Step 1: Create Assistant

Assistants are the core of Voiceflow. Each assistant is powered by the data you add to it.

Step 1: Create a New Assistant

In your workspace click the 'New Assistant' button on the top right. From there you can choose whether to build an AI Assistant or Design for an NLU Platform.

• AI Assistant: If you are deploying to a channel directly with our API, or want to use AI features in your assistant use this.
• Designing for NLU Platform: If you are working with an existing NLU platform (ie. Dialogflow, Lex, Rasa) and using Voiceflow just for design and export. Use this.

This walkthrough will focus on building and deploying an AI Assistant. If you are designing for an NLU platform. Visit our guide to building for export here.

Step 2: Upload data to knowledge base

When the AI Assistant is created, go to the Knowledge Base. Here you can click on 'Add Data Source' on the top right and add as many PDFs, URLs, or Text files as you want to power your assistant.

Once uploaded, your AI Assistant will use these documents to answer questions via GPT. The more documents you upload, the better the answers will be.

Note: There is a 10mb file limit per upload.

Once you've uploaded some initial documents, press 'Preview' on the top of the screen. This will allow you to ask the Knowledge Base questions and see what the responses are. Clicking sources will allow you to see where the answers are being pulled from.

Step 3: Add a step to the canvas

Designing on Voiceflow starts at the step-level, which includes adding, editing, and connecting steps/blocks on the canvas and creating paths in your assistant. Drag a step from this menu into your assistant.

The AI steps allow you to send prompts to different AI models like GPT-4 and Claude. You can specify if you want to use the knowledge base or not to answer the prompt.

Details on each step can be found here.

Step 4: Run a prototype

Once you have a basic assistant, run a prototype to make sure that it works the way you intended. If it isn't working the way you expect, check our guide to debugging here.