From one creator's page to a whole team's AI output, governed.
Bailey started as the fastest way for one person to put an AI-made site online. This week it grew a spine for teams: organizations, a compliance control tower, and a data plane hardened to enterprise standards, without ever asking a non-technical creator to become an admin. New here? Read this as a guided tour: each stop shows what changed and how you'd actually use it.
The anti–Shadow-AI play, at team scale.
When everyone on a team can spin up a site with AI, most of it becomes shadow IT: data collected nobody can see, pages nobody can turn off. Bailey's answer is the opposite: everything a team ships is visible, accounted for, and one click from off. This week we built the machinery that makes that promise true for an organization, not just a person.
Your team's workspace, with real boundaries.
Bailey began with a simple assumption: one person, one page. The moment a second person shows up (a co-founder, a teammate, an intern), that assumption cracks. Who owns the page? Who's allowed to change it? This week we answered that, and the answer is one decision everything else hangs off: a project belongs to the organization, not to whoever happened to create it. From there the team model clicks into place: who can see a project, who can edit it, and how a page moves from one person's hands to another's.
Orgs, activated
Create an organization, invite members, claim a domain, switch the active org from a global topbar selector: the whole lifecycle, end to end.
Org creation · members · invites · domainVisibility & rights
Projects can be visible or hidden, with view/edit grants by role and by person, enforced on every project endpoint, not just in the UI.
Grants · role + personFork to your own
“Fork this page” turns any opt-in shared project into an editable copy in your personal space: a starting point without a scary blank page.
Fork → personal · read-only source- ✓ Internal pages (audience: org): pages served only to org members via an SSO viewer
- ✓ Audience selector: public / org / password, right on the site's access settings
- ✓ Org logo uploadable on the workspace rail badge (private asset)
- ✓ Tenant-scoped projects & folders with an org-creation cap
A teammate built a little “office lunch” poll for the 3rd floor, and you want your own for the 5th. No hunting for files, no pinging them for a copy: you open their page and hit Fork this page.
Bailey hands you an editable duplicate in your own space, with a clean, empty data store. Change what you like, publish it at your own address. The original stays untouched, and nothing leaks between the two.
The view a CISO actually asks for.
A fleet of AI-made apps is only safe if someone can see it whole. The new control tower maps every app in an org by risk × exposure, flags where personal data is really flowing, and puts bounded kill-controls one click away.
Risk × exposure map
Every app placed on a two-axis map so the risky, publicly-exposed ones surface first: a table-first UI built for scanning, not clicking around.
Fleet map · table-firstPII tripwire
A content-level tripwire watches what pages actually store and collect: a caller-independent backstop that catches personal data even when nobody declared it.
Persisted shadow · collect + KVBounded controls
Governance actions (including turning collection off) are exposed as bounded, auditable controls from the tower itself.
Control actions · auditedIf it collects, it stays in the clear.
The moat got deeper this week, but here's what matters to you: it got deeper without adding a single step you have to think about. An email box, a booking form, a newsletter sign-up: that's someone's personal data the instant a stranger types into it. Bailey now treats it that way from the very first visitor, and it quietly makes the safe path the easy one.
You add an email field to your page and hit publish. Bailey pauses and asks, in plain words: what's this for, and how long do you keep it?
You type “to tell people when we launch, keep it 12 months.” Done: you're in the clear, and you never opened a legal PDF. Skip the question and the page simply won't collect. That's the “a reason, or it doesn't go live” rule, working for you.
Under the hood, the guarantees got sharper:
- ✓ Self-describing manifest: per-field PII classification; a page returns its own live data contract
- ✓ Consent notice enforced at runtime: declared, then actually required before collection
- ✓ Hardened reconciliation: anti-downgrade + schema stability on every publish
- ✓ Right to erasure covers KV too: one person, gone everywhere
- ✓ Per-tier write quotas: a free site ≠ a paid one, enforced in the store
- ✓ Data-plane durability: fsync + anti-divergence between memory and disk
- Collects
- waitlist · emails
- Classified as
- personal · derived
- Kept
- auto-deleted at term
- Readable by
- You. Nobody else.
Counted, never tracked.
A whole new brick landed: cookieless analytics. Traffic, top pages, referrers and UTMs: measured without a single cookie, so there's no banner to bolt on and nobody gets followed around. Measurement that fits Bailey's promise instead of contradicting it.
Governed primitive
A first-class analytics bucket type in the golden chain, gated by plan on both the serve-time and control sides, behind a rollout flag.
Premium entitlement · v0 cookielessAnalytics tab
Traffic on a continuous timeline, top pages, referrers, UTMs, plus CTA-click and custom event tagging, right in the dashboard.
Dense series · event taggingReadable by your AI
A governed read_analytics tool means you can just ask your assistant “how did the page do this week?” and get real numbers back.
Your numbers live wherever you already are, including inside your AI. Just ask:
No dashboard to open, no cookie banner to bolt on. Your visitors were counted, never tracked.
Govern from the UI. Or just ask.
Governance stopped requiring a terminal. The dashboard was redesigned flat/SaaS around a rail-and-main shell, and the agent got smarter: a suggestions engine that spots the gap between what a page intends and what it actually declares, and nudges you to close it, in plain words.
- ✓ Data governance from the UI: no terminal; view & edit auto-email replies in the Data tab
- ✓ Recent activity feed + a preview mini-funnel on the home overview
- ✓ Kill switch: “stop collecting” per site, reversible, from the UI
- ✓ Suggestions engine: intent−reality delta, with non-tech tutorials and a “why” on each nudge
- ✓ Brand cascade: default → org → project, with a per-project brand editor
- ✓ Site preview in chat: the agent shows a rendered thumbnail inline after publishing
- ✓ TypeScript strict + typed API boundary, Error Boundary per view, WCAG zoom, mobile pass
- ✓ Lean, self-healing agent skill: connect once, it bootstraps its own playbook
Say your new contact form collects emails, but you never told Bailey why. Instead of a cryptic red error, a friendly nudge appears in your control room: “This page gathers emails. Want to add a reason so it stays in the clear? Here's how.”
One click opens a two-line, plain-English tutorial. That's the suggestions engine: it watches the gap between what your page intends and what it declares, and helps you close it before it becomes a problem: no jargon, no terminal.
The unglamorous work that lets us sleep.
Enterprise trust is earned in the boring layers. This week: real backups, a security-gated CI, and hardening on the identity path: the parts nobody notices until they're missing.
- ✓ Data-plane backups: encrypted, off-site, with a tested restore runbook
- ✓ CI test gate on release + security scans; multi-arch build fixed (no more emulation)
- ✓ Owner-identity hardening: identity validated before the authorization filter runs
- ✓ Toolchain refresh: reachable upstream fixes picked up
- ✓ Dogfood: the docs site now runs on Bailey's own SDK for its page-view analytics
Where the line points.
Review before publish & finer roles
The org-ownership foundation opens the door to review flows and role-based access across a team.
Cookie consent → richer analytics
Consent unlocks bounce, unique visitors and flow: the next brick on the analytics foundation.
A distributable CLI
Remote transport, one-click login and deploy tokens: Bailey from your own terminal (in flight).
Collaborative primitives
The read & coordination half of the data plane: shared boards and beyond.
Publish what your team makes, governed by default.
Connect your AI once, then just ask. Or drop your files by hand. Either way, everything you ship stays visible, accounted for, and one click from off.