Hi everyone, this is Graham from Alhians

AAI and we have an exciting new uh model

that we'd like to share with everyone.

It's the Devastrol model that we work

together with Mistl to create. And what

I want to show you here is how we can

use this model uh completely locally on

our computer to create an AI agent that

allows us to go in and you know fix bugs

or implement features or something. So

the first thing that we need to do is we

need to download the model. Uh there's a

couple ways you can do this. You can do

it through the hugging face uh CLI or

just go and download it

directly. But I in the sake of time I've

downloaded the model. So we have it

right here. And I'll I'll go into this

directory. And we can see we have the

gguf uh format uh that we can do. And

for using the model, I'm going to use

lm's lm studio. And so what we do here

is we do lm's

import. And then we do the name of the

model. And in terms of the name of the

model, let us call that

um devastrol Q4 km

ggf. And so yeah, this is importing the

model and so I'll do

yes and uh I will do don't

categorize. So great. Now I've imported

the model into LM Studio and I'm going

to look over here and we can see indeed

the model was imported into LM

Studio. And so then we can go to the

developer settings and select the model

to load. And what we will want to do

here is we'll want to load it with some

custom settings. So I on the Mac I hold

down the option key and click here. This

gives us some custom

settings. The next thing we want to do

for context length is we want to make

sure the context length is not too long

because if it uh is too long, it might

use all the memory of my computer. Uh

but if it's too short, then we won't be

able to get all the nice long context

that's needed in order to do well at

coding tasks. So I'm going to copy in

something about uh 32,000 here which is

kind of a good number that balances this

off. So loading in the model now and

this will take a little

bit. So once we have the model loaded uh

the next thing that we need to do is we

need to set up some configuration so

that you can actually run the model uh

together with open hands with the model

as a back end.

And we're basically going to have to run

some commands. Uh the first command that

I will run is just uh exporting the

openhands

version. The next thing that I will do

is I will copy the model name from uh LM

Studio. So I just click in here and copy

that. And then I do export LM studio

model

name. And uh it's called

uncatategorized, but it's actually the

Devstral

model. And then the final thing I want

to do is I want to copy in the LM Studio

URL. So there we go. And don't worry if

you can't follow along with all of this.

I will put a link uh below the video

where you can click through and see all

of these steps step by

step. So the next thing I need to do is

I need to get the appropriate version of

OpenHands uh through a Docker pull. So

now I'm pulling this through Docker. Uh,

I actually did this before I started the

demo to save some time, but it might

take you a little bit of time if you do

it

yourself. And then the final thing uh

before we actually start running the

program is I'm going to export some uh

settings uh so that open hands already

starts up with the model set by default.

So I'm just going to do that there and

that's just a copy

paste. Okay, so let's get started. Um,

now we're starting up the open hands uh

kind of coding agent and we can click

through to this URL that it provided

with us and here we

go. So this opens up the repository and

I am going to launch something here and

it's uh starting up the container that

we will be using to run the agent in.

And this will take a little bit uh the

first time you run

it. Okay. And now we can see that the

agent is awaiting user input. And so the

next thing we need to do is we need to

come up with a task for the agent to do.

So what I would suggest you do here is

you just come up with the next kind of

coding task that you would be interested

in doing. So for me, you know, I'm

constantly coding on open hands. So I

have this uh pull request open that I

started but unfortunately it is not

passing the tests here. So we have the

Python unit tests and so I

will take this PR

ID and

say please clone the repo associated

with this pull

request and check out the branch. then

get started

working on running piunit tests. That's

the name of the testing file and fixing

any

errors. So right now the agent has

started working. Again, this will take a

little bit of time the first time you

run it uh because it will be reading in

uh kind of the long instruction that

the agent is provided with. And it

particularly takes a long time on my

computer because I have a old Mac M1

processor. So, I'll I'll wait, you know,

a minute or two for this to boot up.

And there it got started. So we can see

it's cloning the repository and checking

out the branch that it needs to be

cloning. And the best part about all of

this is we're not using any external uh

API services. Uh the only thing that

we're doing is we're generating directly

on my computer and it's uh just my Mac

M1 using the the M1 processor.

And we can see that uh you know the the

model is running here and all of this

was generated with the

model and it's uh successfully cloned

the repo. So we can see that it cloned

the repo in here and now it's going to

get started on uh checking out the

branch and

uh looking for the the unit tests and it

is off to the races.

So the great part about open hands is uh

it's you know completely local. You can

run it on your computer. Typically we do

it through calling an API like quad or

gemini but now with the new devstral

model you can run it entirely locally on

even my you know several years old

computer

and we can you know work and and make

the models better and better until you

know everybody can have their own coding

agent on their computer that does

whatever task they want. So, uh, hope

you're excited about this as I am. And

if you're interested, check the link,

uh, that you will see in the video and,

uh, below the video to learn more about

Open Hands in Devstral. Thanks a lot.