Gnani has launched Artha, a sovereign AI stack designed for Indian enterprises, featuring Evon v3.3, a 30-billion-parameter open-weight LLM optimized for Indian languages. The stack also includes Plexus, an agentic platform for building self-hosted AI agents.
TL;DR
- Gnani introduces Artha, a sovereign AI stack with Evon v3.3 and Plexus for Indian enterprises.
- Evon v3.3 is a 30B-parameter open-weight LLM supporting 11+ Indian languages, with a tokeniser reducing token consumption by up to 50%.
- Plexus enables enterprises to build and deploy AI agents with data residency in mind.
What happened
Gnani, a Bengaluru-based voice AI startup, has unveiled Artha, an end-to-end sovereign AI stack tailored for Indian enterprises and public institutions. The stack comprises Evon v3.3, an open-weight foundational LLM trained from scratch, and Plexus, an agentic AI platform built on top of the model.
Evon v3.3 is a 30-billion-parameter model that supports over 11 Indian languages. Its weights are released under the Apache 2.0 open-source license, allowing developers to download them for free from Hugging Face. A key feature is its rebuilt tokeniser, which requires approximately 20% fewer tokens per Indian-language word than the GPT-5 family tokeniser and less than half of what byte-level tokenisers like DeepSeek, Llama, and Qwen use.
Plexus, Gnani's agentic AI platform, enables enterprise customers to build and deploy AI agents using natural language prompts. These agents support tool calling and can operate autonomously across documents, systems, and conversations. Customers can choose from underlying models on Plexus, including Evon v3.3.
Why it matters
The Artha stack targets sectors such as banking, insurance, and government, which have stringent data residency requirements. Evon v3.3 is designed for self-hosting on a single node, allowing organisations to maintain sensitive customer data within their own infrastructure.
For developers, the open-weight nature of Evon v3.3 and its availability on Hugging Face provide a cost-effective and customizable solution for building AI applications in Indian languages. The reduced token consumption can potentially lower AI costs significantly.
For startups and investors, Gnani's inclusion in the India AI Mission, approved in 2024 with an outlay of Rs 10,372 crore, highlights the company's strategic importance in developing sovereign AI capabilities for India. This could open up opportunities for collaboration and investment in the Indian AI ecosystem.
Key facts
- Gnani launched Artha, a sovereign AI stack with Evon v3.3 and Plexus.
- Evon v3.3 is a 30-billion-parameter open-weight LLM supporting 11+ Indian languages.
- Evon v3.3's tokeniser reduces token consumption by up to 50% for Indian languages.
- Evon v3.3 is available for free on Hugging Face under the Apache 2.0 license.
- Plexus enables enterprises to build and deploy AI agents with data residency in mind.
- Gnani is among 12 local entities selected under the India AI Mission, approved in 2024 with an outlay of Rs 10,372 crore.
- Artha targets sectors such as banking, insurance, and government.
- Evon v3.3 is designed for self-hosting on a single node.
Context
The launch of Artha comes at a time when there is a growing emphasis on developing sovereign AI capabilities in India. The India AI Mission, approved in 2024, aims to foster the growth of the AI ecosystem in the country and promote the development of AI technologies that cater to local needs.
The focus on Indian languages and data residency requirements is particularly relevant for sectors such as banking, insurance, and government, where compliance with local regulations is crucial. The open-weight nature of Evon v3.3 and its availability on Hugging Face also aligns with the global trend towards open-source AI models, which can drive innovation and collaboration in the AI community.