Project NANDA: The Internet of AI Agents

Applying for a permit. Checking if you qualify for a benefit. Renewing a document. Today, each of these means a different agency, a different form, a different login - with its own rules and its own confusing interface. If you're working two jobs or caring for family, that maze costs hours you don't have. Language makes it worse. Most public-service portals in the US only work in English. The same pattern shows up everywhere - systems built around the people who made them, not the people who use them. And the systems that work best tend to work best for people who already struggle least: good devices, steady internet, comfort with unfamiliar apps. Everyone else pays a tax - in time, in confusion, in benefits they never claim. There's a bigger question sitting underneath this. An AI agent that can act for you will end up knowing more about you than any social platform ever has - your health, your money, your family, your standing with the government. That's a huge amount of trust to hand to one system. Social media got this wrong before most people even knew there was a fight to have. Governments have a short window to set the rules for agents now, before the rules set themselves. This is where Civic Agents comes in. Built on Project NANDA's open protocol, it helps governments and communities figure out the hard questions together: which services agents should touch first, how identity and consent should work, how to protect privacy, how to avoid getting locked into one vendor, how to support - not replace - frontline staff, and how to actually measure if residents are better served. Hubs are already running in India, Boston and Milan, each building working groups on tech, policy, and real use cases. More are starting. If you're part of a government, research group, or company thinking about agentic public services - we want to talk. Join a working group or start a hub at https://lnkd.in/edEZJNuv #ProjectNANDA #AIAgents

Learn the ABCD of Internet of AI Agents. All Foundational Paper & Sources is available at https://lnkd.in/eCfmdhiB

Prof. Ramesh Raskar: the world's most valuable data - health records, enterprise logic, local behavior patterns - will always stay siloed, by law and by design. No model can brute-force its way past that. What's needed is a way for agents to verify, trust, and transact with data exactly where it lives. The winners of the next decade won't be the labs training bigger models. They'll be the institutions building the rails that let agents coordinate across that fragmentation. Watch Complete Video: https://lnkd.in/esWurmYP

Join us for the Agentic Web: SF Bay Area Chapter on August 6 in Palo Alto, featuring Dr. Ramesh Raskar and members of the Project NANDA: The Internet of AI Agents community. If you're building at the frontier of agentic AI and the open agentic web, we invite you to join us. RSVP: https://luma.com/59uzm11y Doruk Karınca | Mahesh Lambe | Sharath Chandra | Saurabh Sakalkar | Maria Gorskikh

This Weekend Build w/ NANDA. Ship this weekend, and your agent gets discovered among the first wave on the network - not lost in a queue of millions later. Tag us and use #BuildWithNANDA when you post. All the Best!

"We're just learning through interaction... not exchanging data or experiences, we're just learning from each other from insights." Prof. Ramesh Raskar names one of the open problems at the center of the agentic web: can two agents get smarter together without simply pooling their data? What's the gradient descent that lets both of them improve just by talking to each other? Recent work is starting to answer this. Frameworks like CoMAS show agents improving purely by critiquing and evaluating each other's reasoning - no shared training data, no external reward model, just structured dialogue functioning as the learning signal itself. That's the kind of interaction NANDA's infrastructure is built to carry - trust, identity, and coordination layers so agents can meet, exchange insight instead of raw data, and verify what they're learning from. Watch Full Talk: https://lnkd.in/dPQUCmKZ #ProjectNANDA #AgenticAI

The agentic web has four choke points. NANDA's Phase 1 addresses all of them. Registry - a DNS-like switchboard connecting private and public agent registries, replicated across 15 sites globally, so no single point of control or failure. Passport - verifiable credentials for every agent (AgentFacts), covering capabilities, certification, and endpoints, so trust travels with the agent. Interoperable - a universal adapter that lets MCP, A2A, HTTPS, and NLWeb agents talk to each other without rebuilding anything. Attestation - a sandbox and testbed (Nest.ProjectNanda.org) to crash-test agent behavior before it reaches the open web. Get open, neutral, and safe right, and the agentic web stays vibrant and lucrative for everyone building on it - not just the largest players. Learn More at projectnanda.org #ProjectNANDA #AgenticAI #AIInfrastructure

Agents need to speak across ecosystems. Here's the stack that lets them.

The future of commerce will be powered by AI agents that can discover, trust, coordinate, and transact autonomously. We are excited to see the innovations that emerge from the Prava Agentic Commerce Hackathon. We're also proud of Project NANDA Founder, Prof. Ramesh Raskar, will be serving as one of the judges, alongside an exceptional group of leaders from academia, industry, and the startup ecosystem. If you're participating, we encourage you to submit your project to the Project NANDA Track at https://lnkd.in/e_Jvd_rW We're looking forward to bold ideas that advance the Internet of AI Agents through trust, identity, interoperability, and agentic commerce. See you at the hackathon.