Can shhh.bot read room messages?

No. Text and image bytes are encrypted in the paired browsers before Cloudflare receives them. The service stores ciphertext and limited routing metadata.

Is shhh.bot post-quantum secure?

New delivery-only rooms use hybrid ML-KEM-768 and P-256 protection, but the implementation requires independent audit. Conversation rooms and pairing remain classical in the current preview.

Private handoffs for humans + agents

Your agent has something to tell you.

Q: What is shhh.bot?

shhh.bot is an end-to-end encrypted browser room for one human owner and one authorized AI agent browser to exchange text and images.

Q: Does shhh.bot prevent the AI provider from retaining data?

No. shhh.bot keeps the final exchange outside the visible AI chat transcript, but the AI provider may still retain prompts, outputs, tool activity, screenshots, safety records, or reasoning traces involved in processing it.

Give your agent a private room for text and images, so the final handoff does not have to live in the visible transcript of your AI chat. Everything is encrypted in the browser before it leaves the device.

Create your private room See how it works →

No email. No password. One human. One authorized agent browser.

Agent · just now

I finished the image study. Sending the final direction here so it doesn’t live in our chat history.

concept-final.png

You

Received. Delete this room tomorrow.

End-to-end encryptedHybrid PQ deliveryPasskey onlyNo public discoveryOwner-controlled deletion

A separate channel

Outside the chat window.
Not outside the provider.

shhh.bot is being created so an agent can deliver final text and images through a separate encrypted room instead of pasting them into the AI harness’s visible chat thread and saved transcript.

This is a boundary, not invisibility. The AI provider or agent harness may still retain the instructions, generated output, tool and browser activity, screenshots, safety records, or reasoning traces involved in processing the request.

A quieter handoff

A private room in three deliberate steps.

Agent shows a QR

The authorized agent browser opens a single-use pairing screen. No human account is exposed.

You approve on your phone

Scan, use your passkey, choose access, and create one room from your own device.

Browsers encrypt locally

Text and image bytes are encrypted before Cloudflare receives or stores anything.

Honest security

We can’t read your room.

We still can’t see the content of your room. shhh.bot stores ciphertext and limited operational and accountability metadata. Room keys remain in the paired browsers.

Your AI provider may still retain prompts, generated output, browser activity, or reasoning traces. shhh.bot cannot erase information already retained by an endpoint.

Read the complete security model →

Early access

One private room

$0 during preview

One human + one agent browser
Encrypted text and images
Passkey-only account
Revoke, delete, and replace

Create room

Coming soon

Multiple agent rooms

Every agent, one quiet place

$99 / year

Multiple simultaneous rooms
Independent agent permissions
Configurable expiration
All shhh.bot features

Plain answers

Questions worth asking.

What is shhh.bot?

An end-to-end encrypted browser room for one human owner and one authorized AI agent browser to exchange text and images.

Can shhh.bot read the messages?

No. The paired browsers encrypt content before Cloudflare receives it. The service stores ciphertext and limited delivery metadata.

Does pairing prove the other party is AI?

No. V1 proves that the human approved a particular browser. It does not cryptographically prove that software—not another human—controls that browser.

Is it quantum-safe?

New delivery-only rooms use hybrid ML-KEM-768 plus P-256 encryption. That implementation still requires independent audit; conversation rooms and pairing remain classical.

Does this keep data away from my AI provider?

No. It keeps the final handoff out of the visible AI chat transcript. The provider may still retain prompts, outputs, tool activity, screenshots, safety records, or reasoning traces.