iostack.ai Portal

Retrieval Augmented Generation (RAG) is a technique where a user's query can be used to retrieve semantically similar text chunks previously stored in an external service. This is useful for retrieving additional information to help the LLM form a coherent and accurate response to the user's query.
Content Retrieval is a technique where bulk textual information can be retrieved by calling external services, whether that service uses vector similarity or merely looks up data in a database. In a sense, RAG Retrieval is a form of Content Retrieval - and vice-versa - with RAG Retrieval merely being aided by the use of vector similarity lookup.
The Custom Content Retrieval/RAG Integration can therefore be used to integrate with any RAG repo as it integrates with the rest of the iostack platform in an identical manner to RAG integrations.
RAG and Content Retrieval Integrations are triggered at each turn prior to calling the LLM for the Stages that have been configured to use them.
Information gathered by RAG or Content Retrieval is injected directly into the LLM Prompt, differing from other 'input' integrations which retrieve information from external services and instead populate variables.
You can observe retrieved information in the LLM Prompt section of the Sandbox as you build and test your Agent.

We currently support specific Pinecone and Weaviate RAG methods that make some assumptions about the nature of your RAG repo and thus come with caveats.
If your RAG repo is not supported by the canned methods, you can create your own RAG Retriever by implementing a specific Custom JS Content Retriever Integration. If you have a spcific RAG repo or RAG configuration that you would like supported as a canned RAG Integration, please contact us.
For more information, view the documentation for each RAG Integration type and the information on authoring Custom JS Content Retrieval Integrations.

The Custom Content Retrieval/RAG Function Integration enables authoring of almost any retrieval method you can think of (RAG or Content Retrieval).
Custom Content Retrieval/RAG Scripts can make use of fetch functionality (for REST calls) and/or GraphQL functionality.

Defining Custom Content Retrieval/RAG Integrations

In much the same way that Variables need to be defined before they can be used, you will need to define a specific Custom Content Retrieval/RAG Integration before it can be used in your Agent.
This is to support configuring a single Custom Content Retrieval/RAG Integration with keys and other boilerplate parameters and then specifying the use of them throughout your Agent, customizing each use according to its location and context.
To define a Custom Content Retrieval/RAG Integration:
1. Navigate to the Integrations section of the Agent and click on the ’+’ icon at top right.
2. Select an Custom Content Retrieval or RAG Integration type and fill in the details.

Once defined, an integration will be available for selection in the Agent's list of integrations.
You can use the Integration definition and configure it for each Stage and Transition where you require it as well as in the Common Stage Settings of your Agent.