Background
As a Coveo MVP I was fortunate to have early access to the Coveo API beta. Early in development it was called the chunk API as that was the technical term for what you did with it. It essentially allows you to call an API with a query and get back the chunks of text from content you have in your Coveo index. This is the R in RAG (Retrieval Augmented Generation) pattern of Generative AI. Now that it is released it has an official name "Coveo Passage Retrieval API".
Why do you need it?
The passage API is helpful when you are building a custom application or UI that needs knowledge from your Coveo index of content. Say you have a PowerPoint file in SharePoint that has a case study with details on a project. You can ask an AI Chat UI or AI Agent a question and it can generate an answer based on this content. The API is available to get the text chunks or "passage" so you can send it to an LLM model like GPT-4o to create a response.
Why did Coveo create it?
Coveo already has an out of the box Generative AI solution called Coveo Generative Answering built into their search UI. They use AI/ML Models such as Semantic Encoder and Relevance Generative Answering to deliver an all in one solution that takes away the technical barriers to indexing, embedding, and prompt engineering. Soon after launching customers and partners were asking - what if I want to build my own front end solution but leverage all of this plumbing that Coveo built? The answer is the passage retreival API. You get all the benefits of Coveo (source Indexing with connections to any system, chunking and embedding into a vector index, vector search and enterprise search just to name a few).
How can I learn more and use it?
Here are the steps I'd suggest on your Coveo AI Journey:
- Start a free trial or create a new demo organization - https://platform.cloud.coveo.com/login
- Index content from the Sources tab
- Setup RGA (Relative Generative Answering)
- Setup Passage Retrieval API
- Check out the API Documentation
- Build a test application that calls the API