Turn Claude into your designer▏
94 wrapps on the board · every one runs on your own AI · all free
All free. Everything in one place.
Every wrapp dials one operator — no wiring, no glue.
Run every wrapp on your Claude — or a local model (Ollama, LM Studio). No per-app fees.
The operator patches the call — never owns your data, never listens in.
It holds your AI on your machine and runs the wrapp store from your menu bar.
Download for Mac →Click the Switchboard icon, open the store, run anything — native wrapps and God need nothing else.
Open a wrapp →Only for web wrapps: add it, then paste the pairing code once.
Add to Chrome →One thing first: your Claude, signed in through Claude Code — that's the AI your wrapps run on.


94 wrapps and counting — browse the whole board → or build your own →.
Requests route locally, through your own model. No copy of you on our servers.
No server holds your credentials — the operator patches the call and steps out.
Every wrapp asks before it touches a tool or file. Revoke it instantly.
Open source (MIT) — audit it yourself. Team mode's end-to-end encrypted; no one reads your board, us included.
A wrapp is just a page. It asks Switchboard for the model, the tools and the context — so the user brings the AI, and you bring the idea.
Your users run it on their own Claude, or a local model. No inference bill, no backend, no scaling. A static host is your whole infrastructure.
Today that means clone, install, get an API key, configure, run — and most people stop at step one. A wrapp is a URL that works on click.
You never see their credentials or their data — the broker gates every call on their machine. Nothing to store, nothing to leak.
One line is the whole onboarding — it reads the contract and wires your app. No install, no signup.
Read thelastprompt.ai/switchboard/llms.txt and wire my app to Switchboard.Switchboard injects window.claude into your page. Replace your model call with one request — that’s the migration.
Read the full spec →Every wrapp on the board is open source — fork the closest one instead of starting cold.
Browse the source →Every wrapp, the operator, the notch. Bring your Claude. Nothing to pay, ever.
Optional cloud inference and zero-knowledge backup. Still your data, still private.
N people, N Claudes, one board — end-to-end encrypted. No login, no server.
@anthropic-ai/claude-code, then run claude to log in.