A large language model with tools — the same class of AI as a coding agent
Stunning is powered by a frontier large language model that has been given a software engineer's toolkit. It reasons about your request, then acts on the project through those tools.
Reasoning before writing
The model thinks through the structure of the app first — what data it needs, which screens, how they connect — and only then starts producing code.
Tool use is the key difference
A chatbot can describe an app. An agent can build one, because it can reach out and change the filesystem, install dependencies and trigger a deployment.
Guided by a builder's playbook
The model operates under a detailed set of engineering and design instructions — how to structure data, how to handle errors, how to make an interface feel premium — so quality is consistent, not luck.
What the model can actually do
These are the concrete abilities that turn a language model into an app builder.
Writes complete files
Not fragments or snippets. The model emits whole, valid source files with imports in the right place, so the project always compiles.
Uses tools, not just words
It can create files, edit specific lines, read existing code, install packages, run commands and deploy. Each tool result becomes new context for the next step.
Plans multi-step work
A request like 'add a booking flow' becomes a sequence: data shape, then the write path, then the screens, then the polish — executed in order.
Holds long context
It keeps the whole project in view — the data shape, the existing screens, your saved preferences — so new work fits what is already there.
Self-corrects
When a check fails, the model reads the error, reasons about the cause, and fixes the root of it rather than hiding the symptom.
Designs as it codes
It is instructed to build premium interfaces — real hierarchy, depth, motion and a coherent palette — not plain forms on a white page.
Is it the same as Claude Code?
The idea is the same family: an AI coding agent that reads a project, edits real files and runs commands. The difference is the environment around it. Stunning pairs that agent with a managed reactive database, an automatic deploy step, a live preview and a visual editor — so the result is a running, hosted application rather than code sitting in a terminal.
In short: a coding agent is the engine. Stunning is the engine plus the chassis, the dashboard and the road.
See the full stack