While the rest of the world has chats, Deepsek, Mithun and other AI models, India is still waiting for its home-developed AI system. What is changing to some extent on Thursday as Zoho launched its first big language model called Zia. The company, which provides productivity and functioning equipment to organizations, is now revealed with a large language model called Zia that it will introduce its customers. Zia is certainly an AI system developed in a larger house before an Indian company.

At its annual user conference, Zoholix India held in Bengaluru, the company announced several AI-managed products for the purpose of its enterprise customers. The highlights have Zia LLM, which is particularly developed for its business users.

“Our LLM model is specially trained for business use cases, keeping the privacy and governance at its core, resulting in the customers passing at that price, the estimate cost is reduced, while it is also sure that they are able to use AI productive and efficiently,” said Zoho’s CEO Mani Vambu. “Our discrimination comes from offering agents on our low code platforms so that there is a human in the loop for verification and modification. We call this co-construction with the AI agent.”

In addition, Zoho has introduced several new AI-operated equipment, including automated speech recognition (ASR) model, a no-code AI agent builder, and a model reference protocol (MCP) server-all have been designed to help India companies join the AI race and include various AI tools in their workflows.

Zoho announces Zea LLM

Zia LLM Zoho’s in-house is a large language model, which the company says that has been specially trained for business use cases. The model comes in three sizes – 1.3 billion, 2.6 billion and 7 billion parameters – allowing businesses to be selected based on their references and computational requirements.

Zoho revealed that Zia LLM was designed using Nvidia’s AI computing platform. It supports various types of functions, including structured data extraction, summary and even early-based coding. As part of its AI change strategy, Zoho plans to pursue these models by the end of 2025.

The company has shared that Zia LLM has been deployed in data centers of ZOHO in the US, India and Europe, a strong emphasis on placing user data within ZOHO’s infrastructure to protect privacy. While Zoho will continue to support integration with third -party LLM such as Chatgpt, LLAma, and Deepseek, the company says its ownership models will allow it to reduce the cost and distribute AI to its customers to leave their ecosystem.

AI speech recognition in Indian languages

Zoho has also rolled out the ownership speech recognition and translation models for English and Hindi. According to the company, these models are adapted for low computational load and claim to give better performance up to 75 percent on the standard benchmark.

In the future, Zoho plans to include more Indian languages in his speech recognition and translation tools. The company is also working on a new Reasoning Language Model (RLM) to add to its growing AI suite.

Zia agent studio

Then Zia is an agent studio, which is also aimed at enterprise customers. It is an early-based, low-cod platform for the manufacture of AI agents. The studio consists of over 700 underlying actions and more than 25 prebtt agents, such as candidates Scrinar, Deal Analyzer and Revenue Growth Specialist.

The company states that its users can deploy these agents independently or with the defined access control and audit trails defined as “digital employees”. Some of these agents are also specially sewn to Indian businesses. The list includes document verification for PAN such as agents, voter ID, GSTIN, and more. These zia agents are available through marketplace.

In the coming months, Zoho shared that he plans to open the market for developers and partners, allowing them to create and list their own agents.

In addition, ZOHO has also introduced a model reference protocol (MCP) server, which will allow third-party AI agents to reach the company’s internal equipment safely. Currently, the server supports 15 ZOHO applications and is in early access.