The Licensing Decision
When VoicERA reached a state of maturity that we felt comfortable calling production-grade, we had a decision to make: proprietary, open-core, or fully open source.
We chose MIT — the most permissive license available. Anyone can use, modify, fork, sell, and deploy VoicERA for any purpose, including commercial purposes, without restriction.
Here is why.
Infrastructure Should Be a Public Good
VoicERA is plumbing. It is the pipe that connects a citizen's voice to an AI model, and the answer back to their ear. Nobody should need to pay for plumbing.
The value in voice AI systems comes from the knowledge bases, the training data, the fine-tuned models, the agent design, the policy choices. VoicERA does not capture that value. It enables it. Open sourcing the enabler does not give away the value — it multiplies it.
We Need Governments to Trust the Code
A state government will not route citizen calls through software it cannot read, audit, and modify. Full stop. Any licensing model that restricts government inspection is a non-starter for the use cases VoicERA is designed for.
MIT means a government IT auditor can read every line. A security researcher can find and disclose vulnerabilities. An NGO can self-host without legal risk.
The AI4Bharat Ecosystem Is Already Open
AI4Bharat's models are open. Bhashini is government-funded infrastructure. The surrounding ecosystem is built on openness. Adding a proprietary layer in the middle would have been architecturally incoherent.
What Remains Value-Added
Open sourcing VoicERA does not mean there is nothing to offer beyond the code. Organisations that want:
- Supported deployments
- Custom fine-tuning for specific domains
- SLA-backed hosting
- Integration with specific government infrastructure
...can work with teams who specialise in this. But none of that requires the core to be closed.
The Long View
India's digital public infrastructure — UPI, Aadhaar, CoWIN — has succeeded in part because it is open, built on open source, and governed publicly. Voice AI infrastructure should be no different.
VoicERA is our contribution to that stack.