Document Retrievers
What is a Document Retriever?​
A document retriever is abacus own vector store. It allows you to embed documents and then retrieve relevant chunks of documents based on a query
Document Retrievers are only available in these use cases:
After navigating to any of the above project types, you can create a document retriever by navigating to the appropriate section:
Now click on the create button:
And select the appropriate feature group and any other options that you are interested in. Please note that the default options will work best, as Abacus apply it's own heuristics based on best practices and the actual shape of your data.
Note: The feature group needs to be of feature group type
List of Documents, else it won't be available in the dropdown menu
You can now use your document retriever in two ways:
- As part of a Custom Chatbot model training options
- As standalone deployment, via the API
The former is the most common reason to create a document retriever/vector store. The document retriever works in combination with a custom Chatbot to retrieve most relevant chunks of information and then pass them on to the LLM to answer the user's questions. This allows you to create powerful RAG applications.
Finally, you can also test some queries on the document retriever directly, by utilising the playground functionality:
