You need the LLM model to create a recommendation service. It will look at the tickets you've already solved and make a recommendation for new ones. Let's look for a free way to get LLM tokens.

We used this one for our case, and you can use this one too. Feel free to use the Gemini Pro 2.5 Experimental (free) model. If you want better answer results, use any paid model.

OpenRouter offers many free models. You can select suitable one at: https://openrouter.ai/models?max_price=0.

OpenRouter can restrict access to some models, including free ones. In this case you will see messages describing these restrictions while working with API. If this happens, you can try to use another model or use the paid options.
You can find it on the OpenRouter base URL under "Code Examples"

You now have everything you need to create your future app. Let’s talk about how to build it. We've also preselected the embedding model for you – feel free to ask Junie what it is.

Ask Junie about any terms you're unsure of

What is an embedding model, and what is Hugging Face?

To test if it's working correctly, ask Junie for some information from this file. Do not forget to add task-description.md file to the context.

Adding files to context

Usually, Junie is quite good at navigating the project on his own but if you want Junie to be laser-focused, you can hit the "+" button to attach specific files or folders. This is how you give Junie access to custom parts of your project—like your README, API guidelines, or folder structures.

Adding files to context

You can attach:

Config or certain files;
Entire directories (e.g., /src/controllers);
Single files with tricky logic or expanded instructions.

This is a huge help in larger projects. Instead of explaining everything to Junie, just show it.

Test your task-description.md file

What is my LLM model?

RAG-Based Support Agent (Junie)

Stages