AWS brings PolyAI’s Dialog-RSN-1 voice model to SageMaker JumpStart

PolyAI’s Dialog-RSN-1 reasoning model for real-time customer conversations is now available through AWS Marketplace and Amazon SageMaker JumpStart. AWS customers can deploy it as an endpoint inside their own AWS environment.
- Dialog-RSN-1 is available through AWS Marketplace and SageMaker JumpStart.
- The model processes caller audio and responds in real time.
- Endpoints run inside the customer’s AWS account and VPC.
- JumpStart offers latency, throughput, cost and balanced optimization profiles.
- The model supports English and connections of up to 30 minutes.
A model built for live customer calls
Dialog-RSN-1 is designed to handle complex dialog in real time. According to AWS, it can take in a caller’s audio, reason about what the caller needs and respond during the same conversation, in a single pass.
The model is intended for agents that need to respond to the details of a live call, including tone, hesitation and timing. PolyAI says Dialog-RSN-1 is already handling live calls in production. Unlike a system that first converts speech to text and then sends a transcript to a separate reasoning model, Dialog-RSN-1 reasons directly over the caller’s audio.
AWS says this design can help agents respond faster and keep up with a conversation. It also supports barge-in, allowing callers to interrupt, while transcription is produced from the audio itself. The model is English-only today.
Deployment stays inside the customer’s AWS account
Dialog-RSN-1 is the first PolyAI model available through AWS Marketplace. Customers can subscribe through the marketplace and deploy the model as an Amazon SageMaker endpoint in their own AWS account. They can also start from Amazon SageMaker Studio or select the model in the SageMaker JumpStart catalog and choose Deploy.
The endpoint runs inside the customer’s AWS environment, including its virtual private cloud, or VPC. It uses the customer’s AWS Identity and Access Management roles and can be monitored with existing Amazon CloudWatch logging and metrics. AWS says caller audio and transcripts remain within an environment controlled by the customer, supporting data residency and auditability requirements.
Customers retain responsibility for orchestration, telephony integration and business logic. The model handles the live-call understanding and response element, while the endpoint can be reached from the customer’s existing architecture and billed through its AWS account.
One-click deployment and workload profiles
SageMaker JumpStart provides a model card, overview information and sample notebooks for Dialog-RSN-1. A sample notebook can stream recorded caller audio and display the model’s turn decisions, replies and transcripts, allowing teams to evaluate the model with their own calls after deploying the endpoint.
JumpStart also supplies a pre-configured serving stack, so users do not need to select containers or manually guess instance types. The deployment flow offers four optimization profiles: latency optimized, throughput optimized, cost optimized and balanced.
- Latency optimized targets interactive calls by minimizing Time to First Token.
- Throughput optimized targets higher call volumes by maximizing tokens per second.
- Cost optimized minimizes infrastructure cost per request.
- Balanced provides an even tradeoff for general-purpose use.
Before deployment, SageMaker JumpStart displays P50 latency, Time to First Token and throughput metrics. Deployments can be managed from the console or scripted through the SageMaker Python SDK, including for continuous integration and continuous delivery workflows.
Protocol and availability limits
The Dialog-RSN-1 endpoint speaks the OpenAI Realtime protocol. An existing Realtime client can connect through PolyAI’s bridge by changing its base URL. Audio input must be 16-bit PCM mono at 16 or 24 kHz, with G.711 or Opus audio transcoded first.
Connections can last up to 30 minutes. These requirements define how teams must connect existing telephony and voice applications to the model, while the AWS deployment model lets them apply their existing VPC, IAM and monitoring controls.
For users already building on Amazon SageMaker, the announcement reduces the infrastructure work needed to test and run PolyAI’s real-time dialog model. AWS customers can find Dialog-RSN-1 on AWS Marketplace or deploy it directly from Amazon SageMaker JumpStart.


