Tobi. How it worksImportPricing Sign inStart free

Security and privacy

Your own AI. Not a shared bot.

This is a product you point at the things you would not paste into a chatbot: your clients, your commitments, what people told you in confidence. So the design decisions that protect it are worth showing you rather than summarising in a sentence.

01

Your account cannot see another one.

Isolation is not a permission check bolted on top. Every path, every search vector, and every row carries your account id, so a query that forgot to filter returns nothing rather than someone else's notes.

Storage

Separate by construction

Your files live under a prefix belonging to your account. A request for a path outside it is a 404, not a permission error, because the path never resolves in the first place.

Search

Filtered twice, on purpose

The search index filters by account id server-side, and the code checks it again on the way back. If the first ever regressed, the second turns it into worse search results instead of a leak.

And we test it before every release

  • An automated suite seeds two accounts with distinctive markers and then tries to reach one from the other: by listing notes, by searching, by asking a question, by requesting the file path directly, and by exporting.
  • It includes a positive control, so each account must still find its own marker. Without that, a broken index would silently pass every negative check by returning nothing to everybody.
  • It is a hard gate. If it does not pass, the release does not ship.
02

We ask for as little as the feature allows.

SurfaceWhat we ask forWhat we deliberately do not
Calendarcalendar.readonly, one scopeNo write access. We cannot create, move, or delete anything on your calendar.
EmailNothing. You forward to an address.No mailbox access at all. Reading your inbox would need a Google scope we refuse to request.
Google accountYour address, to label the connectionNo Drive, no Contacts, no Gmail, no profile data.
Sign-inA one-time code by emailNo password. There is nothing to reuse, leak, or steal from us.

The email point is worth dwelling on. Reading your Gmail would be the easy way to build this, and it is why several competitors want that permission. It is also a scope that would put a copy of your entire inbox inside our systems. Forwarding is slightly more work for you and dramatically less exposure, so that is the trade we made.

03

What is encrypted, and what that actually covers.

In transit

Everywhere

HTTPS on every request. Plain HTTP permanently redirects rather than answering.

Third-party tokens

Encrypted at rest

Your calendar access and refresh tokens are encrypted with AES-GCM before they are stored, under a key held separately from the database.

Your notes

Not end-to-end, and here is why

End-to-end encryption would mean we could not index or search your notes, which is the entire product. We would rather tell you that than imply a protection you do not have.

04

Nothing you write trains anything.

Your content is never pooled with another user's, never sold, and never used to train a model. The AI providers that generate your answers process your content under commercial terms that prohibit training on it. This is a contractual and architectural position, not a setting you have to find and switch off.

Hostile content cannot make it act. When answering, your notes are treated strictly as data, and the model reading them has no tools. Something malicious pasted or forwarded into your vault can at worst produce one wrong answer. It cannot send, change, or reach anything.

Read the privacy policy
05

The exit is the strongest guarantee we have.

Every security promise on this page is something you are trusting us about. The one you do not have to trust is the exit: your notes are plain markdown files, exportable in full in one click, on any day including your last. They open in free software forever, and any AI model in ten years will still read them. If we ever became a company you did not want to deal with, you would not be negotiating with us about it.

Where does my content physically live?

On Cloudflare's network, in object storage and a database scoped to your account id. We run the infrastructure; you can pull everything out of it in one click at any time.

Can Hivetivity staff read my notes?

Technically yes, and anyone who tells you otherwise about a hosted product is being loose with the truth. Support access is deliberately conspicuous: opening your account puts a red banner across every page saying a support session is active, and those sessions expire after eight hours. We do not read accounts without a reason to.

If that is unacceptable for your work, this is the wrong product and we would rather you knew now.

What happens to my data if I delete my account?

Your notes, the search index built from them, your settings, and any connected calendar tokens are removed. Export first: once it is gone we cannot get it back for you.

Could a malicious email you sent me take over my account?

No, and the design is the reason rather than a filter. Notes are treated strictly as data when answering, and the model that reads them has no tools available to it. The worst a hostile note can do is make one answer wrong. It cannot send anything, change anything, or reach another account.

Do you have SOC 2?

No. We are a small operation and buying an audit would not make the product safer, it would make it more expensive. What we have instead is described above, and you can verify most of it yourself from the outside.

What if you get acquired or shut down?

You keep working, because your notes are plain markdown files and you can hold a copy of all of them today. That is the honest answer to a question most vendors handle with reassurance instead of structure.

Worth pointing at your real work.

Fourteen days, no card, and everything you put in comes back out whenever you ask.