Introducing Cone of Silence: hardened agent pods

Today we’re opening early access to Cone of Silence - agent pods secure enough for the original Chief and Agent.

A pod is your own always-on AI agent: Claude, working for you around the clock on a dedicated, hardened machine, reachable from your phone over Telegram. Message it like a person; it reads, researches, codes, and runs real tools - while everything it touches stays inside a security boundary we built and operate.

The “we built” part is literal.

Two men seated at a table, each under a clear plastic dome lowered from the ceiling, leaning in to hear each other.
The original. Sealed, deniable, and completely unable to pass a message. We fixed the last part.
Each pod runs Claudette, our open-source hardened agent sandbox, plus the controls we deploy in client engagements: OS-level isolation by default, per-project scoped credentials, prompt-injection screening on everything the agent reads, and fail-closed, paired-device Telegram access. Last week we wrote about the tens of thousands of exposed agents running convenience-first defaults on the open internet - Cone of Silence is our answer: the defaults inverted, and somebody on the hook for keeping them that way.

An agent inside a cone-shaped boundary, with one paired device connected and two other approaches turned away.
One paired device connects. Everything else is refused.

It’s a managed service, not a box of parts: we provision, harden, pair your devices, keep it patched, and send you a monthly security review of what your agent did and what we tightened.

A radio-frequency anechoic chamber: every surface covered in grey foam cones, designed so nothing reflects and nothing escapes.
The engineering version. An anechoic chamber, covered in cones, built so that nothing gets in and nothing escapes. Adamantios, CC BY-SA 3.0

The joke and the engineering happen to share a shape. A room lined with cones is the real way to stop a signal leaving, and the reason it works is that every surface is built to absorb rather than reflect. That is the same bet a pod makes: no single wall doing all the work, but every boundary the agent touches built to give nothing back.

And yes - unlike the original cone, ours actually works. The mechanism is open source, so you can check.

Early-access pricing for design partners: $1,500 onboarding (your scoping call is credited) and from $195/month, with your own Anthropic API key so usage stays on your account at cost. Book a scoping call or request early access →