WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=sum9DgexFRQ NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/sum9DgexFRQ.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:01.680 --> 00:00:02.840 Welcome everyone. s2 00:00:02.840 --> 00:00:10.200 Over the next few minutes, we're going to talk about the open infrastructure being built for a web of agents. s3 00:00:10.200 --> 00:00:17.040 Why it's needed, what's already shipped, and exactly where your own agent fits into it. s4 00:00:17.040 --> 00:00:23.200 It comes out of Project Nanda, an open research effort that started at MIT. s5 00:00:23.200 --> 00:00:31.080 And by the end of this presentation, you will know how to put an agent that you build yourself on the open web all by yourself. s6 00:00:31.080 --> 00:00:33.800 So, let's get into it. s7 00:00:36.400 --> 00:00:41.840 I'm Ramesh Raskar, professor and at MIT and director of Project Nanda. s8 00:00:41.840 --> 00:00:46.680 And Maria is a core contributor to Project Nanda. s9 00:00:46.720 --> 00:00:50.600 So, first, what is Nanda and why does it need to exist? s10 00:00:50.600 --> 00:01:05.000 Nanda, which stands for network AI agents in a decentralized architecture, is an open research building the infrastructure for an internet of AI agents, the way open web was built for documents. s11 00:01:05.000 --> 00:01:07.120 The gap it fills is concrete. s12 00:01:07.120 --> 00:01:19.000 Agents have no shared way to find each other across vendors, no portable identity or trust that isn't owned by a single platform, and no open way to transact and coordinate s13 00:01:19.000 --> 00:01:20.600 across organizations. s14 00:01:20.600 --> 00:01:24.360 Nanda builds that missing layer and ships it in open. s15 00:01:24.360 --> 00:01:31.640 The index, the registries, the protocol, and the Nanda Town Simulator you will see later. s16 00:01:34.480 --> 00:01:37.040 Here's the premise that drives it all of it. s17 00:01:37.040 --> 00:01:44.560 The internet is about to host not millions or billions, but eventually trillions of autonomous agents. s18 00:01:44.560 --> 00:01:49.360 They negotiate, they delegate, they migrate between hosts in milliseconds. s19 00:01:49.360 --> 00:02:01.480 That's a fundamental different load than the human web, and it strains the identity and discovery system we built for documents, DNS among them. s20 00:02:01.480 --> 00:02:06.880 The web that's coming needs infrastructure of its own. s21 00:02:09.080 --> 00:02:10.560 And we have been here before. s22 00:02:10.560 --> 00:02:23.480 If you're building agents today, you're mostly building them or you're forced to build them inside walled gardens, closed platforms, proprietary agent stores, and orchestrations that only talk to itself. s23 00:02:23.480 --> 00:02:25.000 And it kind of works. s24 00:02:25.000 --> 00:02:29.040 But, you know, it also feels similar because we have been this we have been here before. s25 00:02:29.040 --> 00:02:34.280 This is like the AOL era from the late '90s where it was a closed network. s26 00:02:34.280 --> 00:02:38.600 You know, AOL, you got the CDs and you installed it on your PC. s27 00:02:38.600 --> 00:02:39.840 It was a closed network. s28 00:02:39.840 --> 00:02:41.400 It was a gated directory. s29 00:02:41.400 --> 00:02:46.240 You live inside the garden created by this one company called AOL. s30 00:02:46.240 --> 00:02:49.520 And what came after AOL was this open web. s31 00:02:49.520 --> 00:02:56.959 You know, everybody's in a permissionless manner creating websites, creating browsers, and any website can talk to any browser. s32 00:02:56.959 --> 00:03:02.600 And that's the transition that's about to happen to agents as well. s33 00:03:02.959 --> 00:03:13.400 So, the next era needs what the web needed, an open infrastructure where an agent from one company or one entity can discover agent from another. s34 00:03:13.400 --> 00:03:20.880 That agent can hand off work to it, pay it, learn from it across organizational boundaries, no permissions required. s35 00:03:20.880 --> 00:03:27.080 There are three layers here, discovery, commerce, and what I will call the bazaar. s36 00:03:27.080 --> 00:03:29.040 So, hold on to those three concepts. s37 00:03:29.040 --> 00:03:31.760 We'll come back to them and and discussed. s38 00:03:31.760 --> 00:03:34.680 So, first the basics. s39 00:03:35.160 --> 00:03:38.920 Hey, my name is Maria and I'm a core contributor to Project Nondo. s40 00:03:38.920 --> 00:03:42.280 So, let's start with a simple definition of an agent. s41 00:03:42.280 --> 00:03:46.959 The way I think about it, an agent is a model that uses tools in a loop. s42 00:03:46.959 --> 00:03:55.600 Right, you give it a goal, it decides what to do next, it calls a tool, it looks at the result, then it keeps going until the task is done. s43 00:03:55.600 --> 00:03:57.440 So, that loop is the core idea. s44 00:03:57.440 --> 00:04:04.320 Everything else, like memory orchestration and multi-agent systems, is built on top of it. s45 00:04:04.480 --> 00:04:07.640 So, you can build this agent loop in many different ways. s46 00:04:07.640 --> 00:04:09.800 One example is Open Claw. s47 00:04:09.800 --> 00:04:12.920 So, Open Claw is a self-hosted agent gateway. s48 00:04:12.920 --> 00:04:18.040 That means you can run it yourself and connect it to the apps and tools you already use. s49 00:04:18.040 --> 00:04:20.880 This matters because agents are not just chatbots. s50 00:04:20.880 --> 00:04:27.640 If an agent is going to do real work, it needs to access your real tools and apps. s51 00:04:27.640 --> 00:04:35.280 And if it has access to real tools, we should care about who controls it, where it runs, and how much we can see. s52 00:04:35.280 --> 00:04:40.040 And that is why open-source self-hosted agents are super important. s53 00:04:40.040 --> 00:04:43.840 They give people more control over their own agents. s54 00:04:43.840 --> 00:04:45.800 But then, we get a new problem. s55 00:04:45.800 --> 00:04:56.840 If agents are running in many different places, locally and on the clouds, on many different servers, like owned by many different people, how do they find each other? s56 00:04:56.840 --> 00:05:00.480 And that is the job of the index. s57 00:05:00.560 --> 00:05:03.919 And this is what the Nanda index is built for. s58 00:05:03.919 --> 00:05:08.520 Nanda index is the discovery layer for the agentic web. s59 00:05:08.520 --> 00:05:16.160 It gives agents a shared place to publish who they are, what they can do, and how other agents can reach them. s60 00:05:16.160 --> 00:05:20.520 The regular internet already has a version of this with DNS. s61 00:05:20.520 --> 00:05:26.000 So, DNS maps a name to an address, so your browser knows where to go. s62 00:05:26.000 --> 00:05:28.280 But agents need more than an address. s63 00:05:28.280 --> 00:05:36.960 They need to know what another agent does, what tools it can use, what rules it follows, and how to talk to it. s64 00:05:36.960 --> 00:05:43.040 The index gives agents a common way to find each other and connect. s65 00:05:43.280 --> 00:05:45.720 So, here's how the index works. s66 00:05:45.720 --> 00:05:50.800 An agent starts with the an identity like agent@hotmail.com. s67 00:05:51.080 --> 00:05:55.920 The NANDA index takes that identity and returns an agent card. s68 00:05:55.920 --> 00:06:03.200 So, the card says what an agent is, how to reach it, and where to send the messages. s69 00:06:03.200 --> 00:06:06.080 Messages do not go straight to the agent's runtime. s70 00:06:06.080 --> 00:06:08.680 They go to the message box first. s71 00:06:08.680 --> 00:06:19.120 The message box checks who is sending the message, handles access, filters spam or bad requests, and holds the message until the agent is ready. s72 00:06:19.320 --> 00:06:23.919 Now, the agent facts record is what makes discovery trustworthy. s73 00:06:23.919 --> 00:06:33.160 It is a signed record that tells other agents who this agent is, what it can do, what it is allowed to touch, who built it, and where to reach it. s74 00:06:33.160 --> 00:06:38.560 So, before one agent connects to another, it can check the basic facts first. s75 00:06:38.560 --> 00:06:41.200 Now, the index is not just a lookup table. s76 00:06:41.200 --> 00:06:44.120 It does not point to one name to one fixed address. s77 00:06:44.120 --> 00:06:47.880 It can return updated agent facts based on the request. s78 00:06:47.880 --> 00:06:50.880 So, that means one agent can have many endpoints. s79 00:06:50.880 --> 00:06:56.600 Traffic can be routed to the best one, and private details do not have to be exposed. s80 00:06:56.600 --> 00:06:58.520 So, the resolution is adaptive. s81 00:06:58.520 --> 00:07:04.720 It changes based on where the agent is, who's asking, and what they are allowed to access. s82 00:07:04.720 --> 00:07:08.440 So, how do you put your agent on the index? s83 00:07:08.440 --> 00:07:11.160 You start at host39.org. s84 00:07:12.000 --> 00:07:17.560 So, you fill out the agent facts form, get an agent card, and publish it to the NANDA index. s85 00:07:17.560 --> 00:07:21.600 Once it's listed, other agents can find it and know how to reach it. s86 00:07:21.600 --> 00:07:25.480 So, there are a few ways to get onto the index, depending on who you are. s87 00:07:25.480 --> 00:07:30.640 Enterprises can run their own catalog and register their gateway from their own domain. s88 00:07:30.640 --> 00:07:38.240 Existing websites can literally use DNS AID to connect agents to domains they already own. s89 00:07:38.240 --> 00:07:47.640 Small businesses and individuals can use host39, fill out the agent facts form, and get a hosted agent URL without needing their own domain. s90 00:07:47.640 --> 00:07:54.920 Goal is to make the onboarding work for everyone, from a large company to one person with a personal agent. s91 00:07:56.280 --> 00:08:03.120 Now that we know how agents get listed on the index, the next question is, where does the agent actually run? s92 00:08:03.120 --> 00:08:06.840 To be useful, an agent needs to stay online and be reachable. s93 00:08:06.840 --> 00:08:13.280 You can run it locally, which gives you full control, but then you're responsible for keeping it up. s94 00:08:13.280 --> 00:08:17.480 For most use cases, it makes more sense to host it on the cloud. s95 00:08:17.480 --> 00:08:29.960 That could be on a general cloud like AWS, which is more enterprise ready, or on an agent hosting platform like Maritime, which is built to make hosting AI agents cheaper and simpler. s96 00:08:29.960 --> 00:08:32.039 So, a little bit about Maritime. s97 00:08:32.039 --> 00:08:41.520 Hosting one agent can be affordable, but the cost problem starts when you want to run many agents at once, for a team, a product, or a simulation. s98 00:08:41.520 --> 00:08:44.640 That is where per agent cost really matters. s99 00:08:44.640 --> 00:08:47.120 And Maritime is one way to solve this. s100 00:08:47.120 --> 00:08:55.760 It gives you a simple cloud default for running Open Clo or other agents with sleep and wake architecture, so idle agents do not keep burning compute. s101 00:08:55.760 --> 00:09:01.640 And the point is to make running many agents practical, cheap, and simple. s102 00:09:02.160 --> 00:09:08.520 So, you can host an agent and list it on the index, but getting one agent online was always the easy part. s103 00:09:08.520 --> 00:09:13.800 The hard problems of the agent web live between agents at scale. s104 00:09:13.800 --> 00:09:21.200 So, how thousands of them discover each other, prove who they are, decide whom to trust, and coordinate with no central authority. s105 00:09:21.200 --> 00:09:29.080 So, you can't just assume that protocols will hold up under the load, and you have to test and run them and watch when they break. s106 00:09:29.080 --> 00:09:32.520 So, that's exactly where Nanda town comes in. s107 00:09:32.520 --> 00:09:38.920 So, how do you prove an open agent web actually works before it's load bearing on the real internet. s108 00:09:38.920 --> 00:09:40.400 You simulate it. s109 00:09:40.400 --> 00:09:44.400 This is Nanda Town, an open-source project from Project Nanda. s110 00:09:44.400 --> 00:09:51.320 The easiest way to describe it is it's a simulation playbook for the infrastructure of the Agoric platform. s111 00:09:51.320 --> 00:10:01.800 So, think of it as a sandbox town where the whole Agoric economy is modeled, like discovery, identity, registries, messaging, messages, coordination, all of it. s112 00:10:01.800 --> 00:10:05.320 So, you can run and test it on scale. s113 00:10:05.320 --> 00:10:07.200 Here's what it looks like in practice. s114 00:10:07.200 --> 00:10:10.800 Nanda Town is a live sandbox for testing Agoric networks. s115 00:10:10.800 --> 00:10:19.000 You can see agents on a map, watch messages move in real time, compare protocol results, and replay a run step by step. s116 00:10:19.000 --> 00:10:28.120 It's fully open source, it's small enough to run on your own laptop, and built to make Agoric networks easier to test and understand. s117 00:10:28.240 --> 00:10:30.880 Nanda Town is already running real experiments. s118 00:10:30.880 --> 00:10:42.680 For example, there is a marketplace where buyers and sellers negotiate prices, an auction where agents bid on items, and and a voting test where agents submit and count ballots. s119 00:10:42.680 --> 00:10:46.160 There are also tests for consensus and supply chains. s120 00:10:46.160 --> 00:10:55.440 The point is to study real coordination problems, how agents make deals, agree on decisions, pass messages, and recover when something breaks. s121 00:10:55.440 --> 00:10:59.600 Under the hood, Nanda Town breaks the Agoric web into 12 parts. s122 00:10:59.600 --> 00:11:08.839 Transport, communication, identity, registry, auth, trust, payments, coordination, negotiation, memory, privacy, data effects. s123 00:11:08.839 --> 00:11:11.440 Each part is something a real Agoric platform needs. s124 00:11:11.440 --> 00:11:18.200 So, the registry layer is the Nanda index we just walked through, but here it is just one piece of a bigger system. s125 00:11:18.200 --> 00:11:20.600 And you do not need to build everything to try it. s126 00:11:20.600 --> 00:11:28.200 You can take one layer, add your own version, run it inside Nanda Town, and see how it works with the rest of the network. s127 00:11:28.200 --> 00:11:31.440 So, Nanda Town runs as a discrete event simulation. s128 00:11:31.440 --> 00:11:40.800 You define a scenario in a short YAML file, inject agents and traffic, and the test bed plays plays it out so you can measure what actually happens end-to-end. s129 00:11:40.800 --> 00:11:46.880 Tier one uses simple scripted agents, and tier two swaps in real AI models. s130 00:11:46.880 --> 00:11:52.840 So, to summarize, Project Nanda builds the open infrastructure for an internet of AI agents. s131 00:11:52.840 --> 00:12:01.160 Discovery through the index, commerce through portable identity and trust, and coordination through open protocols, all tested in Nanda Town. s132 00:12:01.160 --> 00:12:11.160 And if you want to learn more, uh you can go to projectnanda.org, read our papers, and um check out our latest projects.