Choose an agent mode

The agent builder presents three customer-facing choices. Choose based on how much control the workflow needs; provider names are configuration choices, not agent modes.

Builder choice	Good starting job	How it runs	Status
Free-flow — Standard Pipeline	Support answers with knowledge and HTTP tools	Speech recognition → text model and tools → speech generation	Supported
Free-flow — Speech-to-Speech	Natural low-latency conversation when native provider behavior fits	Native real-time speech model	Supported
Flow-based	A constrained sequence with branches and slots	Generated/editable graph and flow runtime	Alpha

Start with the standard pipeline when the agent must inspect structured tool outputs, apply detailed business instructions, or cite knowledge. Try speech-to-speech when natural turn-taking is the priority and the chosen provider supports the required behavior. Use flow-based Alpha for a workflow you can test node by node, including error branches. Do not assume an automated graph handoff connects a human to a phone call. Existing Concierge Voice is not one of these agent modes. It keeps your answer engine and only supplies browser speech input/output. Integration guide. The server also retains legacy/internal coordinators. They are not extra builder choices. Supported STT, TTS, model, voice, and language options depend on provider configuration and account access; confirm the actual combinations in the builder and test the chosen language end to end.

Overview

First conversation