Indian voice AI startup Gnani.ai has introduced Artha, an end-to-end sovereign AI stack designed specifically for Indian enterprises and public institutions, combining advanced large language models and customizable AI agents to address unique local needs around data sovereignty and multilingual workflows.
- Artha combines Evon v3.3 and Plexus for sovereign AI solutions
- Evon v3.3 trained on 2 trillion tokens across 11 Indian languages
- Focus on reducing compute costs and enabling autonomous workflows
What happened
Gnani.ai launched Artha, a comprehensive sovereign AI stack tailored for Indian enterprises and public sector organizations. The platform integrates Evon v3.3, a 30 billion-parameter open-weight large language model trained natively in 11 Indian languages, with Plexus, an AI agent framework for deploying customized workflows. This combination delivers advanced language intelligence coupled with flexible AI agent capabilities to enable automation and data-driven decision-making.
The launch event was marked by the presence of India's vice president, CP Radhakrishnan, in New Delhi. Gnani.ai has made the Evon v3.3 model weights available under an Apache 2.0 license on Hugging Face, promoting transparency and access. The startup revealed plans to expand the Evon model family to larger parameter sizes and increase language coverage from 11 to 22 languages, addressing a broader range of regional needs.
Why it matters
Artha addresses key challenges for Indian organizations adopting AI, including preserving data sovereignty, lowering AI deployment costs, and supporting diverse Indian languages. By providing an end-to-end stack that enterprises can control on their own infrastructure, Gnani.ai aligns with the Indian government’s push for homegrown AI technologies to reduce reliance on foreign systems.
The efficiency of Evon v3.3, which uses approximately 40% fewer tokens for Indian language workloads than comparable models, is critical for reducing operational costs. This is especially important for scaling real-world applications that require fast, accurate reasoning such as underwriting, payments reconciliation, and advertising, all while keeping human oversight in regulated environments.
What to watch next
Gnani.ai is currently engaging with early customers, with some enterprises starting to build solutions on Evon v3.3. Monitoring these early deployments will offer insights into the practical adoption of sovereign AI stacks in India’s enterprise landscape. The company’s roadmap includes releasing larger models with up to 100 billion parameters and expanding language support, which could broaden market appeal and application scope.
Another key development to follow is Gnani.ai’s speech-to-speech AI model, intended to complement Evon v3.3 for low-latency voice interactions. Although no timeline has been announced for this launch, it has the potential to significantly advance Indian language voice AI capabilities in sectors like banking, insurance, and government services.