Mulberry runs a local AI model directly on your hardware — no internet required. For heavier tasks, bring your own API key. On-device brain tuning adapts to how you work. No middleman. No data harvesting. No surprises.
// How This Started
// The Problem
I needed a way to view and test web apps I was building — but I couldn't rely on a live connection. The tools that existed all phoned home. Every preview, every render, every dependency call — gone the moment the network dropped. So I built a viewer that bundled everything it needed: React, Babel, Tailwind, all cached locally. A private offline studio that answered only to the device it ran on.
// The Loop
What started as a workaround became a workflow. Building offline forced a discipline: every UI decision had to be right because there was no hot-reload from the cloud, no instant feedback from a hosted dev server. The friction made the product sharper. UI and UX got tested the hard way — in real conditions, on real hardware, with no net to catch a sloppy assumption.
// The Turning Point
When Cursor got acquired, it became clear: the tools you depend on don't belong to you. Any editor, any AI assistant, any IDE living on someone else's infrastructure can be pivoted, monetized, or shut down overnight. That shift wasn't just a business story — it was a reminder that the developer's environment is personal infrastructure. It shouldn't be rented. This project stopped being a personal tool and became something worth shipping.
// What It Became
One builder. No VC. No team. The PWA viewer grew into a full offline IDE — 40+ file types, local AI inference, a neuron editor that shapes how the model thinks, a governance log of every AI decision made. It's on the App Store. The architecture that started as a workaround is now the product. Private by design, not by policy.
// Under the Hood
Every design decision starts with the same question: does the user stay in control?
// Local Model + BYOK
A local model runs inference directly on your device — instant, private, no network. For tasks that need more power, bring your own API key and connect to the cloud model you trust. Requests go device-to-API, no proxy, no middleman. On-device brain tuning learns your patterns and adapts over time.
// Data Sovereignty
Every prompt, every response, every code suggestion — logged locally on your device. Export your complete AI history in one tap. Rate responses to build your own quality record. Delete everything when you want. No retention policies that aren't yours.
// Offline by Default
Mulberry caches CDN scripts locally — React, Babel, Tailwind — so you can develop and preview on airplane mode. Write JSX, see it render. Edit Python, see it execute via Pyodide. Open a 3D model. All without a single network request.
// See It Run
Captured on-device — no mockups, no simulations.
// Honest Comparison
We'd rather you know exactly what this is — and isn't.
| Question | Typical AI IDE | Mulberry |
|---|---|---|
| Where does your code go? | Their servers | Stays on your device |
| Local AI model? | No | Yes, on-device |
| Who holds the API key? | They do (bundled) | You do (BYOK) |
| AI adapts to you? | Generic for everyone | Brain tuning learns your patterns |
| Can you export AI history? | Usually no | Yes, one tap |
| Works offline? | Rarely | By default |
| Account required? | Yes | No |
| Telemetry? | Opt-out (maybe) | None. Zero. Ever. |
| Renders React on iOS? | No | Yes, offline |
| 3D model support? | No | Native (USDZ, GLB, OBJ) |
// Beta Access — $12/month
Sign up, then complete payment securely via Stripe inside the app. License key delivered to your email.
// Beta price locked when we launch. Early supporters keep $12/mo forever.