Voiceflow

Voiceflow is an AI agent builder for support teams and agencies that run chat and phone agents, with usage billed in credits on top of a plan.

Voiceflow lets a company design, test and run AI agents that answer customers in a chat window or on a phone line. Voiceflow, Inc. develops it and sells it to support teams that want routine conversations handled without losing control of what an agent says. Building mixes two methods: structured workflows for steps that must happen in order, and LLM playbooks for open conversation. The Workflows canvas lays out step-by-step logic, and the Crew step lets several playbooks pass tasks to each other. The Function tool runs custom JavaScript mid-conversation, the API tool reaches outside services, and the MCP tool connects to MCP servers. A knowledge base answers from imported documents, and each playbook can override the project's default model. Tests replay simulated conversations with checks on routing and tool calls, LLM Evaluations score real transcripts, and Transcript Logs show what every step did. Engineers can manage agents from the vf CLI or the Build API instead of the editor. Customer experience and support teams are the main audience, especially where an agent must follow policy and still sound natural. Agencies are the second, with multi-client workspaces, white-labeling and client handoff tools for agents they build and pass on. Support staff edit instructions and tools in the same project that engineers extend in code. A team picks a plan and then pays for usage in credits spent on model responses and calls, adding editor seats, phone numbers and concurrent calls as it grows, while larger companies sign a contract through an account manager. Most deployments begin with support, answering questions from the company's documents and passing harder cases to a live agent in Zendesk or Genesys. On the phone, an agent answers callers and places outbound reminder calls on Twilio, Vonage or Telnyx numbers, started from a backend script or a Zapier step. The chat widget puts the same agent on a website. Before each release a team runs its tests, then reads the transcripts to see why a conversation went the way it did. For a support operation, the result is one agent that serves the website and the phone line and hands the difficult cases to a person.