ComChannels · in effect from 26 September 2026
This policy says plainly what we hold, what we can see, and what we will never do. Where there is an uncomfortable architectural truth, we would rather write it down than imply something we cannot deliver.
ComChannels runs AI over your content on a server. That means our systems must be able to read your content in cleartext at the moment you ask your personal ComChannels agent to do something with it. No product that runs AI on your text server-side can honestly claim it is cryptographically incapable of reading that text, and we will not make that claim.
What we offer instead is a layered trust model — checkable guarantees rather than an impossible promise:
For content that must be impossible for anyone here to read, the honest answer today is that ComChannels is not that product. "Vault Channels" — client-side encryption under a key only you hold — is an idea we are considering, and it does not exist today; nothing in the product currently offers it. It would also be incompatible with the agent and with publishing, which is the trade it asks you to make. We mention it so the boundary is clear, not as a commitment.
Your own content, rendered in your own logged-in browser, is present in that browser. Right-click → Inspect will show it. Hiding it from you would mean hiding it from you. What we do protect is that no other user's content is ever sent to your browser, and that logging out clears the drafts cached locally on your device — which matters on a shared computer.
When you invoke your personal ComChannels agent, the relevant content is sent to our LLM provider to produce a response, and the response comes back to your account. We do not keep a training corpus, we do not run a model of our own, and we do not send your content to any model-training endpoint. Our provider's terms prohibit training on data sent through their API.
The only way any of your content becomes eligible for AI training is if you turn on "Allow AI training" for a specific published Channel. That opt-in covers only that Channel's published versions. Personal Channels, Private Channels, Lines, drafts, chat messages and uploaded files are never eligible, and there is no setting that makes them eligible.
The context we assemble for your agent is built only from your own content. One user's material is never fed into another user's agent, including material they have paid to read.
Your content stays for as long as your account is open, because that is what you are paying us to do. When you ask us to close your account we delete your content within 30 days, and it clears our backups as those backups rotate out within 90 days.
You can ask us for a copy of your data, a correction to it, or its deletion, and you can object to how we handle it. Email privacy@comchannels.sh from your account address. We respond within 30 days. These requests are handled by a person today rather than a button in the product; the right is real either way.
TLS in transit. Passwords hashed with bcrypt. Session tokens in HTTP-only cookies. Uploaded files served through expiring signed URLs. Every content endpoint requires authentication — there is no anonymous read path into private data.
ComChannels is not intended for people under 16, and we do not knowingly collect their data.
Our servers and processors may be located in countries other than yours, so using ComChannels involves transferring your data across borders. We aim to meet GDPR and CCPA standards wherever you are; for a jurisdiction-specific request, email privacy@comchannels.sh.
If we change this policy in a way that affects you, we will tell you by email and in-app before it takes effect — not quietly, and not by updating a date.
Privacy questions and data requests: privacy@comchannels.sh.