At Drone Expo 2026, ShepHertz today unveiled three India-built AI model families for machines and missions: Tatva (तत्त्व) for on-device command, Sanjaya (सञ्जय) for fleet and telemetry intelligence, and Drashta (द्रष्टा) for machine vision. Part of its AgentAnywhere sovereign AI platform, the models are small enough to run on constrained hardware and work fully offline, with no connection to any cloud. Alongside them, the company is opening its trust toolchain under the Apache 2.0 licence, starting today with Shuddhi (शुद्धि), a data factory that gives every AI training corpus a signed, verifiable receipt.
Tatva, meaning “essence“, turns an operator’s plain words in English, Hinglish or Indian languages into exactly one command a machine understands, or into no action at all. It runs on the device itself, from Raspberry-Pi-class boards to laptops and servers, with Edge, Nano, Micro and Indic variants sized from wearables to ground stations. Sanjaya is named after the Mahabharata narrator who could see the whole field of battle. It gives mission control one live picture across fleets of drones, satellites and vehicles, and keeps a fleet coordinated when links fail or its leader drops out. Drashta, “the seer“, reads satellite and aerial imagery offline and maps objects of interest for analysts to review. It serves defence and strategic users inside air-gapped perimeters, with a named officer in command of every consequential step, and is never placed in a weapons-release chain. Its vision capabilities for inspection and agriculture are being built with design partners.
All three follow one rule: the model proposes, a policy outside the model checks, and a person approves anything that matters. Every step leaves a signed record that can be verified offline. ShepHertz sees the next wave of AI users as machines working where connectivity is thin, contested or forbidden. That spans satellites and ground stations, drone swarms that keep working when the link drops, armoured and utility vehicles where the models are never connected to weapons, in-car assistants that work without signal, and phones, smart glasses and rings. For field engineers, farmers and first responders, that means hands-free help that works anywhere. In every case, the customer defines the machine’s command vocabulary, and anything outside it is refused.
The unveiling closes ShepHertz’s three days at the Expo, where its demonstrations drew defence, agriculture, infrastructure and space teams to the stand. These include a 500-drone swarm that elects a new leader on board when its leader is lost and a humanoid robot that takes tasks in plain language and refuses unsafe ones. They also cover border reconnaissance, Farm Watch, Sky Canvas, rooftop solar inspection, and flood search and rescue, each keeping a person in command. The drone swarm and humanoid demonstrations are simulations, while every Tatva output shown in them is real recorded model output.
Tatva is trained from scratch in India on openly licensed, licence-recorded data, with its training corpus built and receipted using Shuddhi. Every model runs inside the customer’s own perimeter, including air-gapped sites. No data leaves for a foreign cloud, and customer data never trains ShepHertz models. Each model also sits within AgentAnywhere’s live trust layer. There, Veil masks personal and regulated data before any model sees it, and Kavach screens every request and response for attacks and leaked secrets. Every answer and command is then sealed with a signed trust receipt that an auditor can verify offline.
Commenting on the launch, Siddhartha Chandurkar, Founder and CEO, ShepHertz Technologies, said, “India cannot be sovereign in AI by renting its intelligence from abroad. The machines that watch our borders, fly our skies and orbit above us must think in India, answer to the people who command them, and prove every step they take. Tatva, Sanjaya and Drashta are built for exactly that. And trust cannot be a black box, so we are opening the tools that make it verifiable to every builder in India and beyond.”
Shuddhi is available today at github.com/agentanywhere/shuddhi. It turns raw text into a clean training corpus and issues a hash-chained receipt proving exactly what went in and what was filtered out. Over the next three weeks, ShepHertz will also open-source Model Hub, the Agent Universal Gateway, the Annotator and Nabhika OS, the node layer that installs AI on a machine and proves what is running on it.
Tatva Edge enters private preview today, with access opening in waves through the waitlist at agentanywhere.ai. Sanjaya and Drashta are available to design partners, and paid design-partner pilots on customers’ own hardware are open now.