Vaani AI Research will launch a self-serve developer platform for building, deploying and managing voice AI agents on 29 October. The company, a foundation speech model firm based in Bengaluru and Singapore, announced the launch on 27 October. Teams can create agents through application programming interfaces, known as APIs, or through a configurable voice harness, without assembling and managing the underlying infrastructure.
The platform is aimed at developers, startups and product teams. Users can describe the agent they want, preview how it sounds, configure its actions and deploy it across phone calls, websites, apps and WhatsApp. Developers can control how agents respond, keep conversational context, use tools and transfer conversations to human teams when a call needs a person.
Vaani AI Research was founded in 2025 by Tushar Shinde, Nitesh Tripathi and Nitish Mishra. It builds speech models and infrastructure for creating and deploying speech systems, and it is backed by Venture Catalysts, the Google AI Accelerator and an undisclosed institutional investor. Its self-serve platform grew out of its work deploying voice AI across enterprise workflows, where a poor conversational experience can cost a business a customer.
What the voice harness does
At the centre of the platform sits Vaani’s voice harness, an infrastructure layer that connects its speech models with the systems needed to run agents in production. It brings together inference optimization, memory and context management, conversational orchestration, telephony, tool integrations and human in the loop workflows. In practice, one stack handles listening, responding, acting and remembering context, instead of a team stitching those parts together from separate vendors.
The harness is designed for the problems that show up in live calls. Callers switch languages mid-sentence, raise unexpected points and expect the agent to remember earlier parts of the conversation. Teams also need room to refine how an agent behaves without relying only on long prompt engineering. Vaani says its platform gives developers precise control over the experience they deliver, so they can adjust behaviour as they learn what callers do.
Opening that infrastructure to a wider developer community lets teams concentrate on their applications and customer experience while Vaani manages the underlying speech stack. The platform supports independent development as well as larger enterprise implementations, and enterprises with specialised requirements can work with Vaani’s forward-deployed engineering team on integrations and workflow optimisation.
Why voice agents fail in live calls
The launch addresses a familiar gap in speech AI: the distance between a successful demonstration and reliable performance in real conversations. A demo usually runs in a quiet room with a cooperative speaker and a script that stays on track. A live call is messier, with background noise, interruptions, changed minds and callers who answer a different question from the one asked.
Vaani says it built the harness inside live enterprise deployments, where every call showed what breaks. That history shaped the platform’s emphasis on context and handover. An agent that cannot remember an earlier answer, or cannot pass a caller cleanly to a human team, creates frustration that ends a sale or a support call. The self-serve tools let smaller teams get those behaviours right without hiring specialists in speech infrastructure.
Co-founder and chief executive Tushar Shinde put the multilingual case plainly. “Seventy-five percent of the world does not primarily speak English, and most voice AI still treats that as an edge case. We built this harness inside live enterprise deployments, where every call shows you what breaks. Today we are opening it to anyone who wants to build a voice agent in the language their customers actually speak,” he said.
Where the voice platform will be used
Vaani’s infrastructure is already used by businesses in India and in international markets. With the self-serve launch, the company aims to make production grade speech technology more accessible to teams building multilingual interactions. Phone support, sales calls and messaging on WhatsApp are among the channels covered from the start.
The company has not published pricing or a list of launch customers. What it has described is the shape of the offer: a developer platform for building agents, plus engineering help for enterprises with complex integrations. The test platform comes in the weeks after launch, when developers start building agents on it and callers start talking to them.
Further detail is available at vaani.dev, where Vaani AI Research lists its platform and contact information for its founders office.