Zoho Unveils Proprietary AI Stack: Zia LLM, AI Agents, and MCP Server

Zoho Goes All in on AI with 25+ Agents, 3 Zia LLMs, and an MCP Server | AIM

The company's new AI stack features its own LLM, ASR model, and a slew of agents. Indian software firm Zoho has announced a suite of proprietary AI tools at the Zoholics India Press Conference, including its in-house LLM called Zia, Automatic Speech Recognition (ASR) models, and a Model Context Protocol (MCP) server. The press conference featured key remarks from Ramprakash Ramamoorthy, director of AI research at Zoho and ManageEngine, along with Mani Vembu, CEO of Zoho. Zia LLM, trained on Zoho-specific use cases, is available in three model sizes (1.3B, 2.6B, and 7B parameters) and optimised for internal deployment across Zoho's product suite. For the training, the company leveraged NVIDIA's AI accelerated computing platform. It supports summarisation, RAG, and code generation, and is currently being tested across data centres in the US, India, and Europe. "This is not ChatGPT and this is not a consumer grade LLM," Ramamoorthy said. "This is an enterprise LLM backbone that sits deeply tied into Zoho's tools. All self-hosted models will be routed to Zia LLM." Zoho says that the LLM is designed to be privacy-preserving and cost-efficient, with different models tailored for structured and unstructured data. "The 1.3 billion parameter model is not a distilled version of the 7 billion. It's built ground up separately," Ramamoorthy said. Zoho also launched English and Hindi ASR models, claiming up to 75% better performance than standard benchmarks. The models are optimised for low-resource environments and are planned to expand to 15 other Indian languages. A major infrastructure addition is Zoho's new MCP server, aimed at enabling interoperability across applications and agents. "We are trying to build an AI stack, we want the infrastructure. We built the AI layer, we built the search layer, we want to build all the tools, all the infrastructure so that we can optimise on cost and accuracy," Vembu said. Zoho has rolled out 25+ AI agents, including pre-built tools like Candidate Screener and Revenue Growth Specialist. These agents are integrated into Zoho's products and can be deployed as digital employees with defined access and audit trails. The company has also launched AI Agents specifically for Indian businesses for verification of PAN card, Voter ID, Udyog Aadhar, GSTIN, Driving Licence, LPG connection and Electricity Bill. Furthermore, support for the Agent2Agent (A2A) protocol will be implemented, enabling Zia Agents to interact and collaborate with each other and with agents on other platforms. The company reported a 32% year-on-year growth in India for 2024. Zoho's most popular products in the country include Zoho One, CRM Plus, and Zoho People, with BFSI, manufacturing, and education sectors driving adoption. Zia LLM and related services will be rolled out to early access customers, with general availability expected by the end of 2025

Zoho Introduces Homegrown AI Model Zia, Multiple AI Agents for Enterprises

The Candidate Screener AI agent can rank the most suited applicants Zoho, the global software-as-a-service (SaaS) giant, has announced several new artificial intelligence (AI) products. On Thursday, the company unveiled its first in-house large language model (LLM) Zia, an automatic speech recognition (ASR) model, more than 25 pre-built AI agents, Zia Agent Studio marketplace, and a model context protocol (MCP) server to let third-party agents connect to its library. These announcements were made on the sidelines of Zoholics India, the company's annual user conference, held in Bengaluru. The SeeS giant's most notable new product is the Zia LLM, its first AI model developed entirely in-house on Nvidia's AI-accelerated computing platform. The models are currently in the testing phase, but Zoho claims these will be available to its customers by the end of the year. Zia is being built in three parameter sizes of 1.3 billion, 2.6 billion, and 7 billion. These models have been trained with Zoho's product use cases in mind can perform a wide range of tasks including structured data extraction, retrieval-augmented generation (RAG), code generation, and summarisation. Zoho claims that these models scored competitively when compared to similarly sized open-source models. Notably, Zoho plans to scale the models' sizes and will begin increasing parameter sizes by the end of 2025. ASR is another area the company is building technologies for. It announced two new ASR models for speech-to-text conversion that support English and Hindi languages. Eventually, Zoho plans to expand the supported languages and add other Indian and European languages. A reasoning language model (RLM) is also in the company's pipeline. The company highlighted that it has developed more than 25 AI agents which are already integrated with Zoho's products. One of the existing agents is Ask Zia, a conversational AI assistant that can analyse data, create reports and dashboards, and assist in building machine learning models. As part of the latest announcement, Ask Zia is also being improved to handle Finance and Customer Support teams. Zoho also says it has improved the Zia Agent Studio, a platform to create AI agents, and made it entirely prompt-based. This means any agents built using the platform will not require extensive coding. These agents can be integrated with more than 700 actions across the company's products. Interestingly, these agents can also be summoned within customer conversations. Among the new pre-built agents is the Candidate Screener, a recruitment-focused agent that identifies and ranks suitable candidates based on role requirements, skills, experience, and other attributes. Another is Deal Analyser that can analyse deals and provide insights such as win probability and next best action. Focusing on India, Zoho has launched several AI agents designed to handle the verification of PAN card, Voter ID, Aadhaar, GSTIN, driving licence, LPG connection, and electricity bill. These agents can be accessed via the company's agent marketplace. Finally, the SaaS giant has also adopted Anthropic's MCP and has created in-house MCP server that allows third-party AI agents to access into Zoho's data and actions from various first-party apps within the user's defined permission structures. Currently, the server has more than 15 Zoho tools available in early access, and more tools will be added in the coming months. Third-party tools can be accessed via Zoho Flow.

Zoho debuts in-house LLM and AI agents; aims to optimise efficiency and reduce costs - The Economic Times

This initiative is a core part of the company's long-term vision to reduce software costs, optimise power consumption, and provide more specialised, business-centric AI solutions, chief executive Mani Vembu told ET.Zoho Corp on Thursday launched a proprietary large language model (LLM) and a suite of artificial intelligence agents, as the software company seeks to deepen its AI capabilities. This initiative is a core part of the company's long-term vision to reduce software costs, optimise power consumption, and provide more specialised, business-centric AI solutions, chief executive Mani Vembu told ET. The company has in the past supported a variety of third-party and open-source LLMs. With the new LLM, named Zia, Zoho is using its own models to power specific use cases within its ecosystem, he said ahead of the launch here. 'Right-sized' approach Unlike consumer-focused LLMs with billions of parameters, Zia is not a single massive model. Instead, Zoho has developed three distinct models with 1.3 billion, 2.6 billion and 7 billion parameters, each "right-sized" and "specifically trained" for particular business applications, Vembu said. This approach aims to provide the best possible outcome for a given task, while also optimising for power consumption and bandwidth -- a crucial consideration for a company that manages its own data centers, he said. While Zoho co-hosts with data centre providers, it manages all its own infrastructure and does not use public cloud services. Zoho's data centres are powered by renewable energy that the company produces more than what they consume, Vembu said. "The goal is to keep the overall cost of software in check, preventing price hikes that are common in the industry. By running these more efficient models, Zoho can continue to offer its AI features as part of its existing pricing plans without charging customers extra," Vembu told ET. The company also launched Zia Agents, which are pre-built AI agents for multiple use cases, along with the Zia Agent Builder and Agent Studio that allow customers and partners to create and publish their own agents to an agents marketplace. "This is based on the belief that a single, one-size-fits-all agent won't work for every business, even within the same industry," he said. Zoho is also introducing Zia MCP (Multi-Cloud Platform) for interoperability. This will allow customers to use other LLMs to build agents that can still access Zoho's APIs, data and action stacks, enabling a "mix-and-match" approach to building customised solutions, Vembu said. Asked how this is different from more established players in the SaaS and enterprise software markets, such as Salesforce and Microsoft, he said while these companies also offer agent builders and third-party integrations, Zoho's focus is on cost optimisation and the efficiency of its "right-sized" models. For instance, Vembu claimed, Zoho's Indic language models, with fewer than 0.2 billion parameters, show a 70% improvement over what is currently available on the market. AI not a replacement When asked about the role of AI in coding and business operations, Vembu said while AI is a powerful tool, it's not yet ready to completely replace human expertise. While AI can generate code, it still requires a human to verify it, he said. He also pointed to the time-saving benefits of low-code tools combined with AI, a strategy that has been a part of Zoho's platform approach. For internal operations, Zoho is using AI to assist employees and increase productivity, for example, by helping support agents respond to more tickets per day or by handling the boilerplate portions of tasks like responding to requests for proposals, he said. AI is a productivity multiplier, not a replacement for human talent, as it still requires a "human in the loop" to ensure accuracy and handle more complex scenarios, Vembu said. Future investments Vembu said while Zoho won't be open-sourcing its Zia models immediately, that is a part of the long-term roadmap. On the company's investment strategy, he said the funds previously earmarked for the abandoned semiconductor pursuit are being redirected to other opportunities, with a significant portion going into R&D. After a year of exploration and a $700 million investment plan, Zoho shelved its plans for compound semiconductor manufacturing in India in May this year. Vembu said Zoho continues to see growth in both its flagship suite and workflow management solution ManageEngine, with security and IT service management being major growth drivers.

AI race: Zoho launches its own LLM, proprietary models for speech-to-text conversion - The Economic Times

Zoho said it is also rolling out automatic speech recognition models for speech-to-text conversion in English and Hindi. The Chennai-headquartered company plans to further expand the available languages, beginning with other Indian and European languages.Homegrown technology company Zoho on Thursday announced its proprietary large language (LLM) model designed for enterprises using its suite of products, the latest move reflecting bold ambitions of Indian firms to build and innovate on their own AI stacks in the global tech race. Zoho said it is also rolling out automatic speech recognition models for speech-to-text conversion in English and Hindi. The Chennai-headquartered company plans to further expand the available languages, beginning with other Indian and European languages. Zoho top management said it has not seen these AI advances impact jobs and hirings. The company further said it grew 32 per cent in 2024 in India, its second largest market globally after the US. The growth was driven by industries such as ITeS, banking financial services and insurance or BFSI, manufacturing, retail, and education. At the Zoholics India -- its annual user conference held this year in Bengaluru -- the company announced proprietary large language model Zia LLM, prebuild AI agents and 'no code' agent builder. Zoho CEO Mani Vembu said, "Our LLM model is trained specifically for business use cases, keeping privacy and governance at its core, which has resulted in lowering the inference cost, passing on that value to the customers, while also ensuring that they are able to utilise AI productively and efficiently." To a question on whether, the company would look to re-engineer or re-purpose its enterprise specific LLM for customers-centric GenAI model at some point in future, to rival, say a ChatGPT, especially given India's sovereign AI vision, Vembu said for the near future, the focus will be on enabling business customers, given vast opportunities and needs in this space. "Based on our experience and research, we will decide on next course of action," he said at a briefing. Zoho said its own large language model is built completely in-house by leveraging NVIDIA's AI accelerated computing platform. "Trained with Zoho product use cases in mind, ranging from structured data extraction, summarisation... and code generation, Zia LLM comprises three models with 1.3 billion, 2.6 billion, and 7 billion parameters, each separately trained and optimised for contextual applicability that benchmark competitively against comparable open source models in the market," the company said in a release. In the near term, Zoho will scale the LLM model sizes, starting with the first set of parameter increases by the end of 2025. "While Zoho supports many LLM integrations for users, including ChatGPT, Llama, and DeepSeek, Zia LLM continues Zoho's commitment to data privacy by allowing customers to keep their data on Zoho servers, leveraging the latest AI capabilities without sending their data to AI cloud providers," the company said. Zia LLM has been deployed across Zoho's data centres in the US, India, and Europe. The model is testing for internal use cases across Zoho's broad app portfolio, and will be available for customer use in coming months. To enable immediate adoption of agentic technology, Zoho has developed a roster of AI agents contextually infused right into its products. These agents can be used across various business activities, handling relevant actions based on the role of the user. These include customer service agent for Zoho Desk that can process incoming customer requests, understand the context, and either answer directly or triage them to a human representative, acting as an efficient first line of assistance. Zoho's platform-wide conversational AI assistant, Ask Zia, is bolstered with additional skills, tailored to data engineers, analysts, and data scientists, while supporting all users within an organisation. At the time of deployment, an "agent" can be provisioned as a digital employee, maintaining the user access permission structure defined within the organisation. Admins can perform behavioural audits as well as performance and impact analyses on "digital employees", ensuring that every agent is working as effectively as possible and within clear guardrails. Citing examples of use cases, it said, several pre-built agents are available for users, such as candidate screener, which identifies and ranks the most suitable candidates for a specific job opening based on role requirements, skills, experience, and other key attributes; deal analyser, which can analyse deals and provide insights such as win probability, next best action, and follow-up suggestions, and revenue growth specialist, which suggests opportunities for upsell and cross-sell for existing customers. The company has also launched AI agents specifically for Indian businesses for verification of PAN card, Voter ID, Udyog Aadhaar, GSTIN, driving licence, LPG connection and electricity bill. These can be utilised for a variety of use cases, such as employee background verification by HR teams or for document verification in financial services organisations.