fetch-ai

An open platform that enables the creation, deployment, and monetization of autonomous AI agents for various applications.

Fetch.ai

Welcome to Fetch AI's video technical guide. This video is aimed at developers, hackers, and all those entrepreneurial folks so that we can show how our tools at Fetch AI can help you reach your goals at a high level. Fetch AI is building an advanced AI system that connects agents, APIs, and databases together to dynamically solve tasks. My name is Josh. I'm a software engineer working primarily on integrations and demos at FI. 

At FI, we have four core products. We've got the AI engine, which is best showcased by Delta V. Delta V is a web chat interface that allows you to communicate with AI agents and the AI engine. Our AI engine leverages LLMs to understand human input and software code and schemas. This allows the AI engine to discover services that can execute this task and then execute that task by contacting the relevant AI agents. The key thing to take from this is that the AI engine allows you to execute tasks on top of the power of LLMs. AI agents connect to this system as a hook to execute that task. 

Another one of our core products is Agent Verse. Agent Verse is a web platform where you can build and run AI agents. These agents then can be registered as a service, which allows them to be found by their AI engine and then possibly your task that that agent will be executed when the AI engine is looking for that kind of service. So let's go ahead and have a little bit of a tour of Agent Verse. 

If I just sign in, we need to sign in to be able to write agents within the system and also deploy them to services. Okay, so let's take a look. On the landing page, we've just got a list of all of our agents that are running. Not our agents, but all the agents that are running and connected. So I'm just going to go ahead and look at my agents. So I've got one, two, I've got five agents running. 

Alice is a demo. Actually, this isn't super important, but basically Alice is one of the agents that would communicate with another agent. Got basically zero dungeon DS as a tarton and storage agent. If this for this brief kind of producty tour, I'm going to just do a new agent and I'm going to use this request data from a website because that sounds kind of cool. Price feed data request agent submit. 

Okay, and let's go take a look at what it's doing. Great, so it looks like this is okay. So it's getting BTC price from API coindesk.com and there's a function called get BTC.

And this uses the request library. This um yeah request library for that URL checks if we get some data. And on interval calls it um every 10 seconds so nothing crazy. Um it doesn't have a model defined. Uh but we could always update that. So this could be a service. Um if I run this I'm expecting this to every 10 seconds give me a price. Do I need to put an API key in? It doesn't look like it. Fine current price is 42,000. Okay cool we can see that's working. 

Uh and let's stop. Um for our agents. Um in fact let's let's keep that running. Um if we go to services we can generally create a service for an agent. Uh I have two agents two services running. I've got sers of tartan and dice roll. 

Uh for now these these aren't super important but they are very important. Uh a service effectively is how the AI engine knows to find your agent. So what you write in your service is is very important. Um but we will come back to that. We've got stats for our agents that are running. 

Uh we've got the mail room. Effectively the mail room is a way for you not to need to run your agent online all the time. So if you have an agent running on your home device you can effectively uh have agent first store messages for your agent that would be sent to it that can be retrieved when your agent comes back online. 

And that ultimately is the Agent verse Suite. If I quickly go back to services and just give you a quick um like intro what what's going on here. I have an agent called s Tartan. I have a description which says this agent represents a board game. 

Um it's a tabletop game. This agent gets a dice roll for the user so it's quite a simple agent. And I have a model on here which says uh a field descript sorry of a model that says the value of the dice is an integer. This value should be looked up from another task or service. 

So I'll save that and if I go to my agents real quick and I go to sell a Tartan you'll see that I have this uh request model here and it just says dice from an integer. And what's happening here is when a message comes through this effectively returns um the value for a dice roll. 

But this this service doesn't have the value of a dice roll it requires it to return it. And I have another agent uh called dungeons dice roll that does return a uh dice roll value based on the number of sides of the dice. 

So the next of our core products is the microagents library. You can find this at github.com fet and it'll be for/ agents.

Microagents library is a python library that allows you to build AI engine uh comparable agents uh and these agents uh can also be can act as multi-agents in a multi agent system uh yes so the library is open source and there's a few things we should just just highlight uh so in Python uh we have um examples um these examples all will run on your local machine if you um compatible with python um and we have at the core rout we've got integration. 

So these are integrations where effectively we're connecting with external kind of libraries and toolkits and that's kind of the core of of microagents um yeah as uh the microagents library is one of our core products when we have this python uh library to use. 

I think it' be a really good idea just to show you a quick demo of how this uh can run locally on your machine. 

So first off I'm going to just create a directory to to store this agent so make the um let's call it demo agent. 

I'm going to change directory into demo agent and then I'm going to do poetry um in it. 

Poetry just manages my virtual um environment um and I'm just saying yes to everything because it's not not too important right now. 

I'm going go poetry uh shell to get into a shell and then in here I'm going to do uh pip uh install you agents. 

My uh poetry uh shell is uh python 3.9 which is perfect uh for micro agent library and I think it's supported up to 3.2 um now I've installed the the agent library I should make a file and let's call it uh agent.

Py um and I then need to create uh well let's just Nano in and create the agents. So Nano agent. py and I have an agent ready to go which uh is named Alice. Alice is um very simple. Alice just responds to a well just outputs to log every two seconds hello my name is um the the name that's defined as part of her context. So I'm just going to give her a random seed uh because it safe to do that. Uh it's good practice to do that and then I will save my changes and let's run the agent. 

So Python agent. py and then Alice start. We've got hello my name is Alice and we also can see that Alice has started up a uh a web server effectively. So Alice can receive messages from other agents um but Alice doesn't have that function to do so. So let's let's let's add that function. 

Um so I'm just going to stop at Alice for a second then and Nano into Alice again with agent. py and I need to do a couple of things here. I need to define the model if I remember rightly. So I'm going to do class U message um of type model yeah yeah yeah. Um and let's just say that this is going to be message of in this type string and then that looks like too many tabs 1 2 3 four. 

And then I need to define the function for Alice to respond to um a message and we can do that with this function which effectively is async Def message Handler. And what we're saying is Alice on whenever you receive a message at your address address uh we will um respond to it. What we should do here I think is also just add in um maybe uh CTX uh do address so we can see um Alice's address yep and let's uh let's run Alice. 

Oh no Alice agent sorry about that. Um let's just remove a I didn't close it that's why and control X yes and let's try again. Oh I need to import models this is good uh third times a try. 

Uh great so we can see when I print out now my name I also see my agent address. Now the agent address is kind of important because in a local setting when you are uh sending a message you need to know the agent's address of who you're sending the message to. Great I realized it would be really great if we could show an example of how an agent communicates with another agent. So I think we're going to go and do that now. 

Um first thing I need to do is uh open another window and uh let's um change the directory into our working directory which a demo agent. Let's um get into the Poetry shell and let's uh create a new agent. Let's call this agent bob.

Py uh let's uh edit on in there to Bob um and let's let's uh paste in our new agent. So this agent is really simple um there's a few things though that you need to be aware of. Firstly, the address is the address of Alice. You can see Alice has been ped B in printing out her address uh every couple of seconds. We need this address because this address is how the how Bob knows how to uh contact her. 

Uh we've got the model message which is the same as Alice. It was defined in the uh Alice's agent script. We then initialize the agent Bob. We give him the name, we give him the seed, and then we say his endpoint is a locost on Port 8 801. And we obviously define his Port as well as 801. We then say on Bob every two seconds send a message and the message is just saying hello. 

But let's just update this and let's say hello uh Alice uh great day today. And then we send this every two seconds. I'm just going to quickly update this to do it every uh four um and then I'm going to save that uh yes and I'll run Bob in a moment. The key thing I want to do also is to show you the changes I made to Alice. 

So um cat agent so Alice had a couple of uh changes as well. Alice imports fund agent if flow from U agent setup. This is quite important because um it effectively gives Alice credit to register to the almanac contract and that means that Alice can be found by other agents. Um and I think they are the core changes. 

Uh so let's let's keep Alice running and let's start uh Bob.

Py okay great so Bob started Al registration up to date and Bob has started on 801. So you now seeing on Alice. Alice is receiving a message and it's from the address of Bob and it says hello Alice a great day today. 

Now of course these messages could be far more complex. They could have uh much more complex objects. Uh and of course Alice could respond. For more examples of this, the best place to go is fetch a/dogues. Uh and there will be plenty of examples there to to get you going. 

Finally, uh we have our blockchain product at fetch a. Uh our blockchain is uh supporting the alac contract like messaging fees uh for the for the agents. This is best seen um really with the AMAC contract. The AMAC contract is a ledger of agents who are registered um discovered by the AI engine and effectively have who have paid uh for the um blockchain. 

We also have KMP which is a python Library which allows you to um basically send transactions and validate transactions in inter smart contracts um with any Cosmos based blockchain. Okay we've signed in so let's uh Define our task. 

I well as we know we've got two agents running um in my account on Agent verse representing a dice roll and a board game called s of tartan. So let's say I need a new a dice roll uh for the board game Settlers of tartan and I assign them to the service group board games. 

Uh it's good to kind of assign services to a service group that best represents them because it means AI engine can look them up faster and like a work out relations. Um and let's let's start. So my message is sent off uh to Del and Del is trying to work out uh what what that means. 

Um is found a task AI engine has found a task called s of tartan which thinks matches what I'm after. So it starts working on this task um it needs gets dose roll. I'm going to confirm it needs to to do that and yeah it should effectively ask me how many dice sides it has and I'm going to say this dice has 80 sides. 

Um now I I need to confirm this here this is kind of like a um a beta feature because we're using llms in the background. Llms are uh you know can be tricky sometimes so it's good in this kind of chain of of communication to confirm the message effectively. 

But this shouldn't exist um once once Del V leaves uh beta. So now I've confirmed the sites uh the subtask effectively will get that number and the subtask will return to the main task sou of tartan the the the dice roll. So I'm going to confirm.

The dice rolls 37 and this this is actually like the the key kind of power of the AI engine. Um the sers of tartan and the dice roll Services they don't know anything about each other. Um they are compatible. Uh only because one effectively returns like the result. 

Uh another is looking for. I think of AI engine is like a platform that connects. Uh services that are independent of each other and this is this is like like for me the hook of of of what we're building. This is like really powerful because it means that. 

Um you can have a chain effectively of independent apis data silos that are independent of each other but represented by agents and therefore that that kind of information can be changed and returned back in in one command. 

Um as you can see though AI. Uh Del V sorry has U finished this effective and it's returned my do rer 37 and I can then use that in in in my game. 

Um and you can see this message here use this value for the current dice Ro colon 37 is the message returned from Settlers of tartan. So let's take a look at these agents in a little bit more detail and and see how they work. 

Okay we're uh back at agent.

Uh I'm going to quickly sign in. Um and let's let's get to these agents. So I've got two agents. I've got Settlers of Tartan and Dungeons and Di Roll. 

So Settlers of Tartan is the kind of core um like entry layer task and it's really simple. This task effectively just gets like says it needs another task effectively to be completed. So I define the model much like we have seen um in in Alice and Bob beforehand. 

Um I define a protocol. So the protocol is very important because a protocol gets registered um and therefore you know uh can be understood uh by um the AI engine. And then we basically say on message um we need this model filled. It needs to be an integer and we then print out the value once it's returned. 

There's a couple of well very important things here that for Delta V to work with your agent you must specify the type of your agent response type.

Final um there are extensions to these and they can be seen on fetch AI for/ docks. The other agent is um the dice roll. Now the dice roll is the same. It has a different protocol um but this this uh dice roll needs the information number of sides and that's an integer. Um this then returns the um uh the number of uh like the dice roll result and AI engine basically in Delta V uses that number to return to um sellers of Taron. 

Now we're going to go back to Services uh because this is really where where it all works. Um and if I see my service um I can see that my title is set as a tartam and that's what we saw at the beginning with Delta V. Um I give it a brief description uh but the key thing is at the dice uh value of the field description I specifically say that this uh value should be looked up from another task or service. This is telling um a AI engine that it needs to go away to get this value before this to to populate this field. 

Um and yeah that's that's that's critical. Uh so of course um I I specify the protocol and I will save uh just just to quickly create a new service so I can just show you show how this works. Um if I go basically zero I have um I set task and again I can set my message uh field description value whatever that means. 

Um but what I'm trying to show here is that this is populated um based on the agent addresses you have in your agent. So once your agent is running and has registered the almanac you can come here and create a service. Yes I will leave page um and going to dice roll which is um similar uh to sers of tartan. I basically uh define the name um I bas I give a description. 

Uh description is important uh but the key thing again is a field description and down here it says the number of SES provided should be from the task calling subtask. And again that means the a engine knows it needs to get a value from the first task to send uh to to the second. And this is this is how we internally would call this chaining. 

Um multiple agents would reference that they need values from the previous or the next agent and that would kind of build this uh this kind of graph of agents communicating. Okay cool so fet Fords is a great place to find the information you require to build the projects that you're wanting to build. 

Uh the best place to go um in the beginning is guides uh guide we have guides for AI agents um Agent verse Delta V and the blockchain layer. Um uh if you're trying to

Build things um locally. You'll want to be looking at AI agents where we've got lots of tutorials on basically all the features you need. I've shown to build build agents and communicate with each other. Get data from external sources um and also have your agent run locally but also be represented by service on AI um on on a agent vers Ai. Then have your locally running agent uh running uh and connectable on Delta V. 

You've course we've got more guides on Agent verse and these these These are the core ones to Define how you would create a service with your agent um uh again we've got an example for the dice roll um and yeah that that's all really cool. 

Then the any the other thing to add to as I see it for the fetch docs is that we've got examples for the code we're running and we're adding this every day um and yeah that should get you moving really quickly. 

Finally if you need uh help with anything you're working on the best place to go would be Discord. There's links to on Fetch and uh and fet docs but it takes us to Discord and um let's sign in. I should be ready and yeah you can ask here for for for help um yeah thank you.

Note, that there may be slight differences in text and video. Watch video for the original content.

Fetch.ai Video