What's next
type a task — your Claude routes it to the right wrapp and saves it to your one listReady to review
Projects
Quick actions
one click, straight into the workRecommended
from what your library holds — names only, never the dataYour library
what your apps can borrow — names only, lending is per appAccount
tokens and Pro are conveniences — never wallsOpen source · runs on the Claude you already pay for
One connection. Not one per app.
Every AI app asks you to start over — a new account, a new subscription, a new copy of everything you've already explained. Switchboard sits in between: you connect your own Claude once, and all 27 apps here run on that one connection. Nothing you type reaches a server of ours.
Free · runs on the Claude you already pay for · no account. brandbrain is the one to start with.
New Additions
All Tools
every tile is the live app — open one and it runs on your own Claude, freeBrand & content
brand systems, the knowledge bank, copy reviewbrandbrain
Consumer brands, built on receipts — white space, every decision locked, launch.
Bank
Notes, tasks and [[links]] in plain .md you own — a second brain you can actually ask.
Redline
Review a landing page like a document — drop comments, your Claude proposes copy & edits the real file.
Marquee
A cinematic scrolling landing page from one line — generated, hero-painted, edited in place.
Validate an idea
ideabrain, opened to your kind of ideaideabrain
Not a company yet? Research, a playbook for your kind of idea, a deck, and who to reach out to.
Marketplace Validator
Two-sided idea? Work the supply, demand, cold-start, take-rate and path to liquidity.
Consumer App Planner
The core loop, retention hook, distribution and how it makes money.
SaaS Thesis
ICP, wedge, sales motion, pricing and the integration moat — pressure-test a B2B SaaS.
Retail Concept
Location, unit economics, ops and footfall — before you sign a lease.
Hardware Reality Check
BOM, manufacturing, margin, channel and the capital it takes — the honest maths.
Feature Case
Fit, adoption, cannibalization and the metric that proves it.
Ads & growth
analyse, generate, and post the growth workAdPulse
Meta ads post-mortem — paste your export or pull it live; your Claude finds the wasted spend.
AdForge
URL in, Meta ads out — three concepts, one recommended, rendered on your Higgsfield.
Adwall
The volume play — six ad directions at once from your lent brand, render any tile on your Higgsfield.
A-Plus
Amazon A+ content in one pass — hero, feature grid, comparison chart, brand story.
Batch
Your YC application, answer by answer — all 8 questions drafted on your own Claude.
Creative
personas, images, reels — on your own modelsIdentity
Compose an AI creator persona facet by facet, then publish it as a context Cast picks up.
Prism
Prompt → image on your Higgsfield connector, on-brand when you lend it a brand context.
Reel
One line to a launch reel — Claude scripts scenes, Higgsfield paints them, renders in your browser.
Cast
An AI persona studio — creators that stay on-model across faces, scenes, and reels.
Commerce
inventory and product photographyStudio
Product shots without the studio — one photo in, finished frames out, on your Higgsfield.
Shelf
Inventory triage — reorder-now, watch, dead weight, and the cash locked on your shelf.
Play & make
record, build, and get on a callTake
Script it, then record it — Claude writes the beats, you capture in the page. Never leaves your browser.
Cartridge
Form → playable game. Describe it, your Claude builds it, you keep the .html artifact.
Huddle
Get on a working call with your own Claude — camera on, warm thread, grounded in a project folder.
After hours
the fun onesNATAL
Your chart, read bluntly — birth data in, sun/moon/rising and an unsentimental daily brief.
Arcana
Three cards, no mercy — draw a spread and your Claude reads the exact cards you pulled.
Three things you set up once, instead of once per app.
what every other AI app makes you repeat — and what happens here insteadOne account
Everywhere else: every app is another signup, another password, another company holding a login.
Here: no account with us, and none with any app. You bring the Claude account you already have.
One subscription
Everywhere else: their model, their key, their markup — a second bill stacked on the AI you already pay for.
Here: your Claude is already paid for and sitting idle. Apps spend that quota, and we never touch the transaction.
One copy of your context
Everywhere else: each app learns your business from scratch, so you explain it again for the third time.
Here: your context lives in a folder on your machine. You lend it to an app for one job, and it hands it back.
Four steps, and it's yours.
the page keeps track of which one you're onPoint at it and it's yours
One link or one folder. Your Claude reads it, you confirm what it found, and every app below can borrow it.This page was connected before it could read for you. Re-approve once to let your Claude read what you point at.
Five things this can't do to you.
written the way we can defend them, not the way that sounds bestNo key in the app
An app never sees a credential, and neither do we. An app is static files; the credential and the connection live in the daemon on your machine.No account
None with us, and none with any app. You install once and bring the Claude account you already have.No second bill
No API key, no markup, no charge from us for inference. Apps spend the plan quota you already pay for. The daemon does meter each app's usage locally so you can set a budget — that meter reports to you and to nobody else.No one in the middle
Nothing is shared with an app's developer, with us, or with anyone in between. Your prompts go to your own model provider under your own account, or to a local model and then nowhere at all.No silent hands
Nothing writes, sends, or spends without a click that no page and no model output can forge.Every one of those is a constraint you can read in the source, not a promise you have to take from us. One more thing we'd rather say than have you find: where an app's work is stored is yours to choose. Bind an app to a folder you own, or leave it unbound and it gets a folder inside the daemon's own state directory — still your disk, still your machine, never our server.
N people. N Claudes. One folder.
Each teammate keeps their own AI. Only files move.Multiplayer everywhere else means everyone shares one model on someone else's machine — and that machine reads every document on it. Team Mode is the inverse. Each member keeps their own Claude, their own approvals, their own machine. What syncs is files: the same folder, in the same plain .md your vault already speaks. One member hosts, the others join by pasting one invite code. Every frame is sealed before it leaves the machine, and the relay that carries it between networks never holds the key, opens nothing, and stores nothing. It's MIT like everything else — a hosted one is running today and prefilled when you host, or you can self-host it.
You will not watch a teammate's Claude type. There's no shared stream, no per-member attribution inside apps, and no roles yet. Concurrent edits merge file by file, last write wins — right for notes and tasks, wrong for two people typing in one paragraph. Your teammate's Claude never becomes yours; inference never changes hands. And a teammate can never answer a consent prompt on your machine.
Using your own AI is free. That's the design, not a trial.
everything here runs on the Claude you already pay for — or a model on your own machine — at no charge from usAll 27 apps, the whole loop, end to end. No watermark, no artificial delay, no export of your own work held back, and no consent or safety feature ever behind a plan. Your context and your files are yours on every tier, because they never left your machine to begin with.
Your own Claude or a local model — the loop is identical either way. Open source, MIT: read the broker or run your own copy. The only meter is the local one, so you can set a budget per app; it reports to you and to nobody else.
If you have no Claude and no local model yet, a token pack covers the work until you bring your own. Buy once; packs never expire. The moment you connect your own Claude they stop being spent.
Said plainly: checkout here is a labelled simulation today — no card, no charge, no payment rails, no hosted pool behind it. Never required; connecting your own Claude is always the shorter path.
Team Mode is free and MIT, and you can self-host the relay right now. A hosted relay carries a team across networks without anyone running infrastructure; one is running today and prefilled when you host.
Free while we harden it. If that ever changes it would be the second and last thing we charge for — and it still would never hold your key, your files, or the ability to read a frame it carries.
Neither of those is a wall. Nothing here makes bringing your own AI the second-class path — that's not a promise invented for a landing page, it's written down with the rest of the rules in TOKENS.md.
Your files, given a face.
the direction we're working toward — shown here as a plan, not a featureToday an app can borrow a folder and read it, once you approve it. What we want next is for the daemon to render that folder — a live view of the real files, never a copy. You'd look at your files the way you look at a page: add something, remove something, and you've edited the real file, though the app never touched your disk.
The honest hard parts: one source of truth means watching the disk and writing back through to it; every format needs a viewer; write-back has to be faithful or it's worse than nothing; and the trust boundary gets wider, not narrower — the part we're slowest about on purpose. None of this is built. It's the direction, and we'd rather show you the plan than pretend it's already wired.
The app is replaceable. Your Claude, your files and your machine are not.
that's the whole idea — read the manifesto