<b>Flowwise recently added a solution for</b> <b>creating AI

agent teams</b> <b>without needing to write any code.</b>

<b>You can simply add team members to the</b> <b>team by dragging
and</b> <b>dropping nodes onto a canvas,</b> <b>and then
providing the roles and</b> <b>responsibilities to each worker
agent.</b> <b>The team members will then work together</b>
<b>to solve the user's task.</b> <b>This is similar to frameworks
like Crew</b> <b>AI and AutoGen without the</b> <b>need to
write any code yourself.</b> <b>In fact, Flowwise uses</b>
<b>LangRoth under the hood.</b> <b>With an AI agent team, you can
pass an</b> <b>instruction to a supervisor agent,</b>
<b>and the supervisor will then delegate and</b> <b>orchestrate
the tasks between the</b> <b>different team members.</b>
<b>In this video, we will have a look at</b> <b>learning the
fundamentals of creating</b> <b>agent teams in Flowwise</b>
<b>by creating two exciting projects.</b> <b>First, we will
create our very own</b> <b>software development team,</b>
<b>and second, we will create a lead</b> <b>research team that
will make use of tools</b> <b>to achieve their task.</b>
<b>If you're new to Flowwise, then check out</b> <b>this video to
set up</b> <b>Flowwise on your local machine.</b> <b>Also,
if you don't see agent flows in the</b> <b>menu, then please ensure
to</b> <b>update your Flowwise instance.</b> <b>We can
create our very first agent team</b> <b>by clicking on "Add
New"</b> <b>within the Agent Flows page.</b> <b>Let's give
our team a</b> <b>name by clicking on "Save",</b> <b>and
let's call this</b> <b>"Software Development Team".</b>
<b>Let's save this, and let's start by</b> <b>adding a supervisor
to the canvas.</b> <b>Within "Add Nodes", go to the
multi-agent</b> <b>section, and add a supervisor.</b>
<b>This supervisor agent is responsible for</b> <b>delegating the
tasks</b> <b>between the different workers,</b> <b>and
orchestrating the work</b> <b>until the task is achieved.</b>
<b>The supervisor node takes in an LLM as</b> <b>input, and
optionally we</b> <b>can also add input moderation.</b>
<b>We can also give our supervisor a name,</b> <b>but I'll just
leave it as "supervisor",</b> <b>and under additional parameters, we
can</b> <b>see the system prompt for the supervisor,</b>
<b>and we really don't have to change this.</b> <b>This prompt is
simply telling the</b> <b>supervisor that it needs</b>
<b>to manage the conversations</b> <b>between the different team
members.</b> <b>We can go ahead and close out of this</b>
<b>popup, and let's start</b> <b>by adding a chat
model.</b> <b>Under "Add Nodes", let's go to "Chat</b>
<b>Models", and for this video,</b> <b>we're going to add
the</b> <b>"Chat OpenAI Model", like so.</b> <b>Let's
connect our chat</b> <b>model to the supervisor.</b>
<b>Then within the "Chat OpenAI" node, let's</b> <b>select our
credentials,</b> <b>and I've already set up my OpenAI</b>
<b>credentials, but if you</b> <b>haven't done so
already,</b> <b>simply click on "Create New", give your</b>
<b>credential a name, and</b> <b>paste in your OpenAI API
key.</b> <b>You can also check out this video, which</b>
<b>shows you how you can</b> <b>create your OpenAI API
key.</b> <b>Simply select your credentials from this</b>
<b>drop-down, then under "Model Name",</b> <b>go ahead and select
the new GPT40 model.</b> <b>Let's also change the</b>
<b>temperature from 0.9 to 0.</b> <b>Now we can start
adding</b> <b>team members to this team.</b> <b>Let's go
back to "Add Nodes", then under</b> <b>"Multi-Agents", let's</b>
<b>add a "Worker" node.</b> <b>The "Worker" node can take in a
list of</b> <b>tools as input, and we also need to</b>
<b>assign a "Supervisor" node.</b> <b>So let's go ahead and
attach our</b> <b>supervisor to this worker.</b> <b>We can
also add a chat model to this</b> <b>node, and this is
optional.</b> <b>If we don't attach a</b> <b>model to the
worker itself,</b> <b>the worker will simply inherit the chat</b>
<b>model that is being used</b> <b>by the supervisor
node,</b> <b>so this OpenAI node over here.</b> <b>But if
we wanted to override the model</b> <b>that is being used by</b>
<b>this specific worker,</b> <b>we can simply attach a</b>
<b>different chat model to this.</b> <b>This can be very useful
for creating a</b> <b>team where different models need to</b>
<b>verify each other's results.</b> <b>For this demo, we'll
simply</b> <b>inherit the same chat model.</b> <b>Let's
give our worker a name.</b> <b>For this team, I</b>
<b>actually want three workers.</b> <b>First, I want a product
designer that is</b> <b>responsible for coming up with the</b>
<b>feature set for our application.</b> <b>The second worker will
be the software</b> <b>developer that is</b> <b>responsible
for writing the code.</b> <b>And the third worker</b>
<b>will be the technical writer</b> <b>which is responsible for
writing the user</b> <b>manual for our application.</b>
<b>Let's start by</b> <b>creating our product designer.</b>
<b>I will call this worker product designer,</b> <b>and for the
worker prompt,</b> <b>we can specify the</b> <b>roles and
responsibilities.</b> <b>Let's enter something like, "You work
for</b> <b>a software development company."</b> <b>"Your
role as product designer is to come</b> <b>up with the features and
product design</b> <b>based on the user requirements."</b>
<b>Let's save this prompt, and let's have a</b> <b>look at the
rest of</b> <b>the inputs on this node.</b> <b>We can also
provide</b> <b>variables in this prompt template</b>
<b>which we can then specify</b> <b>within the prompt
values.</b> <b>And we will have a look at</b> <b>that in
the second example.</b> <b>We can also specify the max</b>
<b>iterations or leave it blank.</b> <b>This will simply tell the
application how</b> <b>many times this node</b> <b>can go
back and forth</b> <b>and make changes to its solution.</b>
<b>Let's simply leave this blank.</b> <b>We can already test our
team out by</b> <b>saving the chat flow.</b> <b>We can then
click on this chat bubble.</b> <b>Let's expand this, and let's ask our
team</b> <b>to build a to-do list app.</b> <b>First, you
will notice a</b> <b>message from the supervisor,</b>
<b>and the supervisor is passing the request</b> <b>on to the
product designer</b> <b>who is now currently</b> <b>busy
with their task.</b> <b>And after a few seconds, we can see
the</b> <b>response coming back</b> <b>from the product
designer.</b> <b>Let's have a look at this.</b> <b>And the
product designer did indeed list</b> <b>all the core features</b>
<b>of our to-do list app.</b> <b>And this also includes some
advanced</b> <b>features, the user interface design, the</b>
<b>tech stack, and more.</b> <b>Now let's add our</b>
<b>other two workers as well.</b> <b>So under add
nodes,</b> <b>let's add another worker.</b> <b>Let's call
this worker the developer.</b> <b>Let's attach our</b>
<b>supervisor to our developer as well.</b> <b>And for the
developer's prompt, let's</b> <b>enter something like,</b>
<b>you work for a</b> <b>software development company.</b>
<b>As an experienced software developer,</b> <b>you build
software solutions using React,</b> <b>Next.js, Tailwind, CSS,
Express,</b> <b>JavaScript, and Node.</b> <b>Let's save
this, and let's also add our</b> <b>final worker, which is
the</b> <b>technical document writer.</b> <b>So let's
attach our</b> <b>supervisor to this node.</b> <b>Let's
call this technical writer.</b> <b>And for the worker prompt, let's
enter.</b> <b>You work for
a</b> <b>software development company.</b> <b>Write a
detailed user manual based on the</b> <b>product design and
code</b> <b>written by the team members.</b> <b>The manual
should be in markdown format.</b> <b>It's safe as prompt, and</b>
<b>let's test our flow by saving it.</b> <b>And in the chat,
let's expand this.</b> <b>Let's clear the chat, and let's ask
it</b> <b>again, build a to-do list app.</b> <b>So we can
see the supervisor passing the</b> <b>message to the product
designer.</b> <b>What is impressive about this is the</b>
<b>supervisor is able to correctly determine</b> <b>the sequence
of events.</b> <b>It wouldn't make sense to ask the</b>
<b>document writer or the</b> <b>developer to go first.</b>
<b>So after we receive the product design,</b> <b>the supervisor
now passes the product</b> <b>design over to the developer.</b>
<b>The developer now put everything in place</b> <b>to build a
software application,</b> <b>and it's also produced the code.</b>
<b>We can now see the supervisor requesting</b> <b>the technical
writer to</b> <b>write the user manual.</b> <b>And finally,
we can see the manual that</b> <b>was produced by the technical
writer.</b> <b>And this is in markdown format.</b>
<b>Building an AI agent</b> <b>team couldn't be any
easier.</b> <b>Now let's have a look at the second</b>
<b>example, where we will build a lead</b> <b>research team
that</b> <b>will be able to use tools.</b> <b>But first, if
you like this video, hit</b> <b>the like button and</b>
<b>subscribe to my channel.</b> <b>Also, let me know in the
comments if</b> <b>there are any scenarios or use cases</b>
<b>that you would like me to</b> <b>cover in a future
video.</b> <b>Let's go back to the dashboard and let's</b>
<b>go to agent flows and let's</b> <b>create our second
example.</b> <b>Let's save this.</b> <b>Let's call this
lead</b> <b>outreach team and let's save this.</b> <b>Let's
start by adding our supervisor.</b> <b>So we'll drag and</b>
<b>drop the supervisor node.</b> <b>The supervisor also needs a
chat model.</b> <b>So let's grab the chat open AI node.</b>
<b>Let's attach the open AI</b> <b>node to the
supervisor.</b> <b>Let's also select our credentials.</b>
<b>Let's change the model to GPT40.</b> <b>And let's set the
temperature to zero.</b> <b>Now this team needs two workers.</b>
<b>So let's go ahead and add those workers</b> <b>and I will
explain</b> <b>what each of them will do.</b> <b>So let's
add our first worker.</b> <b>Let's duplicate this and let's attach
the</b> <b>supervisor to these two workers.</b> <b>So we'll
do this and we'll</b> <b>attach it to this worker.</b>
<b>Great.</b> <b>I'm going to clear the prompt</b>
<b>templates for each of these.</b> <b>And let's give our first
worker a name</b> <b>like lead researcher</b> <b>and for
the second one,</b> <b>we'll call it the lead</b> <b>sales
representative.</b> <b>So what we want to do here is we want
our</b> <b>researcher to go online</b> <b>and build out a
detailed</b> <b>report of a specific lead.</b> <b>So
effectively we want to tell this</b> <b>researcher to go
online</b> <b>and build out a full report for a</b>
<b>specific individual.</b> <b>This report needs to contain
information</b> <b>like the professional experience,</b>
<b>their involvement in</b> <b>projects and companies,
etc.</b> <b>And the lead sales representative is then</b>
<b>responsible for</b> <b>generating personalized
messages</b> <b>based on the results from the researcher</b>
<b>for that individual.</b> <b>So one obvious issue here is that
this</b> <b>researcher will have no way</b> <b>of going
online and performing research.</b> <b>It will only be able to produce
its</b> <b>report based on the data</b> <b>that it was
trained on,</b> <b>which is not up to date and</b> <b>will
not work in most instances.</b> <b>So what we'll do to solve this
problem is</b> <b>to add a tool to this worker</b> <b>that
will allow it to</b> <b>perform Google searches.</b> <b>And
we will get back to that in a second.</b> <b>Let's first complete these
prompts.</b> <b>So for the first worker, I'm actually</b>
<b>going to paste in this prompt</b> <b>and feel free to pause
the video to have</b> <b>a look at this prompt.</b> <b>But
this is something that you could very</b> <b>easily generate with chat
GPT.</b> <b>This is simply a role description for a</b>
<b>lead research agent.</b> <b>And this is simply saying that
this</b> <b>researcher needs to use</b> <b>all the tools
available to it</b> <b>to build out a detailed report for a
lead</b> <b>person working for a lead company.</b> <b>Let's
save this and let's do the same for</b> <b>the sales
representative.</b> <b>And this is simply telling this agent</b>
<b>that is responsible</b> <b>for writing personalized</b>
<b>messages for the lead person.</b> <b>So feel free to pause the
video and have</b> <b>a look at this prompt.</b> <b>Again,
your prompt does not have to be</b> <b>nearly as complicated as
this,</b> <b>but this prompt does</b> <b>seem to work quite
well.</b> <b>Let's save this.</b> <b>And you might have
noticed that in these</b> <b>prompts, we are using these curly
braces</b> <b>and these represent variables, as we will</b>
<b>be entering these variables</b> <b>by going to format prompt
values.</b> <b>Within here, we can now swap out these</b>
<b>values and I'm going to</b> <b>call my company Lion
Research.</b> <b>Let's accept this.</b> <b>And for the lead
person and company, we</b> <b>want to perform</b>
<b>research for Harrison Chase,</b> <b>who works for Lang
chain.</b> <b>So for the company,</b> <b>let's enter Lang
chain.</b> <b>And for the person, let's</b> <b>enter
Harrison Chase, like so.</b> <b>Let's close this pop up and let's do
the</b> <b>same thing for the sales representative.</b>
<b>Our company is called Lion Research.</b> <b>The lead company
is called Lang chain.</b> <b>The lead person is called Harrison
Chase.</b> <b>And for the lead activity, we</b> <b>will
focus on the Lang graph.</b> <b>And as a reminder, this agent
flow</b> <b>solution in Lang flow is</b> <b>actually based
on the Lang</b> <b>graph framework that was</b>
<b>developed by Lang chain.</b> <b>Let's close this.</b>
<b>Let's save this flow.</b> <b>And we can go ahead and save
this,</b> <b>although we haven't provided</b> <b>a tool for
the researcher yet.</b> <b>And we will do that in a second.</b>
<b>Let's just see if this is</b> <b>working so far by just typing
go.</b> <b>Our supervisor is asking the researcher</b>
<b>to write out that</b> <b>detailed report for our
lead.</b> <b>And after the researcher is complete, the</b>
<b>supervisor is now asking</b> <b>our sales
representative</b> <b>to draft up those personalized
messages.</b> <b>And we can indeed see those different</b>
<b>messages being produced.</b> <b>Now the issue here is this
researcher</b> <b>produced this report based on the</b>
<b>training data that is</b> <b>baked into the open AI
models.</b> <b>This data could very much</b> <b>be outdated
and inaccurate.</b> <b>We can greatly improve the results by</b>
<b>giving this researcher</b> <b>access to a Google search
tool.</b> <b>Let's do that now by going</b> <b>to add nodes
and under tools,</b> <b>we can see a list of different tools
that</b> <b>we can assign to any of these workers.</b>
<b>For example, you could use a chat flow</b> <b>tool to allow
the user to</b> <b>call any of your chat flows</b> <b>in
flow wise, or you can use the custom</b> <b>tool to write your
very</b> <b>own tools that could call</b> <b>service
endpoints, for</b> <b>instance, to Zapier or make.</b>
<b>But what we're interested in is this</b> <b>Google custom
search tool.</b> <b>Let's add this to the canvas and let's</b>
<b>connect this tool to our worker.</b> <b>It's also useful to
know that you can</b> <b>attach more than one tool to a
worker.</b> <b>So as an example, let's add this</b>
<b>calculator tool, we can</b> <b>take this tool and also
attach</b> <b>it to the same worker and the worker will</b>
<b>decide which tool to use and when.</b> <b>Right, so now to get
this custom search</b> <b>tool working, we need to</b>
<b>provide some credentials.</b> <b>Click on this drop
down.</b> <b>And for you, there won't</b> <b>be any
credentials yet.</b> <b>So click on create new, then</b>
<b>give your credentials a name.</b> <b>I'll call this Google
custom search</b> <b>YouTube tutorial, but you can</b>
<b>call yours whatever you want.</b> <b>Then we need to provide a
Google custom</b> <b>search API key, as well as a</b>
<b>programmable search engine ID.</b> <b>These are very simple to
set up.</b> <b>First, let's go to the</b> <b>Google Cloud
console.</b> <b>You will be asked to sign</b> <b>in with
your Google account.</b> <b>Afterwards, you should</b>
<b>see some sort of dashboard.</b> <b>Let's create a new project
by clicking on</b> <b>this drop down and</b> <b>let's
create a new project.</b> <b>Let's give our project a name.</b>
<b>I'll call mine flow</b> <b>wise agent flow tutorial.</b>
<b>Let's click on create.</b> <b>We can click on this drop down
again.</b> <b>And then let's select the</b> <b>project that
we just created.</b> <b>Now go to the navigation menu, then
click</b> <b>on API and services,</b> <b>and then click on
library.</b> <b>From the library, let's</b> <b>search for
the custom search API.</b> <b>Click on the custom search API.</b>
<b>And then lastly, click on enable.</b> <b>Once this has been
enabled, click on</b> <b>credentials, then click on create</b>
<b>credentials, and then API key.</b> <b>Let's copy this key and
then back in flow</b> <b>wise, paste in that API key.</b>
<b>So let's go ahead and</b> <b>create our search engine
ID.</b> <b>Let's click on this link and here simply</b>
<b>enter the name of your search engine.</b> <b>I'll just call it
flow wise agent flows.</b> <b>And then under this section,
select</b> <b>search the entire web.</b> <b>And under
search settings,</b> <b>I'm going to enable safe search.</b>
<b>I'm also going to tell it that I'm not a</b> <b>robot and
let's click on create.</b> <b>Now we can click on</b>
<b>back to all search engines.</b> <b>We can then click on the
search engine</b> <b>that we just created.</b> <b>And from
here, we can</b> <b>find our search engine ID.</b> <b>Let's
copy this.</b> <b>And now we can paste this into this field</b>
<b>and let's click on add.</b> <b>Ensure that you've selected
the</b> <b>credential that you just</b> <b>created and save
your chat flow.</b> <b>So let's run this again.</b> <b>But
this time we're expecting the</b> <b>researcher to use the Google
custom</b> <b>search to perform its research.</b> <b>I'll
just enter something like go.</b> <b>The supervisor is passing the
instruction</b> <b>over to the researcher.</b> <b>Once the
team has completed their work,</b> <b>let's go back up to the lead
researcher.</b> <b>We can see that the lead researcher used</b>
<b>the Google custom search tool four times.</b> <b>When we open
up these tools,</b> <b>we can see the search query</b>
<b>that was executed and the</b> <b>response from the Google
results.</b> <b>And we can see that for each</b> <b>and
every one of these tools.</b> <b>It is impressive to see how
this</b> <b>researcher determined</b> <b>which search
queries to use.</b> <b>And this time the results from the
model</b> <b>is actually based on up to date</b>
<b>information from Google.</b> <b>I hope you enjoyed
this</b> <b>video and I would like to know</b> <b>what use
cases you would like me to cover</b> <b>in these tutorial
videos.</b> <b>If you would like to learn everything</b>
<b>there is to know about Flow-wise,</b> <b>then check out
this</b> <b>tutorial series over here.</b>