LLM App Architectures

There are multitude of architectures or pipelines for LLM applications. We discuss here the main ones.

​

The Single Prompt Architecture

This architecture is the simplest. The LLM application is a simple wrapper around one prompt / LLM call.

In agenta you can create such LLM apps from the UI. Or you can use your own code in case that your model is not supported (or you would like to add some custom logic for pre-processing or post-processing the inputs).

​

The Chain-of-prompt Architecture

The chain of prompt architecture as its name suggest is based on calling an LLM and then injecting the output into a second call as shown in the figure.

​

The Retrieval Augment Generation Architecture

​

The Agent architecture

​

Chat vs. Flow