No Electron. No CLI. No Python. No vision model. Just native Swift reading the accessibility tree and driving your real apps. Free. MIT licensed.

No screenshots, no vision loop. It read the accessibility tree in Slack, found the message field by name among 150 elements, typed, and hit enter. Native Swift, end to end.

Free forever. MIT licensed. Native macOS, no browser engine.

I tried Osaurus, but I was completely overwhelmed by the huge number of settings. Honestly, your settings are terrifying. There are so many controls, and the wording is super technical. It’s the total opposite of the chat interface and overall branding, which looks friendly and easy to use.

We added a configuration agent which you can talk to. Whatever configuration you need, just ask for it

Here something u can add that can change this Use agent browser by vercel most apps are electron based it like you control the browser, i have tested it to control most apps including slack plus they have CDP you use your current session either chrome or arc instead of private one

That's a great idea actually

yeah Slack is Electron, that's the joke. 🥸

nice

Thank you! Much appreciated

Does the latest version have this feature?

Yes it does! Settings > Computer Use

No Electron is the way. Really hope we will see a Windows version of this soon!

We're working on it. Fully native Windows

Tim to try

Let me know what you think!

how its doing that without focusing on screen ?

You can read about it here:

wow

we didn't optimize Electron. we just didn't use it

my laptop battery is happy there is no electron in here

We hear you!

cool

Osaurus is my favorite Mac client.

Fancy.

Seeing 'No Electron' is pure music to my ears. Definitely putting this on my radar.

Looking forward to hearing your feedback next!

I spent more time with it and it quickly became my favorite app to run LLMs locally. The extra work needed to go full native definitely paid off judging by how everything runs so smooth. I wish I had a beefier Mac to run bigger models, but I'm definitely planning to upgrade soon!

accessibility tree is great until an app draws custom controls they just don't exist in the tree. half the apps i tested this approach on were basically invisible to it.

It's a huge challenge. The biggest challenge is working with electron apps. We wrote our own evals suite to test across different apps, electrons included. It also includes it's own internal telemetry (not sent to us) to collect data for future training use.

native swift driving the a11y tree is exactly the right call. reliability angle is what kills most of these projects tho, how's error recovery when the target app updates and shifts nodes

It honestly depends on the model. We added our own Eval suites to test all cases, we're seeing positive recovery stats from even smaller local models

consistency my issue with Browser/Computer Use is SPEED. is this higher optimization upside? what's the point other than it sounds cool?

just tested it, very very good im impressed

Thank you!

any chance you can look into image support for custom providers? i am running several models that support images through custom providers and none of them work.

Yup! Something we are working on now

also autonomous execution without sandbox pls i didn't find any option to allow all permissions

Does it always come with its own GUI or can I use it with Open WebUI?

It does come with its own GUI but you can also use any other UI. It exposes the local server that can connect to other harnesses as well.

Cool! I will try it out.

Let’s goooooo

Absolute mad lad!

More native apps 🙏🏼🙏🏼

@OxTrenchor launch $Osaurus, delegate fees to @OsaurusAI

Linux support when ?

Is there specific distro you like?

@eren7262 windows support pls?

@eren7262 It's a work in progress!

Reading the accessibility tree instead of relying on a vision model is a clever way to skip a ton of overhead.

I am trying this right away. Config screen feels like OpenClaw, ngl.

You can use our configuration agent to configure it

Yeah, that's what I did. It tried and it now successfully connected to my Music app. This is awesome!

this is amazing.

Thank you! Let us know if you see any improvements to be made

not available for windows sadly 😓

We're working on the windows version!

that's cool, will wait.

What ?

What is good

Would love to power it on windows and Linux as well. Built on rust. Extremely performant

Sounds impressive! Simplifying tech while keeping it powerful is no small feat. Excited to see what you create!

When will it be available for Windows?

Why no CLI? Can I use it in Claude Code or Pi?

You can read about it here!

I tried @OsaurusAI app but couldn’t find if it exposes a local endpoint for other harness to connect to..

It does automatically.

Thanks for correcting

很好用的绿色龟头