Every AI builder on the market writes code onto a toolchain you still have to own. Generate the app, then install the dependencies. Wire the backend service. Configure the pipeline. Wait for the build. Debug the deploy.
This is 90 seconds, unedited, one take: sign in, describe a component, watch it get built, edit it with AI, push it to production.
How this exists
I spent roughly fifteen years storing HTML, JavaScript, and jQuery in a database. It worked. Interfaces loaded fast, edits deployed instantly, versioning was free. What it never had was a real back end — no API layer, no server methods reachable from the front end. Everything above the database was fine. Everything below it wasn't there yet.
Two years ago I ran what I called the nucleus test. One question: can I build a front end and a back end that never have to be compiled — ever, at any point, for any reason. Not "compiled fast." Not compiled.
It worked.
Every other decision fell out of that one. One controller, one method — the front end passes the back-end method name it wants called. A dynamic execute pulls the code text from the database and runs it. The rest is consequence. Nothing here is a feature. It's what remained when the toolchain came out.
— Mitchell Maynard, founder
Market Map
Cloud IDEs sit on the right (heavy toolchain). AI code-gen tools sit on the left (lighter toolchain, but still emit code). Frames Studio is the only point where AI emits frames, not files — with no toolchain to maintain.
Feature Matrix
| Replit | Codespaces | Cursor | Frames Studio | |
|---|---|---|---|---|
| Toolchain | Cloud-hosted | Cloud VM | Local + AI | None |
| AI output | Code files | Code files | Code files | Frames (executable intent) |
| Versioning | Git on files | Git on files | Git on files | Automatic on every save. Named checkpoints. |
| Cold-start | Container boot | VM boot | Instant (local) | Instant |
| Multi-tenant | One workspace / user | One VM / user | Local only | Tenant-aware by design |
| Onboarding | Account + setup | GitHub + container | Install + license | Open a browser |
Each row reflects an architectural decision, not a marketing claim. Frames Studio is the only column where the answer is structural — not just faster, but different.
Replit moved the toolchain to the cloud.
We removed it.
Cursor generates code that runs on your toolchain.
We generate frames that need no toolchain at all.
StackBlitz recreated Node in WASM to fit it in your browser.
We deleted the part that needed Node.
What actually happens
Code lives in a database, not files.
HTML, CSS, JavaScript, jQuery on the front. Python on the back.
Nothing compiles. Nothing transpiles.
Saving is deploying.
AI emits frames, not files.
AI Component Builder
Type a description. Pick a model — Claude, ChatGPT, Grok, or Gemini. A complete component comes back: HTML on the front, Python on the back, config alongside. Review it or auto-save it. It’s in your application either way.
Code Manager
HTML on the front, Python on the back — not a generated React blob you can’t modify. Every component is a codename. Every save is an automatic version. Open a frame, edit its code, save it. That’s the loop.
DB Explorer
Browse tables, inspect schemas, run queries, insert rows — from inside the same environment you build in. No SSMS, no Azure Data Studio, no separate Supabase project bolted on. Data, Schema, and Query views. Create tables. Insert rows. It’s the platform’s own database, and it’s open to you.
Deploy Manager
This is where “no build step” stops being a claim. Dev and production, side by side. Every component that changed, listed. Every one that didn’t, left alone. Select what to promote. Click. The changes transfer. Nothing compiles. Nothing waits.
No pipeline. No YAML. No deployment window.
Debug Tools
Event Log streams every FramesEvents emission with timestamps and payloads. State Inspector shows open frames, auth state, and bus health. API Tester calls any Python method directly.
The interface
The runtime happens to render as a desktop UI — file manager, editor, terminal, all in one tab. This is a consequence of the architecture, not the argument for it. Everything you'd do on your laptop, you do here. Nothing new to learn.
What if I want out
Every frame is text. Every schema is JSON. Nothing is stored in a proprietary format — the HTML is HTML, the Python is Python, the schema is SQL.
Most platform lock-in is data-format lock-in. There isn’t one here, because frames are the same shape as what you’d have written anyway. What’s different is what runs them.
A one-command export is on the roadmap and not shipped. If you need it before you’d commit, say so when you request access.
Free during beta. Three projects. No credit card.