A local-first AI coding workbench
Point to something in your running app. Describe the change. Wirefold codes it in a sandbox. You try it, review the diff, and decide what to Apply.
Desktop developer preview Your models. Your keys.
A brighter corner
Violet / One good idea
$149.00The problem: a failed payment closes this dialog.
WirefoldCheckout.tsx:42 CheckoutThe change you want
When payment fails, keep checkout open and show the error. Don’t clear my details.
Run starts the agent. Typing doesn’t.
The selected element resolves to Checkout.tsx:42. Add the outcome you want, then Run.
Illustrative example. No model calls, payments, or file changes.
Your project. Your preferred model.
The loop, not the leap of faith
No translating a visible problem into a paragraph of file paths. Start in your app. Stay with the change all the way through.
Select an element in your running app. Wirefold resolves its file, line, and symbol through a source map built for your development app—not a model’s best guess.
A live element becomes a grounded code target.Write the outcome in Draft, then press Run. Your request becomes a Run-bound Intent Contract. The agent works in this session’s sandbox, separate from your active checkout.
Typing prepares the request. Only Run starts the work.The canvas switches to the sandbox preview. Click through the result, inspect the complete diff, and run the checks your change needs. You can steer the agent without starting over.
A preview and a diff. Not a “trust me” message.Apply merges the reviewed sandbox changes into your checkout. If the sandbox changes after review, Wirefold requires a fresh review. Not the right result? Discard the sandbox.
Run is permission to work. Apply is permission to keep it.Give the agent room. Not the keys to everything.
A sandbox is a separate working copy of your project. The agent can change it. You can try it. Your active checkout stays out of the experiment.
src/Checkout.tsxYour branch. Your unfinished work.
Exactly where you left it.
src/Checkout.tsx+ setError(result.message);+ keep the dialog open+ preserve entered detailsAgent edits. Live preview. Complete diff.
Try it before it becomes your code.
Add direction while the agent works. Your Draft, prompt, and stream stay with the session when you open Review.
Apply checks that the sandbox still matches the reviewed state. If the content changes, you review it again.
Apply automatically stashes your uncommitted work for the merge. Conflicts are surfaced, not silently overwritten.
On your machine. In your workflow.
Sandboxes, contracts, and session history live on your computer. Your selected model provider receives the prompts, code and captured application data included in the task context. Wirefold does not automatically redact that data or mask screenshots. Development selection markers stay out of production builds.
A straightforward arrangement
A subscription enables new Runs and explicit Resume. Connect your own model key; provider usage is billed separately. Existing work, Review, Apply and discard stay available without a subscription.
Available from source with repository access.
Public installers are not linked yet.
Review the price, billing interval and terms before paying.
Model usage is separate. You pay your provider directly at its API rates, with no Wirefold markup.
Before you Run
A little clarity before you invite an agent into your project.
A local-first desktop app for AI-assisted coding in existing web projects. It brings your running app, an AI agent, a sandbox preview, and a code diff into one workbench. Select a live element, describe the outcome, Run in a sandbox, then Review and Apply.
Not during a Run. Agent changes stay in an isolated, session-scoped Git sandbox. Each patch checks that the file still matches the version the agent read. Your active checkout changes only after your explicit Apply. Apply automatically stashes your uncommitted work and restores it through the merge flow; conflicts need your attention.
No. Your sandboxes, contracts, and session history are local, but the selected model provider receives the prompts and code context needed for the task. Bring an Anthropic, OpenAI, or Google API key; provider usage is billed separately.
The review snapshot binds the complete diff to the exact sandbox state. Apply checks that the contents still match. This protects against applying different code from what you reviewed; it does not certify correctness or passing tests. Try the preview, inspect the diff, and run the checks your change needs.
Vite + React and Next.js web apps in the desktop workbench. Development instrumentation maps supported live elements to their source. Production builds do not include the development selection markers.
Public installers are not linked here yet. With repository access, you can start the desktop app from source using the instructions below. Sign in and confirm a subscription before starting a Run or resuming one. Reviewing and applying existing work does not require a subscription.
For your next “what if…”
Bring your project. Bring an idea.
Give it a sandbox to happen in.
Already have repository access? Open the Wirefold repository on your machine and run these commands from its root.
pnpm install
pnpm devRequires Node.js 20+, pnpm, and Git 2.38+. Add your model API key in Settings, then open your web project.
Open repository Access requiredNo public installer is linked yet. This starts the desktop app from source.