Build and deploy your first agent with Mastra

hey everybody we will be getting started here in just a few minutes e e all right it's it is the the top of the hour we will still we'll probably wait another I would say two minutes there's probably uh you know normally the first few minutes a bunch of people are still filtering in so we'll give everyone a

little bit more time while we are waiting if you want to drop in the chat you know where you're calling in from So currently I am in sou Fall South Dakota a lot of the time I'm in San Francisco but today I'm kind of back where I where I grew up back home but interested to know where you're all from Florida India Singapore

Ukraine Canada Atlanta New York all right we'll give everyone about 60 more seconds and then we'll we'll kick it off all right let me share my screen and we will get started there's probably quite a few that'll still be filtering in but we'll we'll get it kicked off hopefully you all can see this I guess uh let me know in the in the chat

if you can't for some reason I've been known to share the wrong screen from time to time so I think I got it though all right uh thank you all for attending this uh kind of webinar Workshop we'll talk a little bit about agents talk about how you can build them with mastra and hopefully it'll uh we will all learn something and we'll have

some time for questions at the end but first of all what's the goal we want to build and deploy your first AI agent so after this I will send the recording out it'll usually takes a couple hours and then it'll get sent and you the idea is you should be able either follow along or uh afterwards kind of watch that recording on you know 2x speed and and

go through this if you wanted to deploy something yourself so let's talk about what we're going to learn we'll talk a little bit about some common AI engineering Concepts mostly around what are AI agents we'll talk about you know core MRA Primitives agents workflows evals and who am I I'm one of the founders and the chief product officer at MRA I was kind of in the in team and

product teams at Gatsby and netlify before this I built a service called audio feed and I've been doing open source software for over 15 years now when I first started kind of building Drupal modules way back in the day doing webd stuff uh so feel free to contact or connect connect with me on uh X on

LinkedIn and we can you know continue the conversation and the one thing I will also note is you know I'll be answering some questions along the way if for some reason I you know don't get to them oh Sam's here so Sam will probably answer some questions he's he's one of the other Founders uh and yeah if you drop

questions in we'll answer them as we go or we will also at the end uh have some time to try to answer some and if we for some reason do not get to your questions you can go to our Discord and there'll be Links at that uh either on our website or at the end of the of the workshop so let's talk about what actually are agents and there's kind of

a wide spectrum of what we would consider uh agents but the simplest definition is software with non-deterministic code is kind of the very simplest way to define an a agent it's kind of making a decision and and changing uh based on something that isn't always deterministic but we like to think of Agents as kind of like a spectrum you know something that's very

less agentic is mostly deterministic workflows but maybe there's a few calls to an llm and maybe a few few uh kind of tool calls in there and we'll describe what tool calls are soon something that's a little bit more agentic is the llm actually builds and executes its own plan and has access to maybe a bunch of external tools so it's like a customer

service assistant with access you know to a knowledge base and something that's even more agentic is kind of like an autonomous car as an example actually the car analogy is really good because you can think of there are cars now that have Lane assist that have like autoing those you know are would be very you know far on the less agentic Spectrum right they're they're they help you but

they're not really you know taking full control we're fully you know a mo in San Francisco is fully autonomous that is much more agentic so it is definitely a spectrum and yeah if you have been in enough Whos you know that it doesn't always work perfectly but you know most of the time it's pretty good it's kind of

magical and we built monra to be an agent framework and why you know we kind of started we're building a lot of this stuff ourselves and this tweet uh is kind of a good Summary of why an agent framework is is maybe helpful you know it's like oh all you need is you know an API key to anthropic or open Ai and a loop and a server and then like good API

decisions and like a tool execution sandbox and like an actual database and all this middleware and so you end up with all these different things and then at the end of it you basically just built yourself an agent framework and so a lot of people have been rolling it themselves and what we decided to do is we should just package it up and just make it easier to get further faster so Master is an open

source AI agent framework for typescript you know for JavaScript developers and it comes with agents so you can have tool calling you can have memory you can have tracing tool calling is just a function we'll learn that soon we also have state machine based workflows you can have human in the loop uh so you can suspend and resume

there's we also try to make it the simplest API possible meaning you can call doep do then after it's really uh it's really easy to kind of understand but it also can get complex when you needed to and then we have evals we have storage for rag pipelines we also have a local development playground so we a whole bunch of different things that we package together to make it easy and the

idea around MRA is we want to be opinionated but flexible meaning we come with some opinions out of the box and that's going to help you get further faster but we're flexible enough that you can swap in other things you know we come with the vector DB as you grow maybe you want to switch that out for pine cone or something else that's easy to do um same thing with kind of you

know some of our eval stuff some of our tracing it's all meant to be something that you can either extend yourself or drop in other thirdparty services as you grow or as you might need something specific to what you're building so we're going to spend some time building an agent we're going to create a new poster project we're going

to create an agent that can tell us the weather we're going to create and run an eval metric for our agent we'll just you know kind of highlight what evals are in a really simple way then we'll create a workflow that can plan activities based on the weather we'll show how that workflow works we will then uh spend some time

creating an agent that can answer questions about personal finance transactions the idea is we're going to allow our agent to access some API that gives you know in this case mock transactions but you could think of it like a personal finance Helper and then we'll uh actually deploy this thing and if we have some time we'll we'll even run a test in in a production like environment so a lot of different things

that we're going to try to accomplish today and with that let me let's actually get started by going ahead and creating our first project so on the masterer website we can get started right here and I'm going to go to the command line and I'll try to make this maybe a little bit bigger so you can all see it I know my

screen's kind of large depending on what you're looking at this on so if you can't see please drop in the chat and let me know and we will uh we'll get it adjusted hopefully it is large enough to see and we're going to get started by running that command we just copied from the masterer website and this is going

to go ahead and start installing we have to give it a name and one quick second here I just want to I got to pull up the spreadsheet that we're going to be getting transactions from as well so I'm going to do that in a separate tab here but we got to give this thing a name let's just call this you know March 27th

Workshop we can call it whatever we want we're going to install some dependencies and this is going to take a little bit of time it's going to install you know the core Master package and then once you know once it does do that it's going to give us some more uh some more options or some some more kind of defaults that we can either change or kind of accept so I'm going to

accept the default here I can select what components I want to install I can add tools I can select our whatever provider we want you know this is just some of the defaults we could add a lot more you know we we we are built on top of AIS SDK so any model you can use with aist SDK you can use with MRA so it's most of

the models I will skip uh dropping in my API key I will add an example because that'll be the code we look at first and this is so this is new we we actually allow you to use a masterer doc mCP server so you can actually chat with your uh you can chat with the masterer docs directly in cursor or wind serf and

I'm just going to skip this for now uh we do have some videos on this that we can share afterwards but it is kind of a cool experience so if you have questions or you're as you're asking uh cursor or wind surf to write code for you it's actually going to even have a much higher level of uh accuracy in this case

what we're doing is pretty simple so it's not going to it'll actually just make it take longer because it'll run this agent Loop but it is uh it is really useful when you're just getting started so I would highly recommend doing that but there are a few extra steps to get it to work all right so we now

can go into this Workshop folder and I'm going to copy in just an environment file that has all my API Keys just that's all I'm doing here is just dropping it in that way it can use my open AI API key and uh I guess you can let let us know in the chat if you're team win surf for team cursor I would imagine there's typically more people are cursor but uh

wind surf is definitely becoming more and more popular I uh to be honest I go a little bit back and forth depending on on the week it seems but a lot of people on our team are using wind surf now so let me make this a little bit bigger and you can see that the nice thing out of the box with MRA is it

gives you this folder hey someone said they're using both actually good I'm not the only one I it's it's good to hear Saul that I'm not the only one uh like I said I use it for different different things and I like to go back and forth to see how they how they're both uh changing over time but there's a source masterer

folder and it this makes it really clear what each of these things are right each of these are kind of Primitives that map so we have agents we have tools we have our workflows and then we tie it all together in this index file so we'll spend a little time just kind of going through what how this works before we do that though it's actually kind of nice

just to test it so let's just run mpm run Dev and this will run our local development console and let's pull this up so this pulls up our you know development playground here and we can go in we can see we have our weather agent and tools we have this weather weather tool and we have this workflow that comes out of the box so let's actually test all these things in

isolation and then we'll show the code of how all this works so under agents if I go to weather agent you can see it has a system prompt over here it tells us what it is tells it what model it's using it tells it that it has this tool called G weather which I didn't mean to click on that but I can and then it will bring me to that tool but I'll just ask it what capabilities does it

have so let's ask what is the weather in sou Falls which is where I'm at today okay so you can see it does give me the weather that's probably accurate I should ask for I could ask for Fahrenheit but it's good enough uh for those of you that know Celsius I think that's correct and we can actually then uh I I would say one of the really nice things

is we actually it worked but why did it work and so we have this uh traces UI where we can see this stream call happened and you can see it asked I can see all the different events here what is the weather in sou Falls and I can kind of click in and see that you this there was a stream text there was this tool call that

happened and if I look I can see that the tool actually returned the temperature what it feels like the humidity level the wind speed the wind gust the conditions the location so I can actually dig in and kind of see what caused uh the llm in this case the this agent to make that decision it called the tool what information did it get it back did it actually get back and then

what was the response so it is kind of a nice way to uh see exactly what's happening as I mentioned before we can the nice thing is we can test Tools in isolation so I can go to this weather tool type in sou Falls and this is just going to run that that specific tool so I can test Tools in isolation then hand those tools to agents knowing that they're tested and

ready to go and we also have this workflow and so let's go through this workflow quick and then we'll look at all the code of how this works but this workflow is very similar to what we just did but it has two steps it fetches the weather first and then it plans activities so it's actually going to start uh based on the

weather it's going to make a plan so so I think it's going to be fairly nice this week but in a few days I think it's going to get cold again in sou Falls so let's see what it says you can see it's fetching the weather it did that now it's planning activities it's executing that so you can see there's some uh

information here if I go to the CLI though you can see it's actually streaming out as it's writing this plan so I can kind of see the the results here so on apparently it's going to snow in a few days I didn't know that but that's not fun but it's telling me I should maybe do some indoor activities uh which makes sense so we uh as you can see it's building out this plan and once it's

done and once this is done then it uh it'll actually you know drop the plan in here as well let's look at some code and show how this actually all comes together so I'm open up wind surf here and let's start by just looking at the tool in isolation so there's some types which are just you know typing the response there's this create tool which every tool basically gets an

ID has a description the tool description is very important because the tool description is very informative on how the llm decides what tools to call so in this case you know I always say you should make it so easy that a random person that you ask they can understand what the tool does and when they would want to use it just from the description

so it make it simple enough for any average human and the llms will will do a much better job calling it in the right circumstances and then it has the output schema and it just has an execute function so in this execute function we're basically doing two things first we're getting we're basically geocoding the

location and then we're going out to this API to get the weather and it just returns data so it's pretty straightforward pretty simple and there's some mapping of weather conditions based on the API and then we hand that tool to an agent so in this case we're creating a weather agent we're calling it a new agent weather agent here's the description here's the tools that we

provided and then to wire this whole thing up we at the top level you can see in our Master folder we essentially just export the masterer class and in this case we give it a workflow which we haven't looked at yet but we will we give it an agent and that agent of course has access to that tool we can set a whole bunch of other things here loggers and and a number of other uh things you can change but we'll just

uh call this uh good for now because it is working and as far as the workflow itself this if we scroll down there there's a lot of stuff in here but the most important part is our example does have a little bit of duplicate code which we wouldn't do if we were probably writing it by hand but this does have a new workflow and it essentially has two steps so it has first has a name has what how we're

going to trigger it which is just provide the city and it has a a step and then it'll do this so first step run fetch weather which is just basically running that fetch weather tool and then plan activities which is just calling an agent with a a prompt that says here's how I want you to make the plan okay so we went through agents tools and workflows we have it roughly working I think the next step now is how

do we actually add something that's a little bit more useful to this so let's go ahead and add another tool to our agent and then change your system prompt and actually get it to not just help us with the weather but help us to uh actually get some ual transactions from like a mock transaction list whether this would be like a credit card statement or bank

account or something like that so I have here we'll just use Google Sheets to make it really really easy and you can see we have a uh you know just a handful of transactions of course you'd probably be working with a lot more data but in this case it's nice just to see can I get my agent that's running locally to interact with this and provide me information about this

transaction list which of course could change over time right I can add more transactions uh this you know this would in theory be a living a living transaction list so one of the important things for this is we need to uh publish it to the web I think we already did that so I'll just kind of redo it we'll just do sheet one we will publish it and it this will just give us a link

and you know I'm not going to worry about o here of course if this was real you got to Fig you got to handle the off you don't you don't want people seeing everyone's data but in this case it's a simple example this gives us a nice easy way to just say let's go out to this URL let's get this data and what we're going to

do is we're going to first create a new tool called get transactions and then we will add this tool to our agent and I'm just going to try to use uh you know try to use Cascade in Wind Surf and see if it'll do it for me it it should be able to so I'll say something like I need to create a new tool in this file that is

called get transactions it needs to do a to fetch from this URL and it should return a string containing the the CSV data use the existing weather tool as an example so let's just see how how this does if this works we should have a simple create tool that goes out to that URL pulls in the data and I just wanted to return a string we could do CSV

parsing we could do all that to give it more structured data but it's just a simple enough data structure that an llm can understand it so it's it's doing its work it's doing some editing so get transactions tool get transaction data from Google Sheets that's probably a good enough description I might you know if I was dealing I'd probably write a lot longer description if I was doing this for real

uh you you should think of descriptions as you can be a little bit more verbose than just a couple of words use a few sentences maybe and it's just going to Output CSV data so that looks good it's going to fetch return response text this looks pretty simple and I think it's going to work we'll give it a test but I am

exporting now this get transactions tool so I should just be able to go into my agent import actually what am I doing let's just do it right here get transactions tool and we're going to add this to here and we'll save it and let's go ahead and go back into our playground let's go to our agent let's go to our weather agent you can see I have get transactions I'm going to try

something and see how it does and let's see how much did I spend at Amazon when what was the date in January it's a little dated on this data I guess in January let's see if this actually works oh so it did work I wouldn't always think that this would work right if you look at the system prompt it's actually and I would say about 50% of

the time it works and 50% of the time it's going to say I'm a weather agent I can't give you personal fin financial information but we didn't change the system prompt right we kind of left it as it was which is all around it just being this weather agent so we have this ability to do some prompt enhancement because what we've learned and honestly from firsthand

knowledge and others on the team have learned is that in general writing prompts is kind of complicated a lot of it's like finding boilerplate and being descriptive especially when you're first starting you just want to get like a good enough prompt and then you'll kind of iterate and tweak over time so we did build a prompt

enhancer change this so it can tell the weather so the assistant hopefully I spell don't know how to spell Fahrenheit how do you spell Fahrenheit there we no that's not it either May someone someone in the chat can tell me how to spell Fahrenheit we'll see if it just apparently my spelling is not very good always convert to F from

Celsius we'll see if it does that all right so I'm gonna enhance this there we go thank you thanks Abby appreciate you we'll see if it works with this it's kind of a weird word you know if you don't type it out much all right so here that if we look this is the basically what it's enhanced it to you're a personal assistant specializing in both weather information

and personal financial transaction so it kind of gives it this role definition it gives it core capabilities behavioral guidelines it it was smart enough to write it for me so that was kind of nice I didn't you know I didn't have to uh look it up and then success criteria so it maybe this will be good we'll try it so we can

save it it adds this new version here I can actually in the playground just hit play and try it this doesn't actually write it to my code at least not yet I have to actually if I like it I'll copy it in but I can iterate on it and in check um what is the weather in San Francisco see if this works okay so it does tell me the

fahrenheit the weather API doesn't return that but it will convert it for me so it tells me both so I'm happy with that that that's good uh how much have I I spent at the Apple Store so let's look at the transaction list the Apple Store $1,299 let's ask something that's a just a little bit harder question

so it can summarize I spent you know two transactions on my PayPal account I'm assuming that's right Netflix and Spotify PayPal Spotify Okay cool so I think that's all correct and yeah as you can see there uh it all works I could check the traces let's actually let's create an eval real quick and then we'll run one more test just to show what evals

are uh let's look at the moner docs and if we go to evals that it tells you there's a bunch of different types of evals and there is not actually or there is uh quite a few options for available metrics so you can do like answer relevancy hallucination completeness let's just type let's look at answer relevancy and let's just we

will pull this code in and we'll try it out and we'll also while we're at it we'll also add memory to our agent so right now in this it it doesn't actually store the history of these messages each one is basically it's an individual message so I can't refer back to it and you can see down here it says the agent is not going to remember the previous messages we have to enable memory so

we'll also look at those docs and do that as well and this is where using the mCP masterer docks in your uh in Wind Surfer cursor is actually pretty handy we could just ask it to do it for us and it would know what memory is in MRA and it would probably do a good job with that but we need to start by installing so if you look here

there's a mmra evals so we need to install that and then if we actually so that was installed and while we're at it there's uh there's also memory that we are going to add in as well so I believe that we need to install okay so we'll rerun or we'll just go back in here let's go back to the docs and we'll start with the answer relevancy thing so

we're going to import the answer relevancy metric we're going to create one um actually maybe there's one that's just a little easier contents similarities may be a little easier okay so yeah we'll just look through so basically as you can see you kind of can give you basically can run a measurement of these but you can also just add it directly to an agent so we'll go back

all right we'll stick with answer relevancy that should work we'll grab this we each uh eval can have its own you can select which model you want to use so you could use a different model for your eval than you are for your actual agent itself oftentimes with your uh when you're running an eval you want to use the one of the most powerful models right because it can actually then does

a better job EV valuing so you could use a cheaper model maybe if you get depending on what your agent's actually doing and then maybe you want to run evals with something that's a little bit more uh expressive so we'll just drop this in we already have that so you can see we're going to use 40 mini I could just use 40 if I wanted to so uncertainty weight scale these are

things that we can run and I can just add evals and I believe it's just like that and I can also import memory from Master memory and I can memory equals new memory I think that's it I think that's all I got to do so we're basically just using the defaults you can again build your own memory classes we we have a whole bunch of like options you can configure to say

like how should you store memory and use memory within your agent as it kind of like automatically adds memory back into your instructions but it's all um it has all the dials and you notice I didn't change the system prompt yet so I'm going to go back to the playground and I'm going to copy this and I'm just going to change this system prompt uh maybe it

didn't oh need this one grab the wrong one okay so if all goes well we should now have memory and we should now be able to see some evals as we're kind of going through this so we'll go back here and we'll rerun Dev and we will go back one thing I'm noticing too though is this called weather agent I don't want to be called called weather agent

I'm going to go back and just change that real quick it's no longer a weather agent right just going to call it an assistant agent agent personal assistant agent and then you'll notice I you know because I in the main master class I going have to change this now and change that there but that way it's a little bit more clear what it

actually is and I should be able to just refresh for some reason it did I not save oh I didn't actually save this okay there it is personal assistant agent if I go in here now let's yeah we can see I might have the much longer system prompt I have my two tools let's ask it a question how much did I spend oops on my debit

card okay so it'll pull up that it'll sum it up you can see now I have this chat history so I could start completely new threads I can reference information on the threads as well so we can say you know what we'll see if it makes a tool call the second time so you see how that was a lot faster it's because it had this

past information and it didn't have to make the same tool call so if we look at the traces we should be able to see in this one it you can see there's a lot more because there's some memory things here but we should be able to see generate oh this is actually running the evals as well so that's why we're seeing

some so it's streaming okay so what was the amount on my debit card again and you can see there is no get tool call it's just streaming the text because it already had previously got that information so it already got the information here so when I ran this the stream here it didn't actually need to go back and do the tool call so if I go to evals you can see I

have this answer relevancy metric there's been two evaluations this will run after every time I um I run a chat right now and you can see I can look what was the amount on my debit card again how much did I spend on my debit card and it actually writes the score and so again you can kind of tweak these things to be

you more accurate for what you need so this one says it was actually low relevance because it added a bunch of extra information that maybe I didn't ask for maybe I did want that maybe I didn't but based on this instructions and this this the answer relevancy seems to be a little lower here so we could decide if we like that or not and we can iterate and then the same thing here

what was the amount of my debit card again this one was actually very direct so I would agree that was pretty relevant so you can see over time you can try you can kind of evaluate these things on a bunch of different metrics hallucination completeness answer relevancy all the different kind of off-the-shelf evals that we have but

also you can write your own you can uh verify you know was it you know did it answer in in the way that you actually wanted it to and you could come up with your own scoring system Okay so we've went through memory we went through evals let's see what else we wanted to do oh let's actually deploy this thing so I'm going to go ahead

and just turn this into a get repos repository and then I'm just going to go in and create a repo for this this way you know we'll make it public so anyone who wants to grab the code can look at it uh we'll make it public that so hopefully that code will be up so and we'll share out this repo after the workshop as well but I can

also now go to master cloud and I can deploy this thing so I'll just uh add a new project grab the March 27 Workshop I do need to add my environment variable so I'm going to uh just pull this to the side and grab that and in this case it's just called open AI API key and I'm going to go ahead and everything else is Right want the main

branch deploy and yes you can deploy on other platforms you can deploy to verell netlify cloudflare you know you can kind of build it and deploy it host yourself if you want there's a lot of different options for how you can deploy this we're just showing here so you can easily see that depend no matter where you deploy it it's pretty easy to get it you know get your agent available

through an API the one thing I didn't highlight in uh the playground which maybe I'll do while this is building is that this is all available through an API so whe whether it's running in Dev or it's running you know uh in our Cloud you can actually directly hit this through uh just some API calls so there's a generate there's a stream

and I'll actually just come into Postman and see if we can hit that right now okay so we need so if we look here the Endo is Agents assistant agent generate and we pass in a body that says what is the weather in let's change this to see something different for once it's weather in Austin and if I run this thing you can see this is this is what's

going to run but I'll run it see if it works all right so the current weather is this you can see the actual full request you can see a lot of information about what it did you can see the tool call you can see the results that the tool call Ran but you can see that you can easily access your agents just over

an API so you can connect it to any front end you want whether that's a website a mobile app whatever and in Cloud it it's the same thing so now this is available let's look at the deployment we can see the deployment logs that ran so you can actually see it it built you can see that it has an agent it has

the workflow that we we created and I can actually do the same things now in Cloud I can test it if I want to make sure it's working right I can see traces for anything that's come in through this and evals and all that and can actually uh you see kind of API endpoint just the agents agent ID here's the here's all the information if I want to access this uh as over the

API so that's it I think we kind of went through everything which is good we got some time for questions I think uh Abby and Sam looks like you've been answering quite a few of them but let's go back here and we'll kind of finish it out and then we'll have a few minutes here to answer some questions all right so we actually did

get to memory so I actually you know we did do memory we didn't actually talk a lot about the configuration options for memory so there is quite a bit more you can do but uh we we showed it is does exist it's easy to turn on you probably would want to uh play around with with some of those options though we didn't show human in the loop workflows those are really powerful if you need you know

something to happen you need to wait a human to say yes or no confirm uh provide some feedback and then it would continue on we didn't uh we show anything about using external tool providers we just wrote our own tools uh really the only ones that we really talk about now are just mCP because even compos has their own mCP so we should you you can just

use mCP for pretty much any kind of uh tools that you want at this point there's a bunch of tool Registries we did have a past uh mCP Workshop I think that was last week and uh you know so definitely reach out if on Discord or you know send me an email if you do want to see that you know the recording for that Workshop we didn't talk much about rag or retrieval augmented generation

this is you know we looked at that transaction list and that transaction list was good because it was short but if it was really long if it was imagine that was tens of thousands of transactions well there's a couple ways you're going to need to do that but essentially you're going to have to have a database you're going to have to have some kind of search Rag is really good for semantic search so you could use rag

for some of that you could use a normal database which would still uh it wouldn't be a vector database but would still be able to maybe pull transactions and your agent could decide how to query that data but there's a whole bunch of different things you can do with kind of building these rag systems whether it's traditional databases whether it's Vector databases and then we didn't show custom eval metrics we did talk about it a little

bit but there's a whole bunch you can kind of build on from here depending on the types of Agents or the type of agentic workflows you're building out we do have a few examples uh if you go to our showcase so we have a bunch of examples in our repo if you go to our website and go of showcase there's a few more that we've added so if you're looking to actually see some code see

some examples of how some of this works we do have some pretty complex ones in there some of them are bundled inside an xjs site some of them have separate front ends and back ends so there's a whole bunch of different ways you can architect these systems and we have a few examples on each of those if you wanted if we wanted to deploy a front end for this we could use

assistant UI and have a completely separate front end from our backend running agents so that would be pretty easy to do we could just essentially deploy this example change the API URL environment variable to the path of our agent which is listed on Cloud on moner cloud and it would work and we'd have a a completely uh you know working front end but then of course as you would with

any application we'd have to add you know o and a number of other things but it would be relatively easy to do and it would it would work so there here's some links you can uh check out the GitHub the website please do join on Discord if you're having questions along the way we have a pretty active community so our team is there I'm in there quite a bit

but it's and then we have a lot of great community members as well so it's probably the best way to get uh quick help quick feedback we do have a number of other events we have some that are planned out a few weeks out we're actually about to release some for the next couple weeks so we will send out an email of when

these new events go out there's you know we're kind of starting to get a little more sophisticated in planning them out further ahead so we have a a bunch of them coming up we have a couple guests coming on from other companies that can help uh kind of educate us on on different things that we are not even the experts in but we want to share with

you all and then yeah as mentioned before connect with me on X or LinkedIn and if you uh want to do something right now uh if you haven't already please head over to GitHub and just give us a star that helps us you know enable us to get more people get a little bit more reach more people using MRA more feedback and we can continue to learn and make this uh framework even

better and if you do want to try out masterer Cloud please uh just drop in your your email there and you'll get an email from me that uh basically tells you how you can get in we have slowly rolling this out we have over 2,000 people that have requested to get access and uh so we're kind of slowly letting

people in and that's it let's spend some time on some questions we have about uh 10 minutes for questions so if there are any Obby that you haven't answered we can we can jump in and for those that I don't know if Obby mentioned it obby's the third co-founder so we got we got the all three of us are here so you you got all

of us today hello no I'm pretty good at answering questions so if anyone has more way yeah if we have any more questions yeah Olay if you raise your hand yeah if you want to come off uh and you want to just chat yeah you can you can unmute and talk yeah hi guys uh thanks for the great product I really enjoy it for the

few days already um I have a question about the workflows and I cannot wrap my head around how to use this with ch for example I was using assistant UI and I uh do not quite understand uh for example I want to use like tools on the front end for example this weer let's say user didn't provide it correct address and I want to show him

like uh pick on the map or get my location so like like this use case is there any recommendations or many examples maybe so you you want to use assistant UI with workflows um which one assistant UI we're very close with that project we're working on a UI spec for our workflows so you can have like a native like frontend hook right if you're using react or whatever today it's not

actually like uh really well connected you'd have to not use those hooks and um use monra client's workflow watch function and kind of rebuild itself so it's not a good answer for you but we're aware of it and we are trying to make it better we're also working with another company that you might want to check out

for UI things called co-pilot kit I will put that in the chat here we're working with them as well like in this sense where moner is a backend type of product so we need to emit events to the front end and there are these companies that are trying to do the front end part of it so we're going to integrate with

those so I guess the long story short or whatever we're working on it um if you come to Discord we can try to do workarounds and stuff or keep you up to date on uh how close we are and all that type of stuff okay thank but anyways it is it is possible without using external libraries so like if I use just native master I can build like chat when I can

have back and forth communication and tools you can it's just not going to be like what all the examples you see on you people using like use chat right you we'll have to do something different in the short term okay I got it and one more and we'll figure it out like uh we open up something there the whole team comes and we we help okay thank

you I see I see another question uh in chat which I'm excited to answer why MRA vers Lang chain I feel like this is like you teed it up this is like the softball question um you know we don't really like Lang chain we tried using it you know full full backstory on MRA we were actually building a different product

and we were building it was going to be an AI CRM but we got so far into just building out all the agent pieces and we didn't really like how Lang chain was python first and they kind of you know poured it over to JavaScript but it wasn't well documented it doesn't have all the features of the Python version version and so what we realized is we

were just rolling it ourselves and we decided that you know we were previously at a company called Gatsby where we did a lot of Dev tools more for website development and so we're kind of Dev tools uh guys and that's how we really got here wanting to build some something very specifically for uh building out agents in you know it's typescript first

you know JavaScript first JavaScript only so it's not uh you know we're not trying to support multiple languages like chain which makes you know in my opinion it's very hard to kind of find anything it does feel like they don't necessarily do the best job communicating and being backwards compatible with with all their changes

and so uh that was one of the really driving reasons you know they they have kind of Lan chain Lane graph Lan Smith if you kind of combined all those together that's that's what MRA is is planning to build that's what we have today so again if you are using python I still I wouldn't even recommend Lang

chain if you're using python I'd recommend there's other decent uh python Frameworks that are are better if you're uh using JavaScript then or typescript then I would recommend using us all right can you deploy masterer without using deployers on your server uh if you're on the wait list I will get you off the wait list right

after this call so I'll send you an email if you can uh yeah assuming that I I can figure out who you are on the wait list which I think I can from your name I will uh I will send you an email right after this uh Obby question for you is it possible to deploy to AWS and integrate with bedrock yeah um first the integration with bedrock is through the model

routing um and that's done through aisk so um we have there's libraries for the Amazon Bedrock connection and you'll have your own URL you know typical AWS style and then if you want to do like ec2 or serverless deployment yeah you can do that as well there's no direct deployer for it because working with AWS kind of sucks

um but you can run like we expose all the things you need to build and serve the application come to Discord if you want to know more yeah and to answer the questions and that there's a lot of interest on people on the wait list I will uh give you all the inside uh this is recorded but uh you know the inside secret if you just go to cloud. m.ai you

can get in today so we let you skip the skip the pass uh if you did send me an email I might forget that you were here and you might get another email from me but yeah you're you're all in so go ahead and give it a try we just don't want to you know we're ready for a a lot of people to start using it we just don't want to open the floodgates you know to you know over 2,000 people at

once so we've kind of been rolling out you know 100 or so a day but you all you're all here you stuck around to the end just if you want to try it out go to cloud. master. yeah keep keep it a secret for now in the next couple weeks it'll be you know we'll be moving to public beta and then you know the floodgates will be open to anyone who wants to just you know sign up and join on the spot but there's a few cool features we're

working on that we want to get in before we call it public so it is uh there's a few new things that we will'll be excited to announce in the next couple weeks all right any other questions really excited on the like definitely the Expo plus mustra that is a really cool idea we've had some that have started doing similar things

anything that is especially shows that it's not just for web apps it can be used across uh you know across any kind of front end that you might want to use whether that's you know desktop app we have some that are building desktop apps uh mobile apps all those things so so very cool definitely interested to see what you're what you're building all

right everyone I guess we get to get out two minutes early thank you all you'll get an email in a couple hours with the recording the link to the GitHub repo the slides all this stuff and yeah nice to see you come hang out with us in Discord I'll say that ailon time come hang out with us in Discord that's a it's a cool place see

yall later really appreciate it yeah see you all see you in a bit man