Agenta enables to experiment with the prompts and parameters for any custom application independent from the framework (langchain, llama_index, griptape..), the LLM model provider (OpenAI, HuggingFace, Self-hoster model) or the application logic.
The Agenta SDK is a wrapper around FastAPI. It does two things:
It allows you to create custom playgrounds connected to your data
It takes cares of setting and loading the configuration for your application.
The main idea about the SDK that an LLM application has two parts - the inputs and the configuration . The inputs are the inputs given by the user, these could be a message in the case of a chat application, a long text in the case of a summarization applicatoin.
The configuration are all the parameters that you need to iterate on. For example, the prompt(s), the LLM model, the temperature. In the case of a RAG pipeline, these would include in addition the chunking strategy, the type of embedding.
The idea behind agenta is that you specify in the code the configuration for your application. Then, you work in the playground on this configuration until finding a good one, which you save. At that point, you can use the configuration to run the application.
How to specify the configuration
To specify the configuration you need to
Was this page helpful?