Gnani AI Unveils Artha Sovereign Stack to Bolster Indian Data Independence

Bengaluru-based AI startup Gnani has launched "Artha," a comprehensive sovereign AI stack built to serve Indian enterprises and government institutions. Unveiled in New Delhi, the event featured Vice-President C P Radhakrishnan, marking a significant step for domestic AI engineering. The stack is anchored by Evon v3.3, a 30-billion-parameter large language model, and Plexus, an agentic AI platform. Gnani has made the weights for Evon v3.3 available under an Apache 2.0 open-source license on Hugging Face, allowing developers to build localized applications for free.

Evon v3.3 is engineered to lower enterprise costs by optimizing token consumption, specifically for Indian scripts. It uses 20 percent fewer tokens than GPT-5 tokenizers and significantly less than byte-level systems like Llama or DeepSeek. Utilizing a mixture-of-experts architecture, the model activates roughly 3.5 billion parameters per token. This efficiency is critical for supporting over 11 Indian languages, with the model outperforming larger rivals in the multilingual MILU academic and professional benchmark tests.

The Artha stack addresses the growing demand for data localization in regulated industries, including banking, insurance, and the public sector. Evon v3.3 enables self-hosting on a single node, which helps companies comply with stringent national data residency mandates. Furthermore, the Plexus platform allows for the deployment of autonomous agents capable of tasks like verifying PAN cards or resolving public grievances. Supported by GPU subsidies from the Rs 10,372 crore India AI Mission, Gnani aims to provide a sustainable alternative to costly international proprietary APIs, ensuring firms maintain control over their data infrastructure.