LM Studio Bionic: què fa l'agent d'IA local i quins límits té

If you love local AI, you're going to

love the new LM Studio Bionic. Bionic is

a brand new free app that was released

by LM Studio, and it replaces tools like

Claude Code and Claude Co-work for when

you want to use local models to do the

work for you. So, on top of local models

that you have downloaded and running on

your own devices, you can also use other

open models like Kim K3 or GLM 5.2, but

when you're using these models, you do

have to set up billing. You do have to

pay for these because they're hosted on

LM Studio servers. Now, if you're in the

local AI game, you've probably heard of

LM Studio, but in case you haven't, a

quick summary is that this is a free app

that lets you download and run local

models on your devices. You're able to

chat with the models directly within the

app itself. If you set up LM Studio

across two or three different devices

that you have, you can actually share

the AI between devices. That's really

cool. And then the final cool thing is

that you can actually get a local host

URL that you can then use to plug your

local model directly into a tool like

pi.dev, dev, open claw, Hermes agent,

clawed code and then you can power those

tools with the local model running on

your computer. So now when it comes to

the functionality of bionic well bionic

can do all those things that we just

mentioned in the original LM studio app

plus it can do more. The positioning of

bionic is that it actually has two main

modes. It has the work mode which is

like the claude co-work and it also has

the code mode which is like the claude

code. The work mode creates a virtual

environment on your computer, very

similar to how Claude called Coowwork

does. And you're then able to create and

edit documents directly from your uh

computer. So basically, whatever you

create, whatever you edit lives on your

device. We also get access to MCP.

There's a couple of pre-built MCPS

already. I'll go through an example of

how to set up a new MCP. And then from

our work session, we're able to interact

with that MCP as well. And then for

Bionic Coding, we're able to create

apps, automations, and do computer

control. And we also have this built-in

preview browser panel directly into the

app. So, as we're building our app,

we're able to preview it and then edit

it live as well. The next update is that

you can use real-time voice

transcription inside LM Studio Bionic.

So, this actually downloads a two or 3

gig model onto your device and then you

can just click the um the voice mode

button and it transcribes whatever you

say into the text input box. And then,

like we mentioned before, this is

natively local. So, just like in the LM

Studio app, you're able to browse and

download different models onto your

devices as long as they fit, of course.

And it works across the GGUF and MLX

formats because it's got multiple

different runtimes running under the

hood. And then of course we have open

models, the Frontier open models. We

have GLM 5.2, Kim K3, Deepseek V4. All

of this is actually run on I think US

servers, but you would have to just

double check. And it has zero data

retention as well. Actually, let's just

quickly check now. So, for example,

running Kim K3 in LM Studio Bionic.

There we go. So, our inference servers

are US-based uh with zero data retention

enabled by default. And then the final

two updates which I just mentioned, the

first is that you have built-in web

search which is facilitated by LM

Studio. If you do want fully private

local web search, you might need to get

your own solution set up for that. And

it also has MCP. So the MCP access is

awesome because it now gives you the

ability to basically get your AI, you

know, your local AI model out of the

actual Bionic app, out of that

workspace, then to interact and do work

in different tools that you use. So here

we have the LM studio app and this has

got a very familiar interface of tools

that we're already used to using. In the

center we have our text input box and

basically this whole central panel is

going to be our conversation. We have

the plus button that we can introduce

different contexts. So different files

and folders into our conversation. We

have our model selector. So all the

different local models that I've got

running on my device. We have the

transcription. So you'd have to enable

it in the settings and then you can just

transcribe by using this button. And

then obviously we send messages in using

this blue button. Now on the left hand

side we've got this panel which is uh

basically the storage of all of our

projects and historical chats. Down here

we have our settings which we'll get to

in just a second. And on the right hand

side we have the auxiliary tab which is

basically whenever you're working with

files, you're building websites, you're

building apps, any other files and

folders that you're bringing into your

project that are additional to the

conversation that you're having, they'll

be shown over here. So we'll actually

get to see this later on when we build

out some files and some apps. So how

about we get started with a brand new

project. So when you're creating a

project, you have an option to either go

with a default setup, which is actually

work mode. So I'm going to say work

demo. But later when we create our

second project, I want to be enabling

coding. And this is just basically we're

choosing a workspace for us to be able

to code within. And then we're able to

run commands in that folder as well. So

it's just about permissions. For now,

I'm just going to leave it as not allow

coding. So, this is going to be our work

demo. And that work demo project creates

a separate folder on our computer. And

I'm just going to send a quick message

just to test this out and say hi there.

And there we go. We got our response.

And just to check, this is running on my

Mac Studio. The Jules Max Studio is, I

believe, a 64 gig M4 Max. So, yeah, it

takes a couple of seconds initially to

send the first message, but then the

subsequent responses are typically a bit

faster. And this is QAT. I'm pretty sure

it's GGUF format as well. So now I'm

going to give Bionic the very first task

and I'm going to get it to web search

for the latest AI news. Now let's just

go across to our settings. Make sure

we're in a general panel. And let's make

sure we have web search enabled. Once

again, this is the built-in web search

from LM Studio. And in order for us to

use this, we actually have to have, I

think, account set up or billing setup

or we have to have like some credit on

our actual account to be able to use

this. And uh later in this video, I'm

going to show you basically how to get

some credit as well. So we've got web

search enabled. Let's go back to our app

and let's hit send. And here we have our

response or the summary of the AI news

being generated. So if we just scroll

up, we can see if we just click on this

worked called one tool. And over here we

had the request. So basically Bionic

creates an objective which is the task

to be completed. So get the latest and

most important AI news from the past

week. We have some search queries and we

have the response. And I think we went

through 10 different search results. So

basically all that kind of stuff, all

the stuff that Bionic went through was

then summarized in this response over

here. And now that we have this AI

summary, I want to test out the

functionality of working with documents.

So I'm just going to ask Bionic, can you

please turn this into a professional

document that I can then share with my

team? So I I've actually used Bionic

from day one when it was released, and

there's been about three or four updates

already. There have been quite a few

issues, at least for my use case that I

wanted to use this for. For example, you

can see we've got some failed tool calls

over here. There were times where I

couldn't use models from different

devices like via LM link. But over time,

these issues are getting ironed out. I

think it's only been a week since LM

Studio Bionic was released. So, they are

really focusing on getting it improved

and making it better. So, if you are

running into any issues, just be a

little bit patient and I'm sure those

kinks will be resolved. So we can find

the file that was created over here. So

it's in the AI reports folder. And if we

just open this up, that's beautiful. So

AI market intelligence report July 2026.

We've got the full report over here. And

it looks like it was in markdown format.

And then we're just previewing it

nicely. Now this does also work with

PowerPoint, but from my experience when

you're using like PowerPoint generation

inside something like Claude Co-work, it

does come out a lot nicer and a lot

like, you know, it looks really good.

It's got nice coloring. It has nice

spacing, different fonts. When I've

created PowerPoints using Bionic,

depending on which model I use, it

actually comes out pretty like very very

raw, very basic, but it does actually

create a PowerPoint. So, I think with

some additional prompting, you can

actually get there as well. Now, at this

stage, I want to take this file. I want

to share with my team. So, I'm going to

go across to my settings panel, and I

want to introduce an MCP, which will let

me push that document or at least the AI

summary somewhere where my team can see

it. So, I'm going to go across to

connected apps. And over here, you can

see I've got five different MCPs that

are preconfigured. I just need to set

them up. Basically, just give my API

key. But what I want to do is add a

ClickUp MCP, which is a task manager

tool. So, let's just get that started by

typing in ClickUp. I'm going to go to my

browser and I have ClickUp MCP. So, I'm

just going to copy this URL. Let's come

back into Bionic. And this is going to

be a web address. And then I want to get

an access token. So, I want to get an

API token and paste it into here. So,

let's go into my test space. This is my

ClickUp test space. Go into my settings

and let's find ClickUp API. And let me

copy this API token. Come back into

Bionic. Paste it into here. Now, one

thing I noticed before in my testing

with the ClickUp MCP was that I have to

also add a Xworkspace-

ID header. Then, we just need to go back

into our account and copy this workspace

ID. I think because I got a couple of

workspaces attached to this one account,

that's why um the MCP wasn't working for

me before. But yeah, basically that's

what I'm just doing here. So, I'm going

to add this MCP now. And let's wait for

the status to change up here. There we

go. Connected 53 tools are ready. So,

let's just check if we can access that.

Can you access my ClickUp workspace? So,

with Claude Co-work, you typically go to

this plus button and then you can add

different MCPS. You can either like

toggle them on or off depending on a

conversation, but for for the current

version of Bionic, you don't have to do

that right now. So, we have um I can

interact with your ClickUp workspace

using these specialized tools over here.

Okay, let me just get that. Let me just

get actually a space, a specific list,

and I'm just going to drop down here.

I'm going to go to AI updates. Let me

just copy this link. Can you access this

uh list? Okay. create a test task in

that list. Okay, so looks like we're

creating a task. Okay, looks like we've

successfully created that test task.

There we go. Test task from the AI

assistant. Let's see if we can just

share the latest AI news in a brand new

task in that list. This will effectively

close the loop for me. So, we did some

web search. We were able to find the

local AI news. We created a document and

now we want to push that update to our

team. The only thing that I would really

love for Claude, sorry for Bionic, to

have is like a scheduled task where I

can run this kind of process like check

my emails every day at 9:00 a.m. Get the

latest AI news every day at 9:00 a.m. I

would love a way to do a scheduled work

directly from Bionic and then hopefully

in the future they actually release that

as well. So, looks like we actually

created that. Let's go back into our

workspace. Latest AI marketing

intelligence summary July 2026. If I

just expand this. Okay, so instead of

attaching a file, it's just created the

actual update over here. So yeah, that's

awesome. That works. Actually works very

well. So at this stage, let's say that

we're done with the work demo. And now I

want to create a coding project. So

let's go code demo. We're going to allow

coding. Choose a folder for Bionic to

work in. I have code demo from my

original test that I was doing. Let's

just go demo 2 and click create. Use

this folder and create it as well. So

let me just close this side panel. I can

just close this as well. Now, we're

inside the code demo one. And what I'm

going to do is actually just grab this

URL, which is I'm pretty sure this. And

I'm just going to ask Bionic to go back

into that ClickUp task and then

basically recreate whatever information

was there in a simple HTML page so that

I can view it. Okay, awesome. So, while

this is working, I'm just going to go

into the settings and explore a couple

more things here. So, I'm just going to

go through some of these settings and

just show you some things that you

should be aware of. Now, the first thing

is that we're currently on build 9. So,

looks like there's been nine updates to

Bionic already. Definitely check for

updates very regularly. I just updated

mine maybe like an hour ago, but as you

update, you do go through and iron out

some of these issues. So, definitely be

updating your instance as often as

possible. Also, if you have LM Studio,

you must update the LM Studio to the

latest version as well. So, one issue

that I had was when I was first using

Bionic, I had LM Studio like an outdated

version on my uh my Mac Studio and I

wasn't able to use LM link. So, I

couldn't use the inference from my

studio on my local device. So, if you do

run into an issue like that, just have

all your LM apps updated all the time.

So, for web search, we definitely want

to have this enabled. Even though we're

using through LM Studio, no big deal for

now. Exploration agents, I think if you

are using open models like Kimmy K3 and

GLM 5.2, too, you'll have a better

experience. But if you've got a small 16

gig device, maybe if you spawn multiple

agents, you might bust your context

window and then crash your computer

because we don't have a lot of RAM.

Next, we're looking at voice. So, this

is for the transcription. If you enable

it here and you request microphone

access, you'll be able to download that

model onto your device and you'll be

able to use local transcription for

billing and usage. Just make sure that

you have your own account. I think that

you can create a free account and then

you can use my trick that I'll show you

a little bit later to get like five

bucks of credit. It isn't much credit,

but it's better than nothing and it

takes about 30 seconds to get. We have

connected apps that we saw before. So,

we can just plug in uh pre-existing MCPS

or we can very easily add in new custom

MCPS like we did with ClickUp. The LM

link is the most interesting thing. So,

I've got multiple devices plugged into

my LM Studio and I can share the models

across basically all of those devices.

But most importantly, I can use Bionic

on my 16 gig device and I can get the,

you know, the Gemma 426B from my Mac

Studio, from the Jules Mac Studio, which

is awesome as well. We then have the

explore tab. So, this is very similar to

the LM Studio original app, you can just

browse different models. The main

difference here that I love is that

instead of just downloading it onto your

main device that the app is running on,

you can decide which device to download

that model to. So, for example, I've got

Jules's Mac Studio, but I'm currently on

my 16 gig Mac Mini. And I could download

a bigger model onto that bigger device

and then use it without having to

physically be on the Mac Studio. And

then the final thing I want to look at

is the local host URL. So, if you just

start the server, you can copy this URL

and then you can plug it into your

Claude code into your Hermes agent, your

OpenClaw, whatever other tool that you

want to be using your local model

within. And I'd say those are the main

day zero settings to be aware of. So if

you just download LM Studio Bionic

today, just go into those like five tabs

and you know mimic my settings if you

would like. So if we come back into our

coding session, so worked for 1 minute

and 21 seconds. Let's have a look.

Called four different tools. Failed some

of these commands which could be because

I'm using the Gemma 426B.

Maybe if I used the Quen 3.6, which

might have been a smarter thing to do,

I'd have a bit more accuracy with what I

was doing here. But let's have a look.

I've extracted the information, the

latest AI marketing intelligence

summary. I've got the information here

and then I created this HTML file. Okay,

so I'm pretty sure we can also serve

this file locally. If I just open up

this browser, there's no quick button to

actually serve it locally, but we can

get a local host URL over here. So, I

just need to instruct Bionic to do that.

So, we've just started the local web

server and opened the page for you as

well. So, I didn't even have to I don't

even have to copy this URL into here.

So, let me just close this side panel.

And here we have the HTML page built

out, which is awesome. I mean, uh, like

the it looks like it was created by an

AI model, which is fine. We're just

using a Gemma 426B. And I gave very

minimal instructions, but we were able

to complete the task. We're able to MCP

into a different tool, bring in some

context, bring in some information, and

then create an HTML page and run a local

server and then preview it within the LM

Studio Bionic as well. So in my eyes for

you know the first week of this product

being out this is extremely capable

especially with such a small model like

Gemma 426B compared to using Claude Opus

5 or you know whatever other models that

we're using. So yeah very excited for

this app. And now in terms of getting

some free credits for your account what

I did was I I found the LM Studio Bionic

uh it's actually been 14 days that it's

been released not just one week. And I

found this post on hackernews and

basically the founder of LM Studio,

Yagle, mentions that if you just send

your username, so just create a free LM

Studio username, send it to this email

address and then you basically get like

I think still five bucks of credit

loaded up onto your account. So

definitely give this a go and then you

can have web search enabled. You can

have a couple of credits for Kimmy K3 or

you know GLM 5.2 to give it a bit of a

whirl as well. All right guys, thank you

very much for watching this video. If

you've watched this far and you're

looking for additional help, either like

personal coaching or if you're a

business looking to implement AI, I'll

drop a link below to my website so you

can get in touch with me. Thank you very

much and I'll see you in the next