VoicEra

Open Digital Public Good · Government of India

Open Digital Public Good · Government of India

A phone call, in any language, is enough.English

Open-source voice AI for digital services — so every citizen can be understood by simply speaking, in the language they already think in.

Create your first agent
23+
Indian languages
<2s
End-to-end response
₹0.28
per 2.5-min call
हिन्दीবাংলাతెలుగుमराठीதமிழ்اُردُوગુજરાતીಕನ್ನಡଓଡ଼ିଆമലയാളംਪੰਜਾਬੀঅসমীয়াमैथिलीᱥᱟᱱᱛᱟᱲᱤ

The premise

Voice is the only interface that scales to every Indian.

VoicEra is built as a Digital Public Good: a common substrate any department, state, or partner can deploy, extend, and audit.

  1. Chapter 01

    A billion voices, one gap.

    India speaks in hundreds of languages. Online services and digital services still expect you to type, tap and read English forms.

  2. Chapter 02

    The phone is the interface.

    Nearly every citizen has a phone. A call is the most universal, accessible, and dignified way to be heard.

  3. Chapter 03

    A common platform, freely shared.

    VoicEra is built once, in the open, and made available to everyone — read the code, run it yourself, change what you need.

The pipeline

One call. Open components. Zero lock-in.

Every stage is a swappable interface. Bring your own STT, TTS, LLM or telephony — on cloud or on-prem.

01

Phone

02

Speech-to-Text

03

LLM · Reasoning

04

Text-to-Speech

05

Citizen

23+

Indian languages

<2s

typical round-trip

100%

open source · MIT

On-prem

or cloud, your call

Walkthrough

What happens between the ring and the answer.

  1. 01

    Citizen dials a toll-free number.

    No app install, no data plan, no login. The number becomes the front door.

  2. 02

    VoicEra listens in their language.

    Streaming speech-to-text transcribes in real time across 23+ Indian languages and code-mixed speech.

  3. 03

    The answer is spoken back — instantly.

    Responses are grounded in trusted, verified sources so the citizen hears something accurate, in their language, in seconds. Every call is auditable.

Live demo

A voice you can feel is human.

Real-time transcription, grounded reasoning, and natural speech — on the same device the citizen already carries.

Try the playground
Waiting for caller
Open Source · MIT

Transcript · कन्नड़

Caller: ಇಂದಿನ ಟೊಮೇಟೊ ಮಂಡಿ ದರ ಎಷ್ಟು?

VoicEra: ಬೆಳಗಾವಿ ಎಪಿಎಂಸಿಯಲ್ಲಿ ಇಂದು ಕ್ವಿಂಟಲ್‌ಗೆ ₹1,850 ಸರಾಸರಿ ದರ.

Capabilities

Everything a public service needs — nothing it doesn't.

Multilingual by default

23+ languages. Code-mixed. Real dialects.

हिन्दीবাংলাతెలుగుमराठीதமிழ்اُردُوગુજરાતીಕನ್ನಡଓଡ଼ିଆമലയാളംਪੰਜਾਬੀঅসমীয়াमैथिलीᱥᱟᱱᱛᱟᱲᱤکٲشُرनेपालीकोंकणीسنڌيडोगरीꯃꯤꯇꯩ ꯂꯣꯟबड़ोसंस्कृतम्

Sovereign & auditable

Runs on Indian infrastructure. Every call is logged, redacted, and reviewable.

Real-time responsiveness

Streaming ASR and TTS keep conversation natural — end-to-end response under 2 seconds.

Swap any component

STT, TTS, LLM, telephony — every interface is open. Use your own carrier, or ours.

Analytics

Per-language dashboards, call volumes, success rates, transcripts and summaries — all built in.

Deploys anywhere

On-premise on your own hardware, or on cloud. Same code, same guarantees.

Affordable at scale

₹0.28 per 2.5-minute call. 50 concurrent calls on a single machine. Cost stays flat as you scale.

In production

Farmers, talking to a phone number.

The first deployment of VoicEra is an agricultural advisory line — a farmer dials a toll-free number and asks, in their own language, about crop diseases, weather advisories or mandi rates. VoicEra listens, understands, and answers in the same language, without a human agent on the other side.

Crop diagnosis

Describe the symptoms on your plants. VoicEra suggests the likely disease and a locally appropriate treatment.

Weather advisory

Ask what the weather looks like this week for your village. Get a spoken advisory grounded in real forecasts.

Mandi rates

Ask today's price for a crop at your nearest mandi. VoicEra reads back the latest published rate.

Built in India, open to the world

23+ languages. One phone number. Deployable anywhere.

Every scheduled language of the Constitution — with an architecture that any government, cooperative, or public service can adopt. Hover any language to hear how VoicEra greets in it.

USABrazilUKKenyaUAESingaporeJapanAustraliaIndia · 23+ languages

HI

हिन्दी

Hindi

VoicEra says

नमस्ते, VoicEra से बात कर रहे हैं।

Open source ecosystem

Public code for public services.

VoicEra is developed in the open under an MIT license. Every module — from telephony adapters to policy guardrails — lives in a single monorepo you can read, run, and extend.

COSS-India/voicera_mono_repositoryProduct GuideGitBook Docs
MIT
License
50
Concurrent calls / machine
₹0.28
per 2.5-min call
terminal
# 1. Clone the VoicEra mono-repo
git clone https://github.com/COSS-India/voicera_mono_repository
cd voicera_mono_repository

# 2. Configure environment
cp .env.example .env

# 3. Bring up the full stack
docker compose up -d

# Services now running:
#   orchestrator  -> :8000  (FastAPI)
#   agent-builder -> :3000  (Next.js UI)
#   ferretdb      -> :27017
#   postgres      -> :5432
#   minio         -> :9000  (recordings + transcripts)
#   nginx         -> :80    (public entrypoint)

# 4. Point a phone number at your public host
#    POST /telephony/inbound

Partnership

Every citizen deserves a voice interface to their government.

Deployment partnerships and design collaborations open soon.

Coming soonRead the source