00:00I've spent the last month putting Relevance AI through its paces, testing every feature and pushing the platform to its limits

00:05I've built an AI travel agent, a coach, a multi-agent system around financial advice, which didn't work very well, and I'll explain why a bit later, but in this Relevance AI review, we're going to be covering a lot of ground

00:11First, we'll walk through the platform and its main features, from tools and sub-agents to the flow builder and core instructions

00:17We'll also take a look at integration deployment options and the LLM models that are available

00:23In part two, we'll walk through two real Examples of building an agent, so you can see everything in context and fast track your knowledge of the platform, and lastly, we'll talk about the pricing, the limitations of Relevance AI, how it compares, and when you might want to choose an alternative AI agent platform

00:39But first, though, what is Relevance AI? In short, it's a no-code platform for building AI agents, and what differentiates it from the hundreds of other agent tools out there, like Langchain, Botpress, Autoflowise, etc., is that it's really easy to build multi-agent teams without code

00:55Put another way, what would take me weeks to build using an Agent framework, I can prototype within hours using Relevance AI and deploy within a day or two, but there are some major limitations, which I'll touch on in part three

01:07Firstly, let's jump in and have a look at the platform

01:14So the first thing I'd recommend doing when you jump into the platform is going to the template page and cloning one of these agent templates

01:21You can then look under the hood at how the agent is built, which is going to provide a bit of a demo, a bit more context for how to use different features, and what everything means

01:27It's a little bit confusing if you kind of jump in and there's just like Everything is blank and you have to work everything out at once, so I'm going to go ahead here and let's clone Scott's blog writing agent, so I'm just going to clone it onto our account

01:33This is going to take us to the agent chat UI

01:39I'm not going to trigger it here

01:45I'm first going to show how this agent is built, so there are four things to kind of emphasize here: core instructions, the flow build, tools, and sub agents

01:52Don't need to worry about the other areas at this stage, so core instructions, this is where we tell our agent what it's trying to do

01:58Sometimes we want to be very detailed here, sometimes we can afford To be a little bit more concise, as is the case here, because perhaps we've given this agent a bit more structure and direction, either in the flow builder or via a tool

02:09Sometimes you'll see, like if the agent, for example, doesn't have any tools, doesn't have any flow, like sometimes this can be incredibly detailed, but essentially this is just telling the agent, like, this is what we want you to do

02:22That's kind of, that's kind of it

02:28Next, we have the flow builder, so this is if you have an agent that, so as you can see here, there's, there's nothing here because we're relying quite heavily on the core instructions, and I, I

02:35Think there will probably be a tool attached to this agent, but sometimes you'll see, particularly if you've got kind of lots of if statements and conditions that you want your agent to take into account, you'll see this kind of built out in quite a lot of detail

02:46I like to use the flow builder in situations where I want to make sure the agent captures certain pieces of information before then using tools, so an example that might be, let's say we want to go away and search a database

03:02I might know that I need to get six pieces of information from the user in order to bring back a good result, so I might say ask create

03:09Like an instruction like this, ask the user for X, and then I might then create another instruction saying, "Trigger the tool." Why? That's a use case where, for example, you could rely on that as an instruction, but the flow builder kind of just forces the agent to think of that a little bit more sequentially

03:23Next, we have tools

03:36Now, tools is a huge, huge area, and we'll spend a little bit more time talking about tools in part two when we go through the examples, because there, there's quite a lot to unpack, but at this point, what's worth knowing is tools are essentially the capabilities that we give agents for them to take

03:50Actions outside of Relevance AI, so that might be running a Google search, it might be interacting with a database we run, it might be scraping websites

03:55A really kind of cool thing in Relevance AI is there's a ton of pre-built tools that you can sort of pick up and run with, so you can sort of see here, um, we've got like a bunch of examples

04:07A lot of them are pretty niche, to be fair, but there are some that are incredibly useful, and when we later on in this video, when we look at building our own tools, you'll kind of see like just how much potential there is to do some interesting stuff

04:21And lastly, sub-agents, this for me is Is the thing that sets Relevance AI apart from so many other tools, because it makes it so incredibly easy to build multi-agent teams, like you can just click here and say, allow this agent to interact with, it's a bit of a random example, but a toddler schedule, plan where to live, these agents that I've, I've built, but you can start to kind of see how you could build out a multi-agent team

04:47A lot of other tools in this space make it really easy to build AI agents, but when it comes to building a team of agents that work together to solve a problem, it starts to kind of get a little bit messy, um, little more complicated, and I

04:57Just find the UI for doing this, and Relevance AI is incredibly easy, and each of these agents has its own core instructions, its own flow builder, its own tools that allow it to kind of go off and achieve its objective, and they all can work together as a team to deliver the overall objective

05:08So that's the main thing

05:19There are a few other things I think worth covering at this point: firstly, integrations; secondly, the deployment options; and then lastly, we're going to just touch on the LLM models that are available

05:27So for integrations, I'm just going to jump over to tools quickly because we can see it in a

05:35Little bit more detail, so if I build a new tool, just go here and we're going to add a step

05:42This is probably the best place to kind of get a bit of a sense of the Integrations we have, so I'm just going to show it just so you can kind of see on screen some of the options we've got

05:4911 Labs, hey J, there's a really kind of strong emphasis on text to image, text to audio, text to video, which kind of makes sense

05:55Got lots of kind of data munging stuff, API calls to Hugging Face, HubSpot, Salesforce, Apollo, LinkedIn, Google Calendar, etc., so you can kind of get a bit of a sense of the Integrations

06:11So in my opinion, this is not necessarily A strong area of Relevance AI, there are some really good integrations here, and from my understanding of conversations in the Discord group that Relevance AI run, there's going to be a lot more integrations coming, but compared to other platforms in this space, the range of integrations offered by Relevance AI is relatively limited, so this is something that I just recommend before you kind of get started, just ensuring it has the relevant integrations that you need

06:44One kind of nice catch-all is, rather, AI does integrate with Zapier, it's also got its own API, so if there are things that Kind of fall between the gaps, you might be able to integrate it via Zapier or the API, for example

06:52I wanted to integrate Notion and just did it via API, which was fine

07:00So that's where we are on the Integrations

07:05There are also nine Integrations you can use to trigger your agents

07:09So if we go into, let's go back into this agent, you can see here we've got triggers

07:14So we can, for example, say when I get a certain email in Gmail, I want to run this agent, or when I get a message on WhatsApp for business, I want that to run my agent

07:21So there are both trigger into Integrations and Integrations of where we can send data from our agent

07:28Next, let's talk about deployment options

07:35So again, this is something that is not the strongest area of Relevance AI

07:42Once you've built an agent, you can click up here to share, and you can see we get a URL, which is just a public URL, and then we toggle this on, we get an iframe edlink

07:49So you can of course copy and paste this into a website that you run and have a kind of a public-facing version of your agent

07:56The downside is it's an iframe, so we can't customize the look and feel, we can't really change it

08:02It's essentially going to look like this panel here

08:09So for me, like that was a little bit of a downtick with Relevance AI

08:13Can't easily build public-facing products using this tool

08:18I can build kind of prototypes and internal tools quite easily

08:22Again, I believe this is something that will likely change over time, but this stage, uh, it's a bit of a limitation

08:28Lastly, I mentioned LLM models, so if we just go back into this agent, and let's go here, so in the advanced settings, you can see the language model that this agent is using

08:41So we've got, I think, about 14 different options

08:49They're all kind of either OpenAI or the Anthropic ones, so we've got a pretty good range of models we can use

08:55Again, it's not the largest off

09:01Offering of agent platforms out there, but generally speaking, like we've got the big ones, and in every kind of situation I've built agent with on this platform, like I've not had any issues with that

09:07So those are the main features of the platform

09:13If there's anything that I've missed or is still unclear, please drop a comment below, because if you're wondering, other people probably are too

09:18With that said, we're going to jump in and build our first agent, so we're going to walk through how you might go about building a really simple multi-agent system for planning travel

09:24This is something I've pre-built, so I've got all my content over here so I can copy and paste it, you don't need to watch me type, and there are a few pre-built elements we'll be using

09:30I'll explain as we go, but first of all, we're going to click create a new agent, we'll give it name travel agent AI, this is travel agent specializing in travel for parents of young children, so we're just going to give it a, this is more for our benefit, and then we're going to give it some core instructions, and there are two things of note here

09:51First is that we giving this agent a goal, the goal is to help the user get a trip booked, this is important because every user of this

10:05Agent is going to kind of be at different stages, like someone might know the dates they want to travel and the specific town or area they want to be, whereas someone might just know that they want to get some beach time

10:16Because of that, we want to be able to cater to those different inputs, those different kind of triggers, and help the user get to the same end point

10:22If everyone using this agent had the exact same kind of type of information to start with, we wouldn't need to rely so much on core instructions, we could just kind of have a very structured tool, but in this case with travel, like everyone's going to

10:39Be at a different start point, so that's why we're giving it a very kind of clear goal, so that for one user it might need to ask 10 questions, for another user it might only need to ask two or three

10:44The second thing I note is that we're then giving it a sequence of things it needs to work out, so it needs to first help the user understand, or work out, the destination or the hotel, it then needs to work out flights, it then needs to work out kind of any requests and booking the room, etc

11:03So before we jump into the flow builder, I'm just going to give this agent some tools, so it's definitely going to have to run Google searches

11:09I've pre-built a tool for scraping hotel websites

11:15There is actually one already here, which is extract content from websites, so we'll give it access to that as well

11:20And then lastly, we have a, I've pre-built a tool, which I'll spend time walking through, called travel destination finder for parents

11:28Before I jump into that, just a quick thing here, you'll see that there's an exclamation mark next to all these tools, so with every tool you can, um, determine whether you want the agent to ask for permission to run that, run that tool, or whether you're happy for it to auto approve

11:49This is quite important because for For example, if we gave the agent a tool that allowed them to send emails, we may not just want our, kind of like, while we're testing it, for our agents to just be able to, like, send out hundreds of emails to people we don't know

11:55So, in that situation, like, you might want it to ask for approval

12:00In this case, I'm happy for it to auto-run

12:05It is going to use up some of my credits, so it's worth kind of being aware of that

12:10Um, but I'm not too concerned about that at the moment, so I'm just going to let this auto-run, just so it kind of doesn't kind of get in the way so much

12:17So, I just want to explain a little bit about the travel destination for

12:24finder, so we just jump over here

12:29Um, we can see we've got a custom-built tool, so there are four kinds of parts to Tools in Relevant AI

12:34We've got knowledge inputs, LLM calls, and any output

12:39So, in this case, basically, you could think of this almost like a form

12:43We basically got a bunch of inputs that we want to try and capture from the US, so our agent is going to try and understand this information and pass it to the tool

12:50So, we want to know, for example, the departure airport, the destination, or the type of trip that the user wants to go on, the check-in date, the duration of the holiday, the age of children, how many children

13:04They have the budget

13:18This is almost kind of like if you went into a travel agent, what are the questions that the agent is going to want to ask of you to be able to book a trip? So we're kind of creating inputs for all of those uh bits of information, and as I say, because we're going to give our agent um access to this tool, the agent is essentially going to ask questions to prefill or identify all these inputs

13:32So those inputs we're then going to feed into an LLM prompt, so we're using Open AI, GPT 4.0, and you can see here that we, we basically just built out a prompt, and then we're using variables to pass this information

13:39We're saying we're looking to travel to a destination between check-in date and checkout

13:47We'll have X number of children aged age of children

13:54We're looking to travel for up to X days from departure airport and with a budget or budget, etc., etc

13:59So this is passing it to Open AI, which is obviously going to return a result, and then that result is then going to get passed back to agent

14:10Now, one other thing just to kind of touch on while we're here is we can also give tools knowledge, so perhaps let's say we might, let's say we've got a database of hundreds of hotels that have been ranked really well for holidays for families

14:29With young children, we might want to feed that database as a knowledge for this tool

14:34So, um, you can do that by scraping, like here, we've kind of got a few articles that we've taken, some hotels that are well around, and we're passing to this tool so that when we run this prompt, we can give it this knowledge, and that's going to help improve the quality of the prompt

14:47So that's kind of, that's sort of that I wanted to kind of explain that first, because when we go, when we build the flow builder, we're going to reference a few of those tools

14:54So, although it seems like we're jumping, kind of, we've skipped a step, it kind of makes

14:59sense to set up the tools before we do the flow builder

15:05So we've got our agent, we've given it a goal and instructions, we've given it access to some tools, and we know what these tools do

15:10We're now going to add a kind of a sequence and tell our agent, um, what to do in what order

15:16So the first step, we're going to give our agent the instruction that it's going to ask the user for their budget, dates they're looking to travel, whether any locations come to mind, so this now has a bit of context, right? Like, remember when we were looking at the travel destination tool, it had those inputs, so the agent is going to start trying to

15:40Prefill some of that information, then it's going to pass that to our tool, which is going to give us some options

15:47So I'm going to create an instruction saying, "Once five options have been presented, ask the user whether any of these are suitable." We'll then create a condition that says, "If the user says one of them is acceptable, then use Flight Time Finley," which is another agent that I've built to go off and look at flight times to find some suitable flights for this trip

16:04Also, get the hotel website scraper to find an email address of the hotel, so that was the hotel scraper tool that we had earlier as well

16:14So once the Flight time agent comes back to some flights

16:23We're going to suggest all options for flights for the trip, ranked best to worst, and then lastly, what's going on there? Lastly, we're going to add final instruction: ask the user whether they'd like to email the hotel to confirm they have babysitting services, and a copy of the hotel's email address, and draft an email

16:33So we should be pretty much good to go

16:39We've given the agent some basic details, core instructions, it's got a flow, given it access to tools, we've set it all to auto run

16:44Lastly, we need to give it access to sub agents, so we, because we reference flight time, Finley

16:49We're going to just give it access to that agent as well, so I'm going to click confirm and continue, and we should now be able to test out our agents

17:02I'm just going to say, "I'm looking to travel to Mexico in May 2025." Okay, so our travel agent is working, so it should come back and have a couple of questions that it needs to fill out, that tool that we built

17:17So, what is your budget for the trip? Let's say £4,000, one child who is 2.5 years old, how long do you plan to stay in Mexico? 3 weeks, it's completely made up

17:33Do you have a specific destination in mind? Say, apply Del common, what type of accommodation? Family friendly result

17:39Six, do you have specific dates? Not yet

17:45Okay, so we'll send that off

17:52That should now hopefully run our tool

18:00Okay, since come back asking for a question, before I proceed, could you please provide me with a rough idea of your check-in? Okay, so let's say 2nd of May check-in to, oh, three weeks, let's say 21st of May

18:05Let's try that

18:10So hopefully this now runs our tool, which will run that LLM prompt

18:15So we can see that's running in the background, so it's using our tool, and that should now return an option

18:22I believe we had an instruction that said come back with five options, so let's see if we get that

18:30So that's complete

18:38Okay, so Let's come back with five options, five beaches, Hotel blah blah blah, so it's nice, okay? So if we pick one of these, let's just say, okay, so if we pick one of these, let's say the option two looks great, option two

18:46Now it should now pass us over to the flight time sub agent, which should go off and find some flights for us to Mexico, so it's kind of interesting, like seeing what the agent sent to another agent

18:59Here you go, find suitable flights for a family trip from UK to Grand Rivier mayor PL Del kman checking in, cool

19:05So this is now running some Google searches, we'll wait for that to come back, okay, so BA Virgin 2E, so let's say

19:12Let's go ahead with that

19:18Now, it should run this, the hotel scraper

19:25So, this is trying to find the email address on that hotel's website, of an email address we can contact to then write an email to the hotel, see if they have babysitting services, or if they can make sure that they have a cot, very important if you have young kids

19:39So, let's see if that's going to work

19:45Oh, okay, so in this case it didn't work, which means probably either that the hotel doesn't have an email address, um, on their website, or no, okay, so it has found it, it just hasn't scraped it

19:55So, in this case, I'd probably double check that that's actually correct

20:01And it's not made that up, but it's written an email and hopefully found the email address

20:06So next, we're going to have a go at building a financial advisor agent that does cash flow modeling, so basically takes your income, your expenses, any assets you have, and kind of extrapolate and forecast the value over time

20:19So this is something that we know can be done because financial advisers do this all the time, it's based on data and a set of assumptions

20:26The reason I wanted to share this as an example is I've really struggled to get it working in Relevance AI

20:32I've tried building it where it just has core Instructions, I've tried building it with a flow, I've tried it being a tour, and each one's kind of had its own challenges

20:38So partly, I want to share this just to show some of the limitations, but also, like, if you have any ideas of how you might go about building this in a way that would understand some of those nuances better, like, I'd love to hear how you'd approach it

20:55So I'm going to keep this one quite simple, just with nothing but core instructions to give this agent the maximum amount of flexibility to come back and ask me more questions, where I found this struggle in the past as a flow or tool is it becomes too

21:01structured, and it doesn't ask additional questions when it really should

21:07So we're just going to give it some core instructions

21:11As you can see, these are incredibly detailed

21:16The reason for that is really to help the agent understand the bigger picture

21:21One of the key things that we've got going on here is we've given this agent three different steps

21:26First, this agent needs to understand where the user currently is, what age are they, what investments do they have, anything that would affect their kind of point A

21:38Then, in order to extrapolate kind of the future and how their portfolio might change, we Need to really understand two things in detail

21:45We need to understand cash inflows, like what is their income, how might that change over time, does their portfolio generate any income, and lastly, we need to understand outflows, how much are they spending per year, are there any expensive events in the future, etc

21:56So if we get those three things right, and we understand them in enough detail, we should be able to create a reasonably good model

22:03So I'm just going to confirm that, and we can get this running

22:09So I really need to get better at setting welcome messages for my agents

22:13I end up just like triggering

22:17them by saying, hey, you can

22:21If I just go in here, I'll just show you that quickly

22:24In advanced settings down here, you can, you can give you your uh, agent welcome message, which just then prompts, basically just shows like the thing the agent would prompt

22:28So here we go, which current age, obviously for this one, I'm going to enter completely made up data, otherwise that would be incredibly silly

22:41So I'm going to say, I am 52 years old, I have a house and $400,000 of Vanguard VTI funds

22:49No, I don't have an emergency fund

22:57My net worth is $400, $500,000 and it is 80% liquid and fully invested

23:02Cool, let's see what happens

23:07Okay, okay

23:12Interesting, so it thinks at this point it's understood point A, because it's moving straight onto understanding the cash inflows

23:17So earned income, let's say $80,000 a year

23:24Does your portfolio produce any income? If so, how much do you receive? So let's say the property gets a 5% yield and the VTI funds, I don't know, 3%

23:39Will these income sources increase with inflation? Yes, with inflation

23:43What rate? Um, I don't know, 3%

23:46Let's model that

23:50So presumably now it's going to go onto the outflows, cool

23:53Okay, so it summarized that

23:55What's tax jurisdiction are you in? Let's say UK taxes

23:58Do anticipate paying income tax? And let's say Corporation tax on a company owns the property

24:05Three, are there any planned lifestyle changes? Let's say, I will be getting married on, yeah, okay, cool, next year

24:12Yeah, that's fine

24:18Current monthly living expenses, let's say $5,000

24:24Do you have any own live events? I plan to buy three more properties, each costing 400,000

24:31I realize I've told it the UK and I'm using dollars, so hopefully that doesn't confuse it

24:37Let's see what it does

24:41Okay, cool

24:45So it looks like it thinks it's understood everything, it's summarized it

24:52Okay, now it's got a couple of questions, so what is the current value of your house? Let's say 400,000

24:58How do you plan to finance the purchase? Just say mortgage at 25% deposit and 5% mortgage rate, three, one per five years

25:04Do you anticipate any changes in your living expenses after game mode? Yes, it will double

25:11So if this worked well, it would be really interesting to then have this agent work with kind of like an investment advisory agent

25:18You could have, you could basically build out a, like a multi-agent team for like full kind of financial planning

25:25Okay, so it's come back with even more questions

25:30So how long do you plan to work for retiring? Let's say until I am 70

25:35Do you have any other sources of income or expenses? We Have that, that's a really good question

25:40I'm kind of glad it's come back with that, because in the past, when I've tried to build this, it's kind of just made a bunch of assumptions and run the model before knowing about something else, so that's a good question

25:51Would you like to include any specific goals to your portfolio? Let's say I would like to draw 4% per year

26:03Okay, yeah, so when I've tried to build this previously and run it, I've just, the challenge I found is, when it's been built with a flow builder or as a tool, it kind of becomes a little bit too rigid and just starts to miss a lot of really crucial information, so I

26:16I think this actually seems, like this is better, just having the instructions just in the core instructions and not kind of having too much else going on

26:22It gives the agent a bit more flexibility to do this kind of back and forward, and now it's coming back with the data table and the model, so this is kind of interesting

26:35The downside that I'm kind of seeing here is, if for example this was a tool, we could take very structured data, and we could tell the prompt precisely how we want this table, for example, to come back, whereas this kind of seems like if I ran this three or four times, this would come back a little bit

26:47Different each time, which has its benefit benefits

27:00It takes into account what the user wants in a bit more detail, but it does just mean that the quality of it is going to be a bit sporadic, but generally, we can see this has done a reasonably good job, I think, of taking what we've mentioned into account

27:13And I presume that anything that is kind of not great about this, um, or inaccurate, is going to come down to anything that we missed in the core instructions that we can sort of go back and adjust and refine

27:25So let's just kind of note a few things that we mentioned

27:29I mentioned to the that we wanted to take 4% per

27:33Year and retire at 70, so it looks like it's done

27:38We can see after age 70, top of my head, that looks like about 4% of that portfolio value, and it looks like maybe it's allocated the property purchases in the wrong place

27:45So I guess this is just the classic kind of challenge of LLM prompts, and perhaps changing the LLM model would, if we kind of use something a little bit more intelligent rather than kind of optimized token price or speed

27:54So it might take a little bit of adjusting, but in general, this looks okay, just not highly reliable

28:04But yeah, kind of, I wanted to put this one out there just as a bit of a, like this

28:13One doesn't feel kind of quite right yet, and I sort of struggled to with a few different variations to sort of find a good approach, so any ideas of how you would kind of improve this, yeah, it would be great to hear

28:22But yeah, just kind of another example

28:31So in the next section, we're going to take a look in a bit more detail at pricing, some more limitations of the platform, and how it compares against other alternative agent building platforms

28:39So when it comes to pricing, Relevance AI are pretty reasonable

28:45They've got a free plan that you can use that has a limit of 100 credits a day, that's not a huge amount to put

28:52That into context, the travel agent that we built earlier, each time we run that, that was using about 100 to 150 credits, so you will get through this quite quickly

28:58But what this means is that when you sign up, you don't have to pop a card in, you don't have to worry about being charged before you've built anything that you're actively using, but I would set the expectation as soon as you have an agent that you're kind of actively using or having other people use, you're almost certainly going to need to jump up to the next tier

29:15Now, on their website, the pricing jumps from 0 to 199 a month, but if you're in the platform

29:22Say, I don't know if this is going away, or if this is sort of intentional or not, but there is a, there is a plan in between these, which is the Pro Plan, which is $19 per month as of time of recording this video

29:29As I say, this might change

29:35I suspect there will always be something in between here though, because between the free plan, which is 100 credits per day, up to the next one, which is 100,000 credits, that's quite a leap

29:46And so the Pro Plan is quite nice if you're kind of playing around, or maybe um building sort of like five to 10 agents that are relatively straightforward, 10,000 credits is a reasonably Decent quota now, as you can see, as you go up the tiers a couple of things change

29:59Obviously, the credits that you have increase, the number of users that can access the accounts, if you're running an AI agency, perhaps you've got clients, that's going to increase, and then the amount of knowledge that you feed into your tools increases

30:12I have not come anywhere near exceeding kind of knowledge

30:19I imagine it would be the credit thing would be most likely in most situations to be the thing that is going to determine if you need to jump up to next level or not

30:24Now, in general, in compared to other tools, so things like bot Press stiffy, Zapia, Central, like other kind of no-code AI agent builders, this feels pretty reasonable to me

30:36There's nothing here that kind of makes me think this is significantly overpriced or underpriced, it's all kind of pretty comparable uh to other tools

30:46Now, I do want to touch on limitations, so there's two things in particular that we've kind of touched on briefly already, but I think they are kind of the two main areas of Reven AI that are kind of have their their limitations

30:59So, first one we touched on earlier is the Integrations, at the moment it it feels to me like the the Integrations that kind of on

31:05offer are quite sort of developer focused, and there are some good ones in there, but it's a little bit random

31:17There's a lot of things that are missing that I'd expect, like there is no Google Sheets integration, there's no Notion integration, no Todoist, none, like a lot of things that I'd sort of expect might be sort of within those 10 integrations you might build aren't there, so that's definitely a limitation

31:29One that I expect maybe 6–12 months from this video going live will probably be less of an issue, but definitely something I would be considering

31:42The other limitation with Relevance AI, That that may or may not make this a deal breaker, something we touched on earlier, which is the embed options

31:47So when you have an agent, this is essentially the only UI option that you have

31:53You can embed it, as we kind of mentioned earlier, we can embed it as an iframe, but that's it

32:02There's no customization options

32:12You can, I believe, work with the API to, uh, obviously trigger the agent and potentially catch the payload and display that in a different way, but compared to some of the other agent tools that I've used, the options for doing that kind of easily, like, I, I'd kind of expect maybe here to have a few more

32:21Options to embed this inline and adjust things like CSS, or how it's going to visually appear, but none of that is kind of an option

32:26So, for me, those are the two main reasons why I might consider using a different tool, but besides that, in general, this is a brilliant bit of kit

32:41Like, I think if you're looking to build prototypes of agents, build multi-agent systems, and you want it to take hours rather than days or weeks, this is a really, really good tool to be using, and the amount, the sort of power, is largely going to come down from the brief and how you approach building these

32:49There really is

32:57One of those kind of what you put in, you get out with this tool, and I, I think it's a really incredible bit of kit

33:05In terms of who I'd recommend it for, I'd say particularly if you're an AI agency or running a consultancy, this is a great tool because the speed at which you can kind of build AI agents and agent systems for clients, and you're going to find it pretty tough, I think, to find another tool that is just as powerful, that's also no code and as easy to use, and kind of speed of building these tools is, is kind of just as rapid

33:31I haven't come across anything that's kind of got those three things quite as right

33:37As Relevance AI, but I'd say where it, where I'd kind of consider different tools would be if I need to have a lot of flexibility over the embed options, or if I need something a lot more kind of powerful and flexible, then I might turn more to some framework, something like Flowise or something like LangChain, but the compromise there is my development time, while my deployment options are going to kind of open up, the time it's going to take to develop an agent is going to increase massively because I'm not kind of using this kind of no-code interface

34:09So if you want to try out the platform, I'll drop a link in the

34:16...description below, and if you're interested in the agent space, please do subscribe to the channel where I'll share tips and tutorials on building agents such as this one, where I built a coaching agent that looks at my Daily Journal, critiques my thinking, and then sends me an email.