Integrations
Integrations are the means by which an Agent can be connected to the real world.
There are two basic Integration Types:
-
Content Retrieval/RAG Integrations
-
Input & Output Integrations
Content Retrieval & RAG Integrations
Content Retrieval & RAG Integrations retrieve chunks of text or documents that are injected directly into the LLM prompts, providing contextual data that allows the LLM to respond using information relevant to the task at hand according to the requirements of the Agent.
-
RAG Retrieval is a common technique for performing vector similarity search on a repository of documents or text chunks and retrieving the most relevent documents compared to the user's query.
iostack allows almost any RAG respoitory to be exploited in your Agents via canned RAG Integrations and Custom Content Retrieval Integrations.
-
Content Retrieval Integrations allow other non-RAG document retrieval repositories to be used.
Content Retrieval & RAG Integrations can be specified to be called in specific Stages - or for every Stage.
They are called after the user's response has been submitted and before the prompt is assembled before sending to the LLM.
Input & Output Integrations
In contrast to Content Retrieval & RAG Integrations, Input & Output Integrations enable the sending or retrieval of Agent state (variables) to the real world.
Examples might include:
-
Creating notifications that a user has completed a task orchestrated by the Agent.
-
Updating a business SaaS service with the results of a survey conducted by the Agent
-
Retrieving context dependent data - for example, retrieving details about a user's account so that the author can integrate the details into prompting.
-
Etc.
Input & Output Integrations can be configured to be called according to the Agent's navigation dynamics i.e. when the conversation leaves or enters a Stage or when a particular Transition is taken.