We have now launched Agent Canvas as a

production ready product. And we are

excited to talk more about that. We have

Devin here who will lead on.

>> Yeah, I think I've talked about this the

past two

community meetings at various stages of

like it's in progress. Here's what it is

and then it's in progress. It's in beta

and now it is fully launched and out

there.

Can you go to the next slide, Jamie?

Awesome. So if you haven't seen it

already, hopefully you have and got a

chance to use it. Agent Canvas is our

new hotness, right? So this is a new

face of Open Hands. It's local first

visual workspace run and automate AI

coding agents. That's sort of the

elevator pitch, right? It's your sort of

front end to do all the things with a

flexible system to sort of build your

own,

you know, best practice tool. So just

some of the highlights. This is

certainly not everything, but it's sort

of I went through like our announcement

post and things like that and tried to

distill it into a few bullet points

here. So parallel agents, the idea is

that very easily you can run multiple

agents across multiple backends and, you

know, do it all at once. We support get

work trees. So when you create a new

conversation,

um,

you can have the option of using a work

tree so that stuff happens there kind of

in a silo and it doesn't affect your,

you know, live branch.

Um, and then we also have automations

built in. So we're really excited about

everything that you can do having a

really full featured automation system

and an automation server built in to,

you know, the same place where you would

work and talk with your LLMs. So you can

plug into all your MCPs and use the same

secrets and things like that and it can

do all kinds of really cool stuff there.

So,

you know, by default you can do like

just, you know, cron at different times.

You can do polling so that you have an

automation kicking off like every minute

to check something. That lets you build

like Slack bots and PR reviews and

things. I've been a big fan of using

this on my own Slack instance to have a

Slack bot that I can talk with and run

and do things.

And then, you know, of course, one of

the really cool things is you can bring

your own agents, not only work with

whatever LLM you want to via API, but

the agent team did a really, really good

job of adding support for ACP so that

you can actually use Claude, Codex,

Gemini,

uh anything you really want inside

instead of just open hands as the full

agent. So, um I just talked with a user

this morning who needed that, didn't

even know that he needed it, uh but when

he first started he was like, "Oh, I can

just set this up." and didn't even know

what ACP was before and was able to sort

of like put together so he could use

what he wanted to across like I think

four or five different virtual machines

with different, you know, setups in each

one. So, very very very flexible. And

then of course like local first but not

local only. Um sorry, Jamie's last

bullet point. Um

again, that was sort of like I did

some user interviews earlier this year.

Leonardo was one of the folks that I

talked with as well.

Just asking, you know, what are you

doing these days with agents? What do

you need? What, you know, and one of

those those key things was like, "I need

a local, you know, GUI that's even

better than what we have, but I also

need it to be not just local but, you

know, able to work on cloud environments

and in virtual machines and so I can

sort of like do that work on the go from

a device." So, that was sort of like one

of the

cornerstones

that we started with and also delivered.

So, you can you can do that now. You can

connect to Open Hands Cloud, run it on

your laptop, set up a computer at home.

You can have a really expensive What is

it? Nvidia Spark. So, you have like a

local LLM and you can have your own

siloed local experience. I mean, there's

a lot of flexibility.

Awesome. Okay, now next slide. Great. Um

so, with that I wanted to just sort of

like talk about

where we are now, what what's sort of

focus and then where we hope to go. This

is not just like me, you know, laying

this out there. This is sort of coming

from community conversations and

different inputs. So, the focus right

now that I've sort of put forth is

because the agent team is still working

on some of the ACP stuff. There's a

handful of tickets, so we're doing some

polish there. And there's also agent

profiles that just got merged into the

SDK. And Simon has been working on

getting that also into Agent Canvas. So,

some really exciting stuff there. Um

hopefully we'll see that sort of ship

next week and that's kind of like the

first release that will come afterwards.

There'll probably be a handful of other

items uh, in that release, but that's

sort of like the the anchor for the next

release is what I've been sort of like

calling it, like maybe like one anchor

per release, so that you have a nice

marketable moment.

And then coming up after that, so like

what's up next? Um,

something that I I've I've heard is

automations are pretty good, but they

could be better and we've gotten some

feedback on what that might mean. So,

sort of the free bullet points here are

make it easier to set up and run, um,

potentially adding like a UI to make it

like easier to fine-tune without having

to talk to an agent. There's a lot of

flexibility when you can just say, "Hey,

agent, create this automation that does

this thing, check this repo at this

time, or hook up with this MCP, whatever

you want." But, um, there's some

hesitation about being able to see

exactly what's going to get set up and

stuff. So, UI is one thing that we're

exploring. Um, and then just making them

more observable end-to-end, so um,

looking at adding like a dashboard view,

so you can see like all of your

automations, what their process is, like

where they are, how often they run, if

they've run into any um, errors, and

like if they've created artifacts.

Um, and then just a general like a

deeper SDLC integration, so you can

actually use this as part of, you know,

whatever that looks like for you, you

know, whether it's taking like um,

tickets and then turning that into

issues or just doing some research or PR

reviews, just kind of generally making

that easier.

So, that's sort of the the next bundle.

There's not a clear epic or um, a

singular

a breakdown of what this is going to be.

So, if you have feedback here, I would I

would love to hear it. If there's a PR

that you've already gotten in,

um,

that's like out there that just hasn't

been reviewed or something that's

dealing with this, that'd be great.

And then of course that the most

important one here, um,

in planning what comes after that, I

think that this call and this community

is where we're going to have the the

biggest guiding light, right? Like so,

this is all just like a starting point,

um,

and your your feedback and contributions

are really going to help shape like what

comes after this. Uh, so just a couple

of highlights here. Um, Christopher P,

he's given a ton of feedback on like the

importance of sandboxing and being able

to use like the the current app server

that um, OpenHands has. Uh, so he's done

some work there and I know he he just

posted 15 minutes ago um in the the

project um Asian Canvas Slack that um

he's got a PR coming to use the current

um app server as a back end. So, just an

example of something like, "Hey, this is

something he really wants to see and

he's going to make it happen, you know,

advocating for it." Um

we had a another user, Shane War 1982.

He posted a vision of like a boss

orchestrator. Um we haven't gone further

than that. There's no other PRs, but

it's sort of like he was like, "Hey,

this is my vision of what I want. Like,

I want a boss that I can talk to that

then spin out all these conversations

across my multiple back ends."

I don't know what that could look like,

but I'd love to hear more about it and

um if it's something that others, you

know, resonate with,

you know, we can make it happen.

Um and then Engel has done uh a really

great draft PR on like um

like add-ons for tool visualizations. Um

Vasco did some work that could um make

that I think possible and then Engel

looked at it and said, "Yeah, I think

this could work." I've got a shaping

ticket out there for like a whole

add-ons or plug-in like extension system

for Asian Canvas that I'd love to see

happen. Um so, he just did some initial

work like, "Hey, here's what it could

look like and here's how it could work."

So. I've talked a bunch. Any any

questions on this? Anybody wants to, you

know,

jump in?

>> So, I have questions if no one else has

questions. Um you mentioned if there's a

PR that needs to be reviewed or like

want more attention. I shared your

GitHub handle and also the Slack.

But, I guess my other question is

what specifically do you think is the

best place for contributors to help us

with Asian Canvas? Like, where's the

best place for them to get plugged in,

especially if they're newer and a little

bit overwhelmed because of all the

options?

>> That's a good question. I mean, I have I

have my opinion about that. I would say

Slack's a great starting point cuz you

can jump in and say, "I'd love to get

started." And then generally, I think

folks get pointed to GitHub like, "Hey,

dive in, look at it."

Probably the best is if you care about

something, then

try to get involved there. Uh if you

know, you see a problem that you want to

fix where you want to add something,

that's a good one. I know that you did a

bunch of work to make like great first

issues. If you just want to look at

anything to get involved in, but I'm

curious to hear if anybody else has um

ideas on like what what's worked for

them or you know what they think could

be the best you know path forward to be

more involved.

>> All right, I think that's

>> Yeah, if you have any ideas feel free to

drop them in the chat. We're going to

keep going. I did share a shortened link

and this is how you can easily find good

first issues. I would say something that

comes up in conversation on Slack is

that the best way to find things to work

on is to start using Agent Canvas. As

you use Agent Canvas for like your

workflow, you'll either see that there's

a small UX bug that you can create an

issue and maybe even fix or you could

say, "Hey, I really wish that we had

this feature. Let me create an issue.

Let me talk about it in Slack." So I

think that's just another way to also

get started.