Introduction to Workflows
Break down AI conversations into a visual workflow made up of discrete steps (“nodes”) and branches between them (“edges”).
Workflows is now available to all Vapi users in Open Beta on the dashboard here. Start building more reliable and structured conversational AI today.
Workflows is a new way to build conversational AI. It allows you to break down AI conversations into discrete steps, and then orchestrate those steps in a way that is easy to manage and modify.
Creating Your First Workflow
Begin by creating an assistant on the Assistants page and providing the required information, such as the assistant’s name and capabilities. Once your assistant is set up, switch the model provider to Vapi and click “Create Workflow” when prompted. A modal will appear offering you the option to create a new workflow or attach to an existing one. Choose the appropriate option to proceed to the Workflow Builder.
Step 1: Create an Assistant
Visit the Assistants page. Create a new assistant, give it a name, and select a voice and transcription model of your choice.
Step 3: Create a New Workflow or Attach an Existing One
Click the “Create Workflow” button. A prompt will appear asking you to create a new workflow by entering a unique title, or attach to an existing workflow.
Tips for Building Workflows
- Deleting Nodes and Edges: Click on any node or edge and press Backspace to delete it.
- Attaching Nodes: Attach a node to another by drawing a line from the top of one node to the bottom of another node.
- Save Requirements: A workflow cannot be saved until every node is connected and configured. The system will not allow saving with any dangling nodes.
- Creating Conditionals: To create conditionals, first add a condition node. Then, attach nodes for each branch by clicking the “Logic” tag on the connecting edges to set up the conditions.
Please let us know about any bugs you find by submitting a bug report. We also welcome feature requests and suggestions - you can submit those here. For discussions about workflows and our product roadmap, please join our Discord community to connect with our team.
Nodes
Workflows break down your AI voice agent’s behavior into discrete, manageable nodes. Each node encapsulates a specific function within the conversation flow. Detailed configuration options let you tailor each step to your requirements. The available nodes are:
Outputs a message to the user without expecting a response. Configure this node by specifying static text or providing a prompt for the LLM to generate dynamic text.
Collects input from the user. Define the variables by specifying a name, a detailed description of the expected input, and the data type (string, number, or boolean). Mark each variable as required or optional.
Makes calls to external APIs using GET or POST methods. Configure request headers and body, and define extraction rules to capture specific data from the JSON response. Optionally, enable asynchronous execution so that the workflow proceeds while awaiting the API response.
Transfers the active call to an external phone number. Ensure you provide a valid phone number in the configuration.
Terminates the call, signaling the end of the conversation.
Edges
Edges allow you to create branching paths in your workflow based on different types of logic:
For detailed configuration instructions and advanced settings, please refer to our dedicated documentation pages for each task.