Public services should be aphone call away— in the language you think in.

India runs its public services in 22 scheduled languages and hundreds more. Yet most digital services assume a smartphone, a literate user, and English or Hindi. For a large share of the country — farmers, daily-wage workers, older citizens, women in rural households — the phone in their hand is a feature phone, and the only interface that truly works is a voice.

Voice AI existed before VoicEra. What did not exist was voice AI a government department could own: multilingual by default, open to audit, deployable inside its own walls, and cheap enough to run at the scale of a district or a nation. VoicEra was built to close that gap — not as a product to buy, but as infrastructure to adopt.

• Voice first

  Speech is the most inclusive interface there is. A phone call needs no literacy, no smartphone and no app.

• Language is not a feature

  Every scheduled language and the ones beyond them are first-class citizens of the pipeline, not translations bolted on later.

• Sovereign by design

  Runs on-premise or in a data centre of the department's choosing. Citizen conversations never have to leave the country.

• Open, so it can be trusted

  MIT-licensed source. Fork it, audit it, run it. Public institutions should be able to read the code that speaks for them.

• Build once, reuse everywhere

  A horizontal layer under every vertical. Agriculture today, health, welfare and municipal services on the same stack tomorrow.

VoicEra is a platform for building and running voice agents that answer real phone calls. Describe a conversation in plain language and VoicEra assembles a working agent — speech recognition, reasoning and speech synthesis — that speaks 23+ Indian languages and can be deployed on a phone number, a website or an existing IVR.

It is the horizontal layer under every vertical: agriculture, health, welfare, grievances and utilities are built on the same stack, not rebuilt for each one. Its first production deployment is in agriculture, through BharatVistaar and MahaVistaar, where farmers get crop, weather and market advice over a call.

A digital public good (DPG) is open-source software, open data, open AI models, open standards or open content that adheres to privacy and other applicable laws and best practices, and does no harm. The idea is simple: infrastructure that a society depends on should be owned by that society — inspectable, forkable, and free of vendor lock-in.

India has shown what this looks like at scale with Aadhaar, UPI and DigiLocker. Digital public infrastructure (DPI) of this kind is built once, made open, and reused across sectors and states. VoicEra applies the same approach to voice: one open voice layer that any department can adopt, adapt and run on its own terms.

VoicEra is not a standalone startup. It is one building block in a larger effort to make AI work for inclusion, housed inside an academic institution and shaped by the people who helped design India's digital public infrastructure.

1. 01

   The institution

   Centre for Open Societal Systems (COSS)

   

   COSS is hosted at IIIT Bangalore and was established jointly with EkStep Foundation. It works on population-scale societal problems through open digital public infrastructure, and incubates programmes that turn research into working public goods.

2. 02

   The programme

   AI4Inclusion

   

   AI4Inclusion (AI4I) is a COSS programme that builds AI building blocks for inclusion — designed so that language, literacy and connectivity stop being barriers to public services. It ships open, composable components rather than closed products.

3. 03

   The sibling block

   AI Switch

   AI Switch is AI4Inclusion's open routing and orchestration layer that lets a service call the right AI model — sovereign, open or commercial — through one interface. VoicEra uses the same philosophy for speech: swap any layer, keep the conversation.

4. 04

   The voice layer

   VoicEra

   VoicEra is AI4Inclusion's voice building block: a production-grade, open-source platform for citizen-scale, real-time, multilingual voice services, with streaming STT, LLM and TTS, Indic-first code-switching, and SIP/PSTN telephony — deployable fully on-premises.

VoicEra is built with institutions, not around them. These are the organisations that host, fund the wider inclusive-DPI ecosystem, or run VoicEra in production today.

• Centre for Open Societal SystemsHost institution and programme home
• AI4InclusionProgramme under which VoicEra is built
• EkStep FoundationCo-founder of COSS; societal-platform thinking
• IIIT BangaloreAcademic home of COSS
• Gates FoundationPhilanthropic ecosystem supporting inclusive DPI
• BharatVistaarNational agriculture DPI · first production deployment

Logos belong to their respective organisations and link to their official sites.

A small team at COSS, IIIT Bangalore, working across product, engineering and adoption — with departments, states and the open community.