AICore

AI Core is a web application that provides a user interface for interacting with the Semantic Kernel, a framework for building AI applications. It's much like a simple ChatGDP client, it allows you to search for topics on the internet and then ask questions about the results. You can also upload documents manually or specify a url to load. Think of it as a ChatGDP light. It allows users to create, manage, and execute AI workflows using various AI models and tools.

All of the resources this project uses are opensource and free. So you don't need to get an azure account, etc.

Note:

I will warn you, if you don't have a decent GPU, gtx 3090 or greater the AI will perform very slowly. The better the GPU's and the more of them the better this project performs.

Also, this project assumes your connected to a domain and uses negiotiate authentication to identify the user. I plan to add Forms authentication at some point, but I haven't gotten to it.

What this thing can do?

1. Ocr Documents
2. Read the result to the user
3. Create Images
4. Search the internet
5. Generate computer speech
6. Handle timed reminders
7. Speech to text
8. And much more!

Resources used:

1. Ollama
2. ComfyUI
3. Postgresql
4. EntityFramework for Postgresql

Getting Started

To get started with AI Core, follow these steps:

1. Clone the repository:
2. Set up a SearXNG server. This is linux based but can be run in docker. I found a good article of how to set it up at https://docs.openwebui.com/tutorials/web-search/searxng It's important that you follow the directions because if you don't it won't return json which is required for the web search plugin.
3. Download Ollama and install it.
4. Start Ollama
5. Open a command prompt and type in "ollama pull granite3-dense:8b"
6. Once that finishes type in "ollama pull PetrosStav/gemma3-tools:12b"
7. Once that finishes type in "ollama pull llava:13b"
8. Ok, next step is to install PostgreSQL. I used the latest version which is 16.2. You can find it at https://www.postgresql.org/download/windows/
9. Make sure if your not using the docker build that you add in vector support. You can find the instructions at https://www.postgresql.org/docs/current/pgvector.html
10. Create a database called "semantic_kernel_web" and a user called "semantic_kernel_web" with the password "semantic_kernel_web".
11. Open the project in Visual Studio or your preferred IDE.
12. Restore the NuGet packages.
13. Download Comfy at https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_nvidia.7z 12a. Download the Checkpoint model at https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5/blob/main/v1-5-pruned-emaonly.safetensors
14. Open the folder you installed Comfy in and go to "Models\Checkpoints" and copy "v1-5-pruned-emaonly.safetensors" into it. This is the Model I use to generate images.
15. Start Comfy, if you have a gpu you can run "run_nvidia_gpu_fast_fp16_accumulation.bat" else you can run "run_cpu.bat". If you have a gpu and want to use it, make sure you have the latest NVIDIA drivers installed.
16. Update the config.json file with your PostgreSQL connection string,SearXNG server URL, comfy information and Ollama Url.
17. Run the application.

Welcome Screen

User Interface

Internet Search

Summarizing information it learned from the internet search