Agent Canvas Windows Install Guide (NPM, No Docker) + First AI Agent Chat
201 segments
Agent Canvas is the latest way to use
open hands. Let's learn how to set it up
on Windows using NPM and then set up
GitHub and Slack integrations. Windows
installation with auto sandbox. You'll
need Node and UV installed. Please note
that if you run the agent server
directly on your machine without a
sandbox, the agent will have full access
to your file system. These are the
commands that you'll execute. Once you
have your terminal open, go ahead and
run NPM install {dash} G for global at
open hands {forward slash} agent {dash}
canvas.
>> [music]
>> Once agent canvas is fully installed,
then you can run agent {dash} canvas.
When you see localhost, you know you're
ready to get started. Getting started.
Setting up agent canvas and starting
your first chat. Navigate to
localhost:8000.
[music]
You'll then go ahead and go through the
prompts using your agents. I'm going to
choose open hands.
You'll check your back end. This time it
will be local. Just click next.
Then you'll pick your LLM provider. In
this case, we'll use Anthropic with the
Claude Opus 4.7. I'm going to input my
API key and click next.
From here, you can see our proven
workflows that you can try out with
linear, Slack, Notion,
or you can enter your own prompt. We're
going to create a basic web page
explaining what open hands can do. We're
going to click send and that will open
up a new conversation.
The agent [music] is going to start
working on the prompt given, in this
case creating a web page. It will reply
back in the chat, and you can preview
the web page on the right-hand side.
Getting started. Adding open hands cloud
back end. Let's go ahead and hover down
to the local area and click add back
[music] end.
We're then going to log in to Open Hands
Cloud. You'll see a new tab pop open
with some letters. That's the name of
this connection. You don't need to save
that. We're going to click authorize and
then we'll return back to our agent
canvas. Then you'll be able to see that
we are now connected to Open Hands
Cloud. Create an LLM profile.
First you'll want to navigate to your
settings and click LLM.
From there you'll click add LLM profile.
You can give that profile a nickname.
In this case we are going to call it
MiniMax.
From there you'll pick an LLM provider.
I'm going to pick Open Hands.
>> [music]
>> I'm going to pick the model MiniMax M.3
and I'm going to put my API key in
>> [music]
>> and click save.
Once you've saved, now you have a couple
of different LLM profiles and when you
create a new chat, you can pick which
profile you want to start with.
Creating an automation from [music]
scratch.
You can create an automation just by
starting a new chat. In this case
[music]
I'm going to create a new automation
where every day at
10:00 a.m. Eastern, I'm going to have
all my GitHub notifications [music]
summarized and presented to me in a
chat.
From there you can use any LLM model
that you want. I'm using GPT. If it's a
simpler request, you can also use
something like MiniMax.
And then the agent will go ahead and get
started on that task. Once it's done,
I'm going to return to the automate tab
and I'm going to go ahead and trigger it
to run now only for [music]
demonstration purposes.
Then a chat get kicked off. We're going
to open it to follow along and once it's
done, it'll return that summary of my
GitHub notifications.
In the future, you could even set it up
with Slack so that it would send you a
message. Adding an agent canvas to your
Slack server. You'll want to navigate to
api.slack.com/apps. You'll want to click
create an app
and then from scratch. You'll then need
to give this app a name. Whatever you
want is fine. And then you'll need to
pick a workspace. You'll need to be
admin of that workspace in order to
create an add an application.
Then click create app.
It'll have you return back to this page.
You'll want to scroll down and go to
OAuth and permissions.
Then you'll need to scroll down
and you're going [music] to want to add
an OAuth scope.
You'll see here that these are the
scopes that you need to add for this to
work.
Once you have all of them, scroll up to
install to dummy hands or whatever your
workspace is called and click allow.
Afterwards, you'll scroll down and
[music]
there will be a bot user OAuth token.
Please save that for later. [music]
Return to your workspace and click the
person icon.
From there, you'll want to go to the
integrations tab
and you're going to click add an app.
Once the new page opens up, you're going
to click add for the app which we named
agent hands. You'll want to return back
to agent canvas and click automate in
the left hand. You'll then click Slack
channel monitor. You're going to put in
your team [music] workspace ID, which
you can find in your URL, and you'll
also need to add that OAuth bot token we
saved earlier.
Once you click install, it'll open up a
new chat with a prompt. [music] If you'd
like, you can go ahead and change the
default name and then submit it to start
the chat.
The agent is going to come back and ask
you a few questions like what channel,
in this case we want general, and also
what name to monitor for, and we are
going to use Joe-Bot.
When it's done, you'll see something
similar to this be returned in your
chat. You can then navigate back to
Automate and click on that Slack channel
monitor to see that it is active.
>> [snorts]
>> Now we'll return back to our workspace
and go ahead and send a message [music]
for our bot.
Once the bot has seen this, it'll react
with some eyes and start a thread.
You can either go to the localhost URL
or wait for the agent [music]
to reply back with your answer in Slack.
Adding GitHub to your agent canvas.
First, navigate to the Automate tab side
and then go ahead and pick a GitHub
workflow that you want to implement. I'm
going to pick [music] the GitHub code
review and I'm going to input my
personal access token
>> [music]
>> and click install.
A new chat is going to open with a skill
and I'm going to enter
and then I will need to answer a few
questions like how often do I want this
automation to poll? I'm going to pick
every 5 minutes. I'm also going to have
the agent comment on PRs that are ready
for review, but not drafted.
And also put the agent's review as a
comment to that PR. [music] And then I'm
going to pick a repository for this
automation to monitor.
Now that the automation is done, just
for demonstration, I'm going to go ahead
and go back to the automate tab, and I'm
going to go ahead and click run now so
that I can get a review for my PR.
Once the job is done, I can go to GitHub
and I can check out the comment that was
left.
Join our Slack if you have any questions
or comments. Thanks so much for
watching.
Ask follow-up questions or revisit key timestamps.
This video provides a comprehensive guide on installing and utilizing Agent Canvas. It covers local Windows installation using NPM, setting up various LLM profiles, and creating custom automations. The tutorial also includes detailed walkthroughs for integrating external platforms like Slack for channel monitoring and GitHub for automated code reviews.
Videos recently processed by our community