Introduction to Agents in Dataiku

While the primary task of LLMs is to generate text, this ability to generate text, including structured output (JSON), can be used for building much more advanced applications. In these applications, the LLMs are used in order to achieve tasks beyond “answering based on knowledge”, but leveraging data, and even performing actions.

In these applications, the LLM usually acts as a central “brain” that leverages “tools” to do its work. This kind of AI-based application is often called an Agent.

Types of agents

Dataiku DSS has the ability to create “Simple Visual Agents”, “Structured Visual Agents” and “Code Agents” for various tasks.

In Dataiku Agent Management, however, you will primarily be using “External Agents”: surfacing agents from the 3rd party platforms into the Dataiku Agent Management Flow.

Using agents

Once the agent has been imported in the Dataiku Flow, you can leverage it within the Flow.

In Dataiku, every agent is also available as a “Virtual LLM” in the LLM Mesh. This lets you use it wherever the LLM Mesh is supported.

The Agents system is fully integrated with the rest of the LLM Mesh, including auditing, security, and Guardrails.