Gnani.ai Unveils Sovereign AI Stack Artha For

What happened

ai has unveiled Artha, an end-to-end sovereign AI stack aimed at Indian enterprises and public institutions. The platform was unveiled by India’s vice president CP Radhakrishnan in New Delhi today. 3, a 30 Bn-parameter open weight model trained natively across 11 Indian languages, and Plexus, the startup’s agentic AI platform. 3 provides the underlying language intelligence, Plexus enables organisations to build and deploy AI agents for specific workflows.

While some voice AI workflows are already largely autonomous, Gopalan said regulated use cases such as underwriting and payments reconciliation will initially retain humans in the loop. He expects more routine tasks, such as document and data matching during loan onboarding, to become increasingly autonomous. The launch comes as the Indian government is pushing for development of indigenous foundational models to reduce the country’s dependence on foreign AI systems.

ai was among the startups selected under the IndiaAI Mission in 2025 to develop homegrown foundational AI models. The startup was selected to build a multilingual, real-time voice AI foundational model with reasoning capabilities. In December 2025, it launched Vachana STT, a speech-to-text (STT) model for Indian languages trained on more than 1 Mn hours of real-world voice data.

The model was developed under the IndiaAI Mission and was part of the startup’s broader voice technology stack. ai offers voice AI products for enterprises. Its offerings include virtual assistants, tools that provide real-time guidance to customer service agents, and voice biometrics for fraud detection. The startup caters to enterprises and government bodies, with its products used across sectors such as banking, insurance, healthcare, telecom, and government services.

It said the capital would be used to expand internationally, strengthen its agentic AI capabilities, develop multilingual and industry-specific products, and expand its engineering and product teams.

Why it matters

ai said the integrated stack will allow enterprises to use AI while retaining control over their data and technology infrastructure. 0 licence. Model weights contain the patterns and relationships learned by an AI model during training and determine how it interprets inputs and generates responses. ai cofounder and CEO Ganesh Gopalan told Inc42.

The 30 Bn-parameter model has been trained on 2 Tn tokens, with a focus on reasoning and Indic-language performance. The startup now plans to expand the model family with 70 Bn- and 100 Bn-parameter models, depending on the complexity of use cases. ai is also looking to expand the model’s language coverage from 11 to 22 languages in the near term.

ai’s upcoming speech-to-speech technology in mind, where low-latency reasoning will be critical. The startup has not disclosed a timeline for the launch of its speech-to-speech model. 3 consumes around 40% fewer tokens for Indian-language workloads than comparable models, helping lower compute costs. Gopalan said the focus on efficiency is particularly important as Gnani looks to deploy its models in production rather than build them purely for benchmark performance.

ai, Artha has been developed to address three challenges faced by Indian organisations adopting AI – data sovereignty, the cost of deploying AI at scale, and the ability to work effectively across Indian languages and real-world use cases.

ai has started showing the model to select customers, with five of 20 enterprises that attended a recent customer meeting in Pune already expressing interest and beginning to build on top of Evon. Its potential use cases include underwriting, advertising and payments reconciliation, although these engagements are currently at the early stage and no deployments of Evon have gone live yet.

What to watch

ai expects Evon and Plexus to contribute 20-30% of its business within the next year, even as its existing voice AI business continues to grow. Gopalan said the company is currently adding around 10–15 customers a month to its traditional voice business. ai raised $10 Mn in a Series B round led by Aavishkaar Capital, with participation from existing investor Info Edge Ventures.