Cloud VMs keep getting hacked, so we built the first on-device iPhone agent Introducing Result > Cards & passwords stay on-device > Run openweight models on your compute or cloud > Native Apple Pay & Passwords > Works in the background even when locked

this looks super promising. keeping credentials on-device makes a ton of sense.

cheers!

I mean this is another cloud one.

haha yeah its hard to believe but that's running fully on your iPhone. We've built an entirely new browser agent harness in Swift that runs locally on your iPhone to get tasks done.

Yes. It’s hard. Because

looks like it wasn’t clear: Execution happens on your iPhone - your passwords, cards, etc stays on device. For running a model you can: a. Access the latest models in our cloud b. Or you can pair with a mac to: - Connect your private inference (DGX, Ollama, etc.) - Download a model weights there & tunnel from there

Yes. Now it’s clear. You can’t use the model on iPhone. You can pair with your Mac or other machine. Means “cloud”. Means you need an external machine to run inference. You support slow engines so no proper spec dec etc for Mac. Thanks

Based on our internal tests, models small enough to run on an iPhone were too slow and not capable enough. On Mac, the models we support (GLM-4.7, Qwen 3.6, etc.) perform surprisingly well on most personal agent tasks, reaching 100+ tokens/sec. For cloud inference, we offer zero data retention, and we support frontier open-weight models at speeds of some up to 1,000 tokens/sec.

Does the browser instance and computer use run locally too?

yes! it runs on device and in background (+ when its locked too)

Very good

needs to be oss to even consider or integrate with mcp and chatgpt to recommend to others

Personal on device ai that can utilise Apple Pay nd device vault. Is I think the future on agentic devices

Let's go!

So it’s still a sandbox but not in a VM? The risk is always the OpenClaw one where your agent thinks it’s helping and instead deletes your codebase or apps. I assume this method still has protections from this?

Its a safari based browser instance and doesn't have access to file system or anything of destructive sort. It can browser websites, interact with it, and get work done!

congrats on the launch! super sick

thanks!!

I would love to use it with my @near_ai subscription for guaranteed confidential (TEE) inference. Is there a chance to opt-in for that?

where do you run the cloud openweight models?

we use inference providers with a 0 data retention policy so you have access to all the frontier openweight models with 0 data retention you can also run the models on device on your own compute

This is so cool!

thank you!

@benjitaylor could @bot computer work directly on my chat like this, instead of its own tab hidden in bot profile? 👀

banger

do you have to anchor a macbook to it?

is this a pivot or am I hallucinating

This is amazing!!!!

how does this take control of your device? I thought iOS was way too sandboxed and locked down for something like this

Bringing agent execution onto the iPhone gives Result a clear focus. I think your next demos could reach more people interested in on-device AI. Want to explore a boost?

Great idea and execution!

Thank you!

Need android version

How do you make money?

👀👀👀

6

@grok 这个有安卓的计划吗

Why the age restriction on app store is 16?

Yo! This looks sick

Thanks!

Congrats on launching Result! Keeping browser-agent data on-device is an interesting approach. I own available for sale if it fits a future product. DM me.

Can I hook up my local linux AI server to it?

Won’t allow me to proceed with onboarding without accepting terms of use that state you send data to cloud AI providers. How is this local inference? Is local on my Mac only if paired? Either way, blocking signups over this doesn’t quite jive with your on-device framing

Running browser control on the phone is a meaningful difference from a cloud VM. When the model runs on a paired Mac or in the cloud, which context leaves the iPhone—and do browser sessions, credentials and memory stay local?