Indian tech powerhouse Zoho is making significant strides in the realm of artificial intelligence. On July 17, the company unveiled its ambitious expansion of AI offerings, introducing Zia LLM, a comprehensive in-house large language model. This launch is complemented by a suite of AI tools, consisting of prebuilt agents, a no-code agent builder, and a server, all designed to enhance AI accessibility and functionality for businesses.
Unlike several competitors that heavily depend on external models, Zoho has opted for a more autonomous approach. The newly developed Zia LLM is a product of Zoho’s engineering prowess, crafted from the ground up using NVIDIA’s AI platform. This internal development allows Zoho to fine-tune models specific to business workflows while ensuring user data privacy on Zoho’s secure servers.
Zia LLM: A Trio of Models
Zia LLM comprises three distinct models, each with varying complexities: 1.3 billion, 2.6 billion, and 7 billion parameters. These models have been individually trained to excel in various tasks such as summarization, code generation, data extraction, and retrieval augmented generation (RAG). Zoho’s strategic approach involves deploying the most suitable model based on the context, effectively balancing performance with energy consumption.
The company asserts that its models perform competitively when benchmarked against other open-source LLMs. Currently, these models are undergoing internal testing across Zoho applications and are slated for customer availability in the near future. Zia LLM will operate from Zoho data centers strategically located in India, the US, and Europe.
Speech Recognition Models Tailored for India
In addition to Zia LLM, Zoho has introduced two automatic speech recognition (ASR) models catering to Hindi and English languages. These models are engineered to be lightweight, fast, and precise, particularly benefiting Indian users. This advancement underscores Zoho’s dedication to catering to localized needs while maintaining high technological standards.
AI-Powered Agents and Customization Options
As part of its latest update, Zoho has launched over 25 AI-powered agents that seamlessly integrate into existing workflows. These agents are designed to streamline processes and enhance productivity. Users have the flexibility to create their own agents using the Zia Agent Studio, which supports both no-code and low-code options, providing access to over 700 actions across Zoho’s comprehensive app suite.
Once deployed, these agents function akin to digital employees, adhering to user-defined access rules while being subject to monitoring for behavior and performance metrics. This feature empowers businesses to optimize their operations with precision and ease.
Agent Marketplace and Protocol Innovations
Zoho is also introducing an Agent Marketplace, facilitating businesses to swiftly find and deploy AI agents. This marketplace will soon enable third-party developers and Zoho partners to list their own agents, fostering a collaborative ecosystem.
To ensure seamless integration across platforms, Zoho is implementing support for the Multi-Channel Protocol (MCP). This protocol allows diverse AI agents to interact efficiently, executing complex tasks with shared data and actions. Currently, MCP actions are accessible from 15 Zoho apps, with plans for further expansion. The company is also preparing to launch support for Agent2Agent (A2A), enabling collaborative interactions among different agents within and beyond the Zoho ecosystem.
Rollout and Future Prospects
These innovative tools are being gradually rolled out to early access users, with a general release anticipated by the end of 2025. Zoho is keenly observing the performance of these tools across various regions and industries, intending to finalize pricing based on comprehensive feedback and analysis.
Note: This article is inspired by content from https://www.msn.com/en-in/money/news/zoho-rolls-out-zia-agents-and-no-code-builder-to-simplify-ai-for-business/ar-AA1IMO9b?apiversion . It has been rephrased for originality. Images are credited to the original source.