Make knowledge available to your AI Agent

Large Language Models excel at generating natural-sounding language and plausible but not essentially correct answers. But they often struggle to incorporate domain-specific knowledge. To address this, it’s a best practice to enrich an AI Agent’s context with relevant knowledge, a process commonly referred to as Retrieval-Augmented Generation (RAG).

Cognigy.AI enables this through its Knowledge AI feature, accessible via the sidebar. To integrate this capability with an AI Agent, follow these steps:

1. Ensure you meet the prerequisites.
   Make sure you have both a model for embedding generation and a model for language generation provisioned (see “Prerequisites: Set up your AI Agent’s brain” for details).
   
   
2. Create a knowledge store in Knowledge AI by simply uploading your data, such as a PDF file.
   Knowledge AI ingests the document by breaking it into smaller chunks and associating them with their semantic meaning for more efficient retrieval. For more details, see Understanding RAG and Embeddings. Here is an example PDF file.
   
   
3. Assign the Knowledge to your AI Agent.
   You can assign a knowledge store to an AI Agent directly in its Knowledge & Instructions section in the AI Agent configuration (not the AI Agent Node). This will make the knowledge available to any job the AI Agent is assigned to.

Additionally, you can customize the knowledge processing for each AI Agent Node in the Grounding Knowledge section. Here, you can choose whether the knowledge is considered on every turn or only when the large language model identifies a need for it. This section also allows you to combine the AI Agent’s general knowledge with job-specific knowledge.

Now you can test your AI Agent by asking questions related to the document. For example, in our case, you might ask, “What are you doing for the environment?”.

➡️ To improve the experiences, let's give your AI Agent memory.