No key escrow
We never hold, derive, or read your encryption keys. The relay routes opaque bundles it cannot open. Compromising us does not compromise your data — and that isn’t a policy promise, it’s an architectural one enforced in our test suite.
Capability Assertion Protocol
Every system you run hands out bearer tokens — strings that mean “trust whoever holds this.” They can’t be un-issued, they can’t say what they’re for, and they were designed for a browser with a human in it. Richdale Sigilex replaces them with short-lived capability certificates: bound to a key, scoped to an action, checked on every single call, and revocable the moment you change your mind.
No redirects. No key escrow. Every decision independently verifiable — including by people who don’t trust us.
The problem
Access control was built around a human at a browser, clicking through a redirect once an hour. That assumption is now false in four expensive ways.
How it works
A capability certificate names the key that holds it, the action it permits, and when it dies. It is checked on every call — not once at login.
A certificate is minted against your tenant key, scoped to specific actions and a short lifetime. It names the holder’s public key, so a copied certificate is useless without the private key that never left the device.
The gateway checks the signature and the live revocation state on every call, then fails closed. There is no session to hijack and no cached “already logged in” state to ride.
Each decision is appended to a signed, tamper-evident log. Anyone can pull a single entry and verify it against the signed root in their own browser — no access to our systems required.
What makes it different
We never hold, derive, or read your encryption keys. The relay routes opaque bundles it cannot open. Compromising us does not compromise your data — and that isn’t a policy promise, it’s an architectural one enforced in our test suite.
Kill a certificate and the entire delegation tree beneath it goes with it. Because enforcement happens per call rather than per session, the next request is already denied — you’re not waiting out a token’s lifetime.
The log is an append-only Merkle structure with a signed root. Pull any entry and verify the proof client-side. An auditor doesn’t have to believe our dashboard — they can do the maths themselves.
Hand an AI agent a narrow slice of your authority — specific tools, a short window, nothing more. Every call it makes is verified against that slice and recorded, so “what did the agent actually do” has an answer.
The part nobody else offers
Most security products ask you to believe their dashboard. Sigilex publishes a signed root of its audit log, and ships the verifier to your browser. If our records ever disagreed with the cryptography, the cryptography would say so — in your hands, not ours.
Products
Sigilex is the platform. You don’t have to adopt all of it at once — most teams start with whichever edge hurts most.
The core: capability issuance, delegation, per-call verification and the provable audit trail — with an admin console for the humans who have to run it and answer for it.
Put your MCP tool servers behind CAP. Agents get scoped, expiring authority instead of a shared API key, every tool call is verified independently, and the whole session is reconstructible afterwards.
A standards-compliant OpenID Connect provider, so the applications you already run can authenticate against CAP without knowing anything about capability certificates.
Early access
Sigilex is deployed and running in production for early-access customers. If you’re handing autonomous software real authority — and you need to prove afterwards exactly what it was allowed to do — we should talk.