Northeast India is home to over 220 indigenous languages spoken across eight states. Despite this, government services, citizen helplines, and public information systems overwhelmingly operate in English or Hindi, leaving large sections of the population without direct access in their own language.
MWire Labs has been building the NE-Stack, a set of foundational AI models for Northeast Indian languages including Khasi, Garo, Mizo, Kokborok, Assamese, and others. NE-Agent is the first system that brings these models together into a single, deployable tool.
What NE-Agent Does
NE-Agent is an agentic AI system. Rather than following one fixed process, it automatically identifies the language a citizen is using, decides what kind of help is needed, and routes the request to the right capability:
- Information retrieval, answering questions grounded in a document base
- Translation, currently Khasi to English, with more languages planned
- Speech transcription, converting spoken language into text across eight languages
This routing happens automatically, without the citizen needing to select a mode or menu option.
Why This Matters for Government and Public Services
Most AI systems deployed in India today are built for English, Hindi, or a handful of major scheduled languages. Citizens in Northeast India who are more comfortable in Khasi, Garo, or Mizo are routinely underserved by these systems, whether in agricultural advisory, health information, grievance redressal, or general public service delivery.
NE-Agent addresses this gap directly. It is:
- Fully offline capable, requiring no continuous internet connection once deployed, making it viable in low-connectivity districts
- Free of external API dependency, running entirely on local infrastructure, which matters for data sovereignty and long-term operating cost
- Purpose-built for the region, not adapted from a general-purpose model with no real understanding of these languages
Where It Fits
We see NE-Agent as infrastructure that can sit underneath citizen-facing services: a farmer advisory line, a public health information desk, a district administration helpdesk, an education support tool. The underlying models are the same ones already powering our work with departments across Meghalaya.
We are actively working with government partners to pilot these capabilities in real service delivery contexts. If your department or organization serves communities in Northeast India and is exploring how to make services more linguistically accessible, we would welcome a conversation.
Learn More
NE-Agent is open-source and available for review and evaluation:
- GitHub: github.com/MWirelabs/ne-agent
- Models: huggingface.co/MWirelabs
