Why build a model that says no?
Language models are extraordinary at producing text. Machines do not need text. A drone needs one exact instruction. A robot arm needs one task. A satellite needs one telecommand, sent in a window measured in minutes. For a machine, the most dangerous thing a model can do is improvise, and the most valuable thing it can do is decline.
So we built Tatva Edge to compile, not to chat. Its output is constrained to the command grammar of the machine it serves, so what comes out is a valid command with named arguments, or an explicit no action. There is no free-text path from a sentence to a motor.
The most important word an AI on a machine can say is no.
Three things it does
It obeys
Plain language in, exactly one schema-valid command out, or no action. The machine's own vocabulary, with your operators' phrasing built in.
It stays
On the machine, offline, in India's languages. Raspberry-Pi-class boards, laptops, servers and the browser, in our own runtime.
It proves
Policy outside the model, a named person for anything consequential, and a signed receipt for every step, verifiable offline.
Where it runs
Each setting has its own deep dive.
Drones
Missions that keep going when the link doesn't. Read more.
Satellites
Plain-language commanding, checked against flight rules. Read more.
Launch operations
A console that answers; a flight computer it never touches. Read more.
Robotics and industry
Task commands in the operator's own language. Read more.
Humanoids
Task commands, with balance left to the robot. Read more.
Defence
Air-gapped, observe and assist, a human in command. Read more.
What it is not
Tatva Edge is not a chatbot and will not hold a conversation. It is not on-device perception yet: that comes next, and we will say so when it does. It is never placed in a safety path, a control loop, a flight computer, a driving function or a weapons chain. A valid command is not the same as a correct one, which is exactly why policy sits outside the model and a person approves what matters. And integration with a specific drone, robot, vehicle or controller is work we do with you, on your hardware; we do not claim it out of the box.
Built in India, from token one
Tatva Edge and its sibling Tatva Nano are trained from scratch in India on openly licensed, licence-recorded data: no inherited foreign weights to explain to an auditor. They come from a company that has put AI into production since 2012, and they run inside the same governance as the rest of the AgentAnywhere platform.
How to get it
See it live today, the final day of Drone Expo at Yashobhoomi, New Delhi, running Drone Mission Ops, Humanoid Assist and the new swarm, border, farm and light-show demos.
Join the waitlist. We are opening access in waves: guided demos first, then a hands-on preview, then paid design-partner pilots on your own hardware and command set. Join the Tatva Edge waitlist.
Build it into your product. If you make drones, robots, vehicles or devices and want an on-device, Indian-language command layer under your own brand, the design-partner programme is open now.
Frequently asked questions
What is Tatva Edge?
Tatva Edge is an on-device AI model from AgentAnywhere, built in India, that turns a plain-language instruction in English, Hinglish or an Indian language into exactly one command a machine understands, or returns no action when the request is out of scope. It runs offline on the device, with policy outside the model, a named person approving consequential actions, and a signed receipt for every step.
Is Tatva Edge available now?
Tatva Edge is in private preview. The waitlist is open, access opens in waves starting with guided demos, and paid design-partner pilots on a customer's own hardware are available now.
How is Tatva Edge different from a chatbot or a large language model?
It does not generate free text. Its output is constrained to the command grammar of the machine it serves, so the only possible outputs are a valid command with named arguments or an explicit no action. That makes it small enough to run on the device and predictable enough to sit behind a policy check and a human approver.
Does Tatva Edge need the internet?
No. It runs on the device in ShepHertz's own runtime, on Raspberry-Pi-class boards, laptops, servers and in the browser, with no network connection.
Can Tatva Edge control my drone or robot directly?
It proposes one command; it is never the last word. A policy engine outside the model checks the command, a named person approves anything consequential, and the machine's own controller does the moving. It is never placed in a safety path or a control loop, and integration with a specific machine is done with you as a design partner.
Written by
Siddhartha Chandurkar
Founder, ShepHertz Technologies
Siddhartha Chandurkar writes for AgentAnywhere, the sovereign AI platform built by ShepHertz Technologies, which has put AI into production since 2012.
