AI-SWEET: An Ollama + Open Webui + profile-toggleable-goodies Docker Compose for Windows + NVIDIA

===============

Overview

• This is for local, on your own, private computer. Lots of bad api-key practices here. This is like lab equipment atm.
• AI-SWEET is a "Docker Compose up" minimal default, local ollama-openwebui-flowise docker system for feature extension and access.

Running AI-SWEET

To start the services, run the following command:

git clone https://github.com/Martin-Stankard/AI-SWEET.git
cd ./AI-SWEET
docker-compose up

Point your browser to http://localhost:8080. Flowise is at http://localhost:3000.

Pipelines

Powerful, dangerous stuff as it executes user input raw python code.

To configure:

1. make sure the pipelines service is running via docker ps
2. Navigate to the Admin Panel > Settings > Connections > OpenAI API section in Open WebUI.
3. Add new API URL: http://host.docker.internal:9099 and API key: 0p3n-w3bu!. Test button available under settings, should be green

Your pipelines should now be active. Load into admin settings > pipelines.

• ./Pipelines/filterPipelineHelloWorld.py: This offers good feedback in pipelines services log AND baseline OpenWebUI llm use.
• ./Pipelines/flowise_agentflow_test001.py: Find and change API_URL ( from flowise flow embed dialogue ) and self.name, to mathch a flow named "test001". Select in open-webui, needs a Flowise project to work

Flowise

Flowise http://localhost:3000.

• I recomend this to continue: https://youtu.be/9TaRksXuLWY?si=PXheDAWPwM2wB-99 ( AWESOME Flowise training, vast subject )
• Immediate open web-ui test project, import ./Flowise/agentflow_test001 Agents.json for immediate pipeline feedback ( it just analyzes whether your question requires "search" or "no search". TODO, clean up logging/use autocomplete in webui. )

qdrant TODO, no api key rag vector db, hopefully.

• for vector embeddings/fancy rag
• http://localhost:6333/dashboard#/collections

postgres and pgadmin

• for record manager on flowise knowledge base "upsert"
• http://localhost:8888/browser/

Add Google Search via Open Web-UI settings (same credentials inside flowise)

To add Google search functionality:

1. Start here: Google Programmable Search Engine
2. Create a new search engine
3. Navigate to the Open-Webui Admin Panel > Settings > Web Search
4. Enable "Web Search Engine"
5. From the Google search engine you created, get the "Google PSE Engine Id" and the "Google PSE API Key". (Keep that secret)
6. Save settings
7. These seem to expire and have usage limits, so consider early if debugging.
8. Same credentials for a flowise Google custom search tool.

IMPORTANT: In the Open Web UI, normal chat interface, there is a '+'. Choose Web search there.

TODOs

• simple llama vision agent, run as: python PictureNamer.py -f < some directory full of pictures with messy names >; -- rename image files -- accompanying namer.config contains name - vision plain text prompt. -- liking a

• better tts, kokoro82m out of open webui is feeling like a need

• 1000 points of a new cloud app, software creation checklist