Decagon Introduces Alto to Control Voice Agent Conversations

Decagon introduced Alto on November 11, 2026, a Decagon Labs model that controls the conversational layer of its duplex voice agent, deciding when to speak and what to say. The company reported that Alto lifted resolution rates by more than five percentage points on live customer calls while delivering average responses in about 700 milliseconds.

Decagon said the rest of the agent’s logic remained effectively unchanged during the resolution comparison. The announcement did not identify the customers, sample size, baseline rates, or measurement period for that result. Alto extends Decagon’s model specialization beyond speech quality to conversational control: deciding whether an immediate answer is justified and how to keep the caller engaged while a reasoning model finishes the work.

The Answer-or-Wait Decision

A conventional cascade voice pipeline handles each turn in sequence: transcribe the caller, reason about a response, then generate speech. Decagon’s duplex agent instead runs a fast conversational model in parallel with a more powerful reasoning model. The first listens and speaks; the second handles complex decisions and actions behind the conversation, allowing the agent to keep interacting while work is underway.

That parallelism creates a judgment call on every turn. Waiting for the reasoning model can disrupt the flow, but answering too soon risks an unconfirmed statement, such as telling callers their reservation has changed before the system has confirmed it. The speed advantage of a smaller conversational model therefore comes with a reliability problem that duplex architecture alone does not solve.

Alto continuously tracks what the caller has said and what the rest of the agent is doing, grounding its responses in a shared call record. It waits for confirmation before announcing that an action is complete. During a lookup or tool call, it can request the next piece of information or check in with the caller instead of relying on generic filler. If a result arrives as the caller asks something new, Alto can answer the question and supply the result together.

Training for Ambiguous Calls

Decagon started with a small open-weight model to keep responses fast, then fine-tuned it on simulated calls informed by production experience. In reinforcement learning, Alto generated several possible responses at each conversation point. The scoring process prioritized whether the model made the correct answer-or-wait choice, then assessed whether its wording followed business policies and reflected confirmed information.

Decagon concentrated training on ambiguous cases after finding that Alto already handled most situations correctly. It also varied policies and tools across simulated businesses, aiming to teach judgment that transfers between situations rather than a memorized script for one company.

Before any caller heard the model, Decagon reported testing it across more than 90 million simulated calls. It then ran Alto in shadow mode alongside live production traffic, where its responses could be evaluated without being spoken to callers.

Two Specialized Voice Models

Alto follows Voice 3 and Chord, announced on October 1, 2026. Voice 3 combined Chord’s speech generation with the duplex architecture and support for more than 70 languages. Chord shapes how Decagon’s agents sound, while Alto shapes how they conduct a conversation. Decagon previously credited Chord with an independent resolution increase of up to seven percentage points.

Together, the models divide voice performance into two training targets: speech naturalness and conversational judgment. Alto’s reported result connects the second target to resolved customer calls, but its central operating constraint remains the answer-or-wait decision—using the speed of the conversational model without letting a natural-sounding exchange get ahead of confirmed actions.