System26 started as a simple question for us: What actually becomes possible when Apple’s new on‑device AI shows up across iPhone, iPad, Mac, and Vision Pro? Not in a keynote. In a real app. In our hands.
So we built a playground. Not a polished consumer product. Not a stealth startup. Just a fast, opinionated demo that lets us (and our clients) feel what’s now possible with:
- An on‑device large language model
- On‑device image diffusion (text‑to‑image generation)
- Visual intelligence: camera‑powered understanding of faces, bodies, and objects
And we did it the way we build for clients: small team, AI‑accelerated workflow, human judgment at the center.
What System26 actually is
System26 is a native playground app that runs on:
- iOS
- iPadOS
- macOS
- visionOS
From one shared codebase.
Think of it as a “show, don’t tell” lab for Apple’s new on‑device intelligence stack. Apple’s own models now power experiences like Writing Tools for rewriting and summarizing text, image generation via Image Playground, and smarter, context‑aware actions across apps. With the Foundation Models framework, those same on‑device models are now available to developers as building blocks for offline, privacy‑preserving features — with no per‑call inference cost.
System26 pulls those ingredients into one place so we can demo three things live, on your device:
1. On‑device language
In the app, the local LLM can:
- Rewrite and clean up text
- Summarize content
- Help with small, focused tasks
All without sending data to a server.
This matches where on‑device LLMs shine in the real world: short, private, low‑latency tasks like smart replies, summarization, and UI “helpers” that run directly on phones and laptops.
2. On‑device image diffusion
System26 can generate images from text prompts locally. No upload. No round‑trip to a cloud model.
This is the same class of tech behind modern mobile text‑to‑image systems, where optimized diffusion models can create visuals on‑device in under a few seconds.
For a product team, that’s the difference between “tap and wait” and “tap and feel it happen.”
3. Visual intelligence from the camera
The app also uses Apple’s visual intelligence stack to:
- Track faces and body pose in real time
- Recognize objects in the scene
- Power translation‑style experiences via the camera
This mirrors where pose estimation and object recognition are already being used today — from form‑checking fitness apps and physical therapy tools to sports training and real‑time motion analysis.
System26 isn’t meant to be “the product.” It’s the conversation starter. A way for us to sit with you, open the app, and say: “Okay. Now imagine this inside your business.”
Why on‑device AI matters for your product
You don’t need a playground app. You need outcomes: speed, privacy, cost control, better customer experiences.
Here’s how the same building blocks we used in System26 translate into real‑world wins.
On‑device LLMs: instant intelligence where connectivity is fragile
On‑device LLMs run directly on hardware — phone, tablet, laptop — instead of in the cloud. That means:
- Speed: responses in milliseconds, not seconds
- Privacy: sensitive text never leaves the device
- Resilience: your app keeps working in planes, warehouses, or store backrooms
In practice, we’re seeing (and shipping) patterns like:
- Field tools that can’t depend on Wi‑Fi
- Inspectors, technicians, sales reps: summarize notes, generate visit reports, or translate key phrases offline while standing in front of the equipment, not back at a desk.
- Inspectors, technicians, sales reps: summarize notes, generate visit reports, or translate key phrases offline while standing in front of the equipment, not back at a desk.
- Customer‑facing flows that feel personal, not creepy
- Local models suggest copy, rewrite messages, or summarize long threads without streaming customer data to a third party — especially important in regulated industries.
- Local models suggest copy, rewrite messages, or summarize long threads without streaming customer data to a third party — especially important in regulated industries.
- “Quiet” intelligence around your existing UI
- Think Apple’s Writing Tools, which can summarize or rewrite text inline, or notification summaries that cut through the noise.
- Your app can do something similar: turn complex status pages into one clear sentence, or auto‑draft human‑sounding responses for your team to approve.
On‑device image diffusion: visual creativity without the wait
Text‑to‑image models are no longer just cloud services. Research like MobileDiffusion and production apps from companies like Adobe show how fast, on‑device image generation is becoming part of everyday creative tools.
Patterns we’re excited about with clients:
- Retail & e‑commerce
- Instantly generate on‑brand banner concepts, promo variations, or simple lifestyle mockups directly in a merchant or marketing app — without exporting assets to another tool.
- In‑store experiences
- Help associates create quick signage, recommendations, or inspiration boards with a prompt, right from a tablet on the floor.
- User creativity without heavy infrastructure
- Let your customers create avatars, stickers, or lightweight AR elements — all rendered locally — instead of relying on a GPU‑heavy backend.
Visual intelligence: your camera as a real‑time sensor
Computer vision is already transforming industries like fitness, healthcare, and retail by turning raw video into actionable insight.
With on‑device models, that intelligence can move directly into your app:
- Fitness & coaching
- Pose estimation to give instant feedback on form, reps, or alignment — without ever uploading video of someone’s body to the cloud.
- Retail & inventory
- Shelf‑scanning tools that run on a phone or tablet camera to detect stock‑outs or misplaced items, instead of relying solely on fixed ceiling cameras and heavy infrastructure.
- Try‑before‑you‑buy experiences
- Virtual try‑on for fashion, cosmetics, or accessories using pose estimation and face tracking — a pattern already driving conversion lifts in fashion and beauty.
System26 doesn’t do all of these things out of the box. But it lets you feel the underlying capabilities: real‑time tracking, instant responses, offline behavior. From there, it becomes much easier to imagine the version that belongs inside your product.
Built in ~16 hours: what AI‑accelerated development actually looks like
We talk a lot about shipping faster with AI. System26 is one of the cleanest examples.
Roughly speaking:
- A few focused hours to define behavior, flows, and visual direction
- One AI‑assisted coding sprint to generate the first working prototype
- The initial version — with the on‑device LLM feature — came together in about 20 minutes once the specs were in place
- Another set of passes to layer in image diffusion and visual intelligence
- A mid‑way refactor to lean fully into Apple’s liquid‑glass UI patterns
- A couple of hours to localize the app into 10 languages using AI‑assisted translation
All told, around 16 hours of build time plus localization — spread over what, in another era, would have easily been a multi‑week effort.
The key is not “AI built this for us.”
The key is:
- We were precise about what we wanted
- We used AI as a fast, tireless developer assistant
- We kept human judgment on architecture, UX, and quality
AI accelerated the work. It didn’t replace it. That’s exactly how we build for clients: tools in service of expertise, not the other way around.
What this unlocks for you
System26 exists for one reason: to make the conversation about on‑device AI concrete.
When we sit down with your team, we don’t have to hand‑wave about “Apple Intelligence” or “edge models.” We can:
- Open the app on an iPhone or Vision Pro
- Show you offline language, image, and vision features working in real time
- Map those same capabilities directly onto your customer journeys and internal workflows
From there, we can quickly answer questions like:
- Where does on‑device AI really move the needle for you — speed, privacy, cost, or experience?
- Which use cases should stay in the cloud, and which could run locally for near‑zero marginal cost?
- How do we design something useful, humane, and trustworthy — not just “AI‑powered” on a slide?
And if you’re the type who likes to see under the hood, we’re releasing System26 as open source on our GitHub. Clone it, explore it, star it, borrow pieces of it.
Ready to see what this looks like in your world?
If you’re curious about:
- Bringing Apple’s on‑device intelligence into your product
- Using AI to shorten build cycles from months to days
- Designing experiences that feel fast, private, and human‑first
We’d love to show you System26 live, then sketch the version that belongs inside your business.
Let’s build something that actually moves the needle — and feels good to use.