Hey Sonora Realtime devs,

This month’s roundup for developers is here. Let’s get into it.

Product Changelog

Deploy and run voice agents in India

Agents can now be deployed on Sonora Realtime Cloud to the new Mumbai, India region (ap-south) and use co-located Dictate 3 speech-to-text models through Sonora Inference. Co-locating the model and the agent lowers latency for users in India and surrounding regions.

Automatic routing: Sonora Inference routes requests to the Mumbai Dictate 3 deployment when the model and language match a supported configuration. No code changes required.

Built for India-based experiences: shorter round trips and support for local language configurations, including Hindi, on supported models.

For more tips, read the guide to building performant voice agents in India. Co-located Dictate 3 models in the US and EU regions are coming soon.

sonora-realtime-agents 2.3

This release focuses on stability: more granular observability controls, fixed edge-case hangs, and improved reliability under load. STT and TTS fallback, shutdown, and logging behavior are more predictable.

The record parameter on AgentSession.start() now supports granular options: record everything, record nothing, or record audio only without logs, traces, or transcripts.

Content & Community Highlights

Complete workshop: Sonora Realtime 101

Building voice agents in Python? In this video workshop series, Priya Natarajan walks through the end-to-end fundamentals and the production patterns for shipping faster, debugging issues, and keeping agents reliable.

Got an hour? The full course ends with a working voice agent that runs locally.

The Sonora Developer Community is live

The new developer community forums are the place for technical support and discussion with other Sonora developers. Ask questions, share what you are building, and get help from the Sonora team and the community.

Halcyon TTS 1.5

Halcyon TTS 1.5 is now available across Sonora Realtime through Sonora Inference, Agent Builder, and the Agents SDK plugin. The model suits low-latency voice pipelines and handles structured text such as dates, phone numbers, and email addresses.

Sonora Inference includes Halcyon’s default voices. For cloned voices, use the plugin with a Halcyon API key. Cloned voices through Sonora Inference are coming soon.

To hear the model in action, Marcus Oyelaran let an agent run his tabletop role-playing campaign:

Video avatars for Sonora Realtime agents

The Lumen avatar plugin generates a real-time video avatar from a single image and streams it into a Sonora Realtime room. See it in action in Priya Natarajan’s salary negotiation coach demo.

Sonora Realtime in the Wild

Voice AI for healthcare

Clinic Notes is a project built by a team at the Harborline Hackathon that uses Sonora Realtime agents to listen to a patient visit and generate structured outputs such as clinical notes and referrals.

Voice AI on smart glasses

A community demo runs a Sonora Realtime agent on a pair of developer smart glasses. Voice AI, but wearable.

Voice Infra Meetup

Join an expert panel on what it takes to run voice agents in production, with speakers from Sonora, Cobalt Cloud, and Larkspur. Food and drinks included.

Happy hacking!

Let’s build together

Join the developer community
Read the docs
Follow the blog
Watch the video library