Supported Bot export and import, allowing export of Bot name, description, icon, and Modelfile as a single Bot file, similar to an application configuration file; installation can automatically import default Bot files
5. Allows users to choose the sorting method for Bots in chat rooms
6. Supported Bot icon replacement
7. Added Kuwa API server compatible with OpenAI API
8. Provided default examples for connecting to cloud multimodal APIs: gpt-4o-mini-vision, DALL-E, Gemini pro 1.5 vision
9. Supported setting the upper limit of uploaded files via Web interface
10. Supported installation and execution in environments with Web proxy within enterprises
11. Supported acceleration of model inference using Intel GPU
12. Added automatic installation and update scripts for Docker version, thanks to @wcwutw
13. RAG Toolchain default Embedding model replaced with Microsoft's intfloat/multilingual-e5-small model, licensed under MIT
14. RAG (DocQA, WebQA, SearchQA, DB QA) added display_hide_ref_content, retriever_ttl_sec parameters
15. Increased support for downloaded tools' default models, including Meta Llama 3.1 8B with Function calling and lightweight Google Gemma 2 2B

Bug Fixes​

1. #21: Docker version does not generate https:// links after a reverse proxy, thanks to @Phate334
2. #23: Two-minute timeout issue, thanks to @x85432
3. #24: Modelfile parsing issue
4. #25: Importing Prompts does not apply Modelfile
5. windows\src\tee.bat is misjudged as a virus
6. RAG reference data does not display original file names
7. Updated Windows version dependency download link

Feature Updates​

1. Customized Bot Permissions: Configure the Bot's readable and executable permissions at system, community, group, and individual levels
2. Customized Upload File Policy: Admin can set maximum upload file size and allowed file types
3. Tool Samples: Added samples for Copycat, token counter, etc.
4. Pre-defined Model Profiles: Provided profiles for LLaVA and other fine-tuned models
5. UX Optimization: Beautified icons and chat lists
6. Updated Default Models: ChatGPT Executor is connected to GPT-4o by default, Gemini Executor is connected to Gemini 1.5 pro by default

Bug Fixes​

1. File name with whitespace parsing issue when uploading
2. Language is not saved after logout
3. Dependency issue of Llamacpp Executor
4. Color and line breaks not supported in Windows version logs
5. The first message in the group chat is always sent even using multi-chat single-turn Q&A
6. Windows version DocQA default parameters may exceed the context window

New Tutorials​

Customizing RAG Parameters Tutorial: https://kuwaai.tw/blog/rag-param-tutorial
Customizing Tool Tutorial: https://kuwaai.tw/blog/rag-param-tutorial

Hi everyone, Kuwa v0.3.1 is out, and this update mainly focuses on multimodal input and output, which now supports both speech and images. Combined with the previously launched Bot system and group chat functions, this allows for practical functions such as meeting summaries, speech summaries, simple image generation, and image editing:

1. Supports the Whisper speech-to-text model, which can output transcripts from uploaded audio files, and features multi-speaker recognition and timestamps.
2. Supports the Stable Diffusion image generation model, which can generate images from text input or modify uploaded images based on user instructions.
3. Huggingface executor supports integration with vision-language models such as Phi-3-Vision and LLaVA.
4. RAG supports direct parameter adjustment through the Web UI and Modelfile, simplifying the calibration process.
5. RAG supports displaying original documents and cited passages, making it easier to review search results and identify hallucinations.
6. Supports importing pre-built RAG vector databases, facilitating knowledge sharing across different systems.
7. Simplified selection of various open models during installation.
8. Multi-chat Web UI supports direct export of chat records in PDF, Doc/ODT formats.
9. Multi-chat Web UI supports Modelfile syntax highlighting, making it easy to edit Modelfiles.
10. Kernel API supports passing website language, allowing the Executor to customize based on user language.
11. The Executor removes the default System prompt to avoid compromising model performance.

Hello everyone, after receiving feedback and suggestions from the community, we have launched the official version of kuwa-v0.3.0 to better meet your needs.

The main differences from the previous version, kuwa-v0.2.1, are the addition and enhancement of features such as Bot, Store, RAG toolchain, and system updates, as well as a new chat and group chat integration interface:

1. Bot allows users to create Bot applications with no code, and can adjust System prompt, preset chat records, User prompt prefixes and suffixes to implement different functions such as role playing, executing specific tasks, or using Ollama model files to build more powerful applications;
2. Store allows users to build and maintain their own shared Bot application store, and users can also share Bots;
3. RAG toolchain allows users to create their own vector databases by simply dragging and dropping local file folders, and then use the existing DBQA function to perform Q&A;
4. The new integrated interface not only directly supports group chats and single-model chats, it can also import Prompt Sets or upload files at any time, and can also be used for related RAGs;
5. Windows version adds SearchQA, which can be used to organize website Q&A by connecting to Google search;
6. Added Docker startup script to simplify Docker startup;
7. Executor can be directly connected to Ollama to use the models and applications supported by Ollama;
8. You can use update.bat to quickly update to the latest released version without re-downloading the .exe installer

Hello, friends of the community,

As planned, we have launched the kuwa-v0.3.0-beta1 version for you to experience in advance 😁

Hello everyone,

The TAIDE model released the Llama3-TAIDE-LX-8B-Chat-Alpha1 version today. Friends who use the Kuwa TAIDE version only need to update to the latest v0.2.1 version to experience the latest version of the TAIDE model. In addition to updating the TAIDE model, this version also expands the support for local models and fixes some minor problems, hoping to provide everyone with a better user experience.

Hello friends,

The TAIDE model was released today, and we are happy to release a customized Kuwa system for Windows that has the built-in TAIDE LX 7B Chat 4bit model.

Hello to our community friends,

After collecting everyone's feedback, we plan to roll out the long-awaited RAG feature in v0.2.0. The RAG part has been internally tested, so we are releasing v0.2.0-beta to invite everyone to test it out and see if it meets your expectations.
In addition, this update also provides a way to connect with TAIDE API and TAIDE models.
At the same time, we have also adjusted the system installation script and fixed some known bugs, making the entire system more stable, easier to extend, and easier to use.
If you have any suggestions or if you think there is room for improvement, please let us know!