Dhenu Platform is live: agentic AI for agriculture, built on NVIDIA

Today at the India AI Impact Summit 2026 in New Delhi, KissanAI launched Dhenu Platform.

Dhenu Platform is the first agentic AI platform purpose-built for agriculture. It is built on the NVIDIA NeMo framework and NVIDIA GPU infrastructure, and it ships with a 20-agent suite that any agricultural business can deploy to its farmers across voice, WhatsApp, mobile, and APIs. Setting up a working agent takes a day, not a quarter. There is no data-science team requirement and no bespoke integration project.

The launch closes out three years of model work at KissanAI and opens the platform layer that the team has been building toward since the first KissanGPT release in March 2023.

The problem the platform is built for

India’s agricultural economy includes input manufacturers, distributors, retailers, processors, farmer producer organisations, and large agribusinesses. Many of them want to reach farmers directly and build a longer relationship. Few can do it consistently at scale.

The economics break down on the operating side. A field officer can cover about 10 farmers on a good day. Helplines run 20-minute waits at peak. WhatsApp groups for distributors stop scaling beyond a few thousand contacts. Meanwhile farmers are making decisions worth lakhs based on what the last person told them, the recommendation of a neighbour, or guesswork.

The information that would help the farmer mostly exists. India has 726 Krishi Vigyan Kendras, a network of state agricultural universities, decades of agronomy literature, and the catalogues of every major input manufacturer. What the farmer lacks is access at the moment of decision: the knowledge sits in institutions, PDFs, and the heads of the senior agronomists at every agribusiness.

Dhenu Platform is the productisation of the model layer, language pipeline, agricultural knowledge base, and deployment infrastructure that solves this in a repeatable way.

What a business gets on day one

A business that signs up for Dhenu Platform can do six things on day one.

1. Build a custom agent for their own product catalogue. Their products, their pricing, their recommendation logic. The agent will not recommend a competitor’s brand. It stays inside the operating geography, the approved label rates, and the seasonal compliance constraints the business already operates under.
2. Deploy across voice calls, WhatsApp, mobile apps, and APIs. The farmer reaches the agent through whatever channel the business already uses. The platform does not force a new channel on either side.
3. Support 22+ Indian languages out of the box. Dhenu handles the language work, including code-switched dialects, vernacular variations, and audio quality that drops well below telephony spec. About 40% of the farmers calling cannot read, so voice is how most of the conversations happen.
4. Run product discovery, lead generation, and retailer routing. The agent is not a help desk. It runs the demand-discovery and sales motion that an agribusiness would otherwise need a hundred field staff to execute.
5. Get real-time analytics on every farmer conversation. Leadership sees which products are being asked about, which regions have shifting demand, and which farmer concerns are spiking before the field team has reported them.
6. Hand off cleanly to humans when escalation is appropriate. Dhenu answers about 80% of calls and messages with the agent itself and hands off the remaining 20% to human teams when judgment, empathy, or a relationship matters.

Creating an agent on the platform is easier than filling out a form. The platform comes with our curated agricultural knowledge base and the domain-specific Dhenu LLM family built for Indian agriculture. Closed-loop evaluation against held-out farmer interactions is currently sitting north of 95%.

Why now: the platform layer was the missing piece

Getting agricultural AI models into farmers’ hands turned out to be harder than building them.

Across two years of working with large agribusinesses, the same friction surfaced over and over again. Pilots that worked in a controlled environment got stuck at the deployment stage. Data-science teams that the business did not have became the rate-limiting step. Integration timelines that were quoted in months stretched into years. Maintenance budgets that nobody had planned for started showing up. The model was rarely the bottleneck; the path from a working model to a deployed service was.

Dhenu Platform collapses that path.

Upload the raw materials. Product catalogues, technical manuals, agronomy guides, FAQs. The platform automatically curates them into structured product listings and a knowledge base the agent reasons over. There is no data-preparation project and no model-training cycle. The platform handles guardrails, updates, security, scale, and the language layer. Adding a new SKU is uploading a new document.

Built on NVIDIA

Dhenu Platform is built on NVIDIA infrastructure end to end. The Dhenu LLM family was pre-trained and instruction-tuned on NVIDIA GPUs through KissanAI’s NVIDIA Inception relationship, using the NVIDIA NeMo framework for the training and the serving stack. The vision encoder that powers crop-protection diagnostics, the language pipeline that handles the 22+ Indian languages, and the agent orchestration that runs the deployed services all live on the same NVIDIA stack.

KissanAI has been a member of the NVIDIA Inception Program since 2024. The compute relationship, the access to the NeMo team, and the engineering reviews are what made it possible to push from a research model to a production platform on the timeline we did.

Recognition and partnerships

KissanAI is a member of the NVIDIA Inception Program, Microsoft for Startups, and the AI Alliance. The company is a participant in the AI AgriBench Consortium, which is establishing benchmarks for agricultural AI systems, and was shortlisted in the IndiaAI Innovation Challenge in the agriculture category.

Across the work, KissanAI has shipped four firsts in agricultural AI:

• March 2023: KissanGPT, the first voice-based vernacular copilot for agriculture, deployed before the major AI labs added voice.
• December 2023: Dhenu 1, the first domain-specific language model for agriculture from any startup, unveiled at the GPAI Summit.
• 2024: Dhenu Vision, a multimodal model that identifies crop diseases from photographs.
• October 2024: Dhenu 2, with over 1.5 million curated agricultural instruction pairs across model sizes and a climate-aware reasoning layer.

Each of these was built because enterprises kept asking for more than a research model.

Traction

• 150K+ farmers served across 15+ Indian states.
• 22+ Indian languages supported, with regional dialects.
• 95%+ accuracy on the closed-loop evaluation described above.
• Global Fortune 500 agribusinesses as enterprise customers.
• Sub-2-second response time on the voice path.

Geographic focus

Dhenu Platform starts with India and expands to other emerging-economy smallholder markets in the second half of the year. The geographic order is deliberate: the smallholder farmer is the hardest user to build for. “We chose to harden the platform on the toughest distribution first,” says founder Pratik Desai. “If the agent works in Marathi for an onion grower in Nashik on a five-year-old phone over a flaky 3G connection, it works for the larger commercial systems too.”

How this fits with KissanAI

Dhenu Platform is one of two products in the KissanAI portfolio. The second, built for large-holding agriculture teams in North America, gets its own introduction next month. Both share the underlying harness, the agronomy methodology, and the same engineering team, with different tradeoffs on top of that shared foundation.

How to get started

The waitlist for early access is at dhenu.ai. Enterprise inquiries should go to hello@kissan.ai.

A particular thanks to Chintan Papaiya, Lokesh Desai, Nirali Patel, Pravin Mali, Drashti Patel, Satyam Tiwari, and the rest of the team that shipped this, and to the NVIDIA Inception, Ministry of Agriculture and Farmers Welfare, IndiaAI, AgTech Alchemy, Skyhaven Group, and DeepRoot Strategies teams whose support shaped the launch.