Architecture
How LayerT is built
Written for the engineers and security reviewers who have to decide whether LayerT is sound. It says what the product does, how, and where the limits are.

Describes LayerT as built on 18 September 2026. Anything not built is labelled.
LayerT has two parts. A browser extension runs inside the Chrome profile your company already manages. A console and API is where admins, approvers and compliance people decide things. They talk only over HTTPS. There's no proxy in your network path, no agent on the operating system and no new browser.
The five rules we build by
LayerT's engineering follows a written constitution. Every spec and every change is checked against it.
- Fail closed. If LayerT can't prove its policy is current and genuine, it blocks the actions that policy covers. It never quietly lets them through.
- Separate the people who configure from the people who grant exceptions. IT writes rules. Compliance grants lasting exceptions. Nobody approves their own request.
- Sign what the browser trusts; keep the audit trail readable. Policy is signed with Ed25519 and checked in the browser. The audit log is kept in plain words, so an auditor can read it.
- Use the real protocols. SCIM 2.0 for your directory, OpenID Connect for sign-in, RFC 6238 for authenticator codes, Web Crypto for signatures. Tests replace the data sources, never the protocols.
- A feature is done when a real browser proves it. Each flow has scenarios that drive real Chromium with the real extension loaded.
What runs where
| Part | Runs in | What it does |
|---|---|---|
| Service worker | The extension, in Chrome | Enrolment, device credentials, policy sync and signature checks, audit batching, shared-account grants |
| Content scripts | Each page, in Chrome’s isolated world | Watch only the fields your rules name; fill shared-account sign-ins; draw LayerT’s prompts |
| In-page prompts | Closed shadow roots on the page | The block card, the account menu, notices and the session chip. The page can’t read or click them |
| Console and API | LayerT cloud | Rules, shared accounts, Inbox, Sessions, the audit log, SCIM and sign-in |
| Postgres | LayerT cloud | Policy versions, audit events, directory data, shared-account secrets as ciphertext |
| Object storage | LayerT cloud | Session recordings, encrypted on the device before upload |
| Scheduled jobs | LayerT cloud | Expire requests and sessions, apply retention, re-sign policy before it expires |
The details are in Components and data flows.
What LayerT’s servers can see
A security review should know this up front: LayerT is not zero-knowledge. To sign someone in to a shared account, LayerT’s servers decrypt that account’s password for an approved, time-boxed session and send it to the person’s enrolled browser. To play back a recording, the servers decrypt it for someone allowed to watch. Both actions are checked against roles, bound to a person and written to the audit log.
How far along it is
LayerT is pre-launch. Everything on these pages is built and tested in development unless it carries a label:
- Off by default built, but a company has to switch it on (recording, watermark, auto-submit, two-author review)
- Early access built, but not yet approved for production (gateway routing)
- Coming soon in progress (Edge, Firefox and Opera; cloud key management; force-install from a signed package)
- Proposed designed, not built
There are no customers or production deployments yet, and LayerT holds no certifications.