Skip to content
SILO.RED, trust in a less certain world
Company

Trust in a less certain world.

Software now acts on its own, with real permissions, on real machines. We think every organisation that lets an AI agent in should be able to see what it does, stop it when it should be stopped, and prove both afterwards. SILO.RED is how.

What we believe

Watch the agent, not the prompt.

What an agent was asked matters less than what it then did. So SILO sits on the machine, below the API, and judges agents by their behaviour.

  • Behaviour over signatures. New agents appear every week; how an agent behaves is harder to disguise than what it is called.
  • Proportion over panic. Responses match the evidence, and the serious ones wait for a person.
  • Honesty over reassurance. A tool that cannot see something says so.
  • Your estate, your data. Self-hosted, so the record stays with you.
WHERE IT STANDS

SILO.RED runs today.

Endpoint agents on macOS, Linux and Windows. Cortex self-hosted, federating over mutual TLS, with licensing, signed distribution and enrolment in place. Tested continuously by our own red-team tooling and a simulated estate of hosts, gateways and storage.

We are working with a small number of design partners ahead of general availability.

Company background

Built by people who have run systems that could not be allowed to fail.

The team behind SILO.RED has spent decades building and securing systems where a mistake is expensive: operating-system kernels, defence and intelligence, global banking, law and public health. SILO.RED applies that experience to a new kind of risk.

Operating-system kernelsLow-level platform engineering, which is where SILO watches agents from.
Defence and intelligenceThreat modelling and security for organisations with determined adversaries.
Regulated industriesBanking, law and healthcare, where controls have to be shown, not asserted.
Applied AIBehavioural models and AI systems in production, the root of SILO’s scoring.

SILO.RED is designed and built in our own AI development lab, where it is attacked every day by the red-team tools built alongside it. The company is Agencio APAC Pte Ltd, based in Singapore and working with organisations in the UK and Asia.

We keep the people behind SILO.RED out of public view. A company that defends against adversaries does not publish a list of whom to target. If you would like to know more about the team, ask and we will tell you directly.

Ask about the team →

Research

The thinking behind it, written down.

SILO.RED rests on a series of papers covering its philosophy, architecture, detection methods, threat models and hardware trust. They are available to partners and investors on request.

I

Philosophy

Why AI agents need a watcher of their own.

II – III

Architecture and methods

The layers, the score and how detection works.

V

Threat modelling

How an agent turns against the estate it serves.

IX – X

Guardian and hardware trust

Protecting the watcher, and binding it to the machine’s keys.

XI

The confused deputy

When an agent gets another service to act for it: what can be seen, and what cannot.

ON REQUEST

Ask for a paper

Tell us which and why, and we will send it.

Request →
DESIGN PARTNERS

Run it with us

Organisations putting agents into production and ready to shape the product.

Get in touch →
PARTNERS

Bring it to your clients

Managed-service providers, cloud partners and consultancies running AI governance work.

Get in touch →
INVESTORS

Back it

We are talking to investors and strategic partners as the product moves towards general availability.

Find out more →

SILO.RED is a product of Agencio APAC Pte Ltd, Singapore.