Oracle APEX 26.1 and local AI Model

With the recent release of APEX 26.1 using AI is easier on APEX. I wanted to try it but I stil don't want to spend money on that, Im just learning to configure it so I don't want to spend money just by making a "Hello world!" test, if that's your case or you are worried about your privacy and want to run only local models this post is for you, we'll set up a local AI model, enable external access and configure it on APEX.

Computer Specs

The AI model you can run on your PC will depend on your hardware, these are my PC Specs so you can compare, maybe you can run a better model.
  • Processor: Intel Core Ultra 5 245K
  • RAM: 32 GB
  • Graphic Card: GTX 1050 Ti (4 GB VRAM)
  • Operative system: Windows 11 Home single language
As you can see, my graphic card is not the best but it works.

Installing Ollama

We'll use Ollama as our tool for AI, you can install it from here: 

After installing Ollama you need to download the AI model, there are many models available:

For My hardware, Gemini recommended llama3.1:8b (Link) or qwen2.5-coder:14b (Link), llama is not the smartest one but at least is fast and will work for testing that everything is working fine.
If you want to use the same one open a CDM and type
ollama run llama3.1
When it finishes you can test it by typing a message on the console, after testing close it.

Disable AutoStart

Ollama will start with windows automatically and there is no option on settings to disable this, you can open task Manager and go to Start up apps, find ollama and delete it from there, that way you will not load any AI model when you don't need it.

Enable external access

Now you have an AI model running on your computer, if you are running APEX on your computer you don't need this step, just use localhost:11434 when asked for the model URL. But if you are using APEX on the cloud or oracleapex.com you need this to make it available, the good news here is that you don't need to install anything, Pinggy will create the http tunnel for you with a single command. 

First, go to setting on Ollama and make sure the "Expose Ollama to the network" is selected





create a new file, call it pinggy.bat and insert this: 
@echo off
title Pinggy
echo Connecting to Pinggy...
ssh -p 443 -o StrictHostKeyChecking=no -o ServerAliveInterval=30 -R 80:localhost:11434 free.pinggy.io
pause
Line 4 makes the magic and would be enought, we are just creating a batch file so you don't have to copy/paste the command everytime you need it.

Execute the file, It will ask for a password, just leave it empty and press enter.

this will give you a URL to access to ollama services


Now you have the URL to access your service, but due to pinggy free service limitants this link will last only 60 minutes, after that time you will need to create a new one and update url on APEX, but for testing this might be enough, however if you need a permanent url you can buy a plan on piggin or try other similar tools.

Setting APEX

We have everything ready for APEX, on APEX 26.1 go to "Workspace Utilities > All workspace Utilities" select "Generative AI" and create a new service

 

Set these parameters
Parameter Value
AI Provider Ollama
Name Set a custom name for this AI service
Used by APP Builder Set Yes if you want to use this AI service on builder, i.e. help you to build a query
Base URL Set the URL created by Pinggy
Credential Leave it empty
API Key Leave it empty
AI Model your AI Model name, the same you used when installing the model i.e. llama3.1:8b

Before click on "Create" you can click on "Test connection", you'll see that your Pinggy logs this connection:


Now you can click on "Create"

APEX created a web credentials for this AI service, however this is linked to a URL for security, we have to change our AI service URL every 60 minutes so removing this validation will save us to update it here as well, if you want to remove it, go to "Workspace Utilities > All workspace Utilities" and select "Web Credentials", clear "Valid for URLs" item and save.

Testing

Now to test that everything worked go to SQL Commands, click on APEX Assistant and type a simple prompt:


------------------------------------------------------------------
We have received an answer from our local AI and now we can use it for development and test.... Let's celebrate!!!

Author

Comments

Post a Comment