Skip to main content



An assistant is an AI model that you can create to power in-app AI experiences that are helpful, contextual, and relevant. Assistants are composed of sources and settings.

Assistants have a unique identifier which is how they are referenced to in your code.


Assistants can be configured with instructions which are additional prompts that can help with making the assistant more relevant and contextual to your usage. For instance, you might want your assistant to take a certain tone or reference your product a certain way.

Here's an example of instructions you might provide:

You are a helpful assistant for an app called "Super App". Super App is a robust email client for business professionals.

When responding, use a professional tone of voice and keep your answers short and concise.


Sources produce documents which serve as the basis for your assistant's responses. Think of them as source material that the assistant can cite when providing answers.

Sources can consist of crawled public websites or manually uploaded HTML documents. Your assistant can leverage multiple sources at once. After a source is crawled or HTML is uploaded, the result is a set of documents which is the actual underlying content that the assistant will leverage.


Every assistant comes with analytics which logs each query made to an assistant and the answer the assistant responded with in addition to any contextual inputs used (page information, visual screenshot, etc.).