Client work is someone else's confidential material. We host it where you choose, and we never train on it. Access stays with the people your firm has already allowed.
Platform storage is in the EU or the US, whichever you picked when the workspace was created. Inference runs where that model provider runs: the US for OpenAI, Anthropic, and xAI; China for DeepSeek. Enable only EU-resident providers if the work has to stay in the EU.
Hebno does not use your inputs or outputs to train any model. Providers process inference under their own terms, for the models you have switched on.
AES-256 at rest, TLS 1.2 or higher in transit. If you supply your own provider API keys, those are encrypted at rest too.
Owner, admin, and member roles. Connected sources keep the permissions those accounts already have. We record sign-in, admin changes, and usage.
Your firm, your clients, the people in the work, the numbers: each one travels as a placeholder and comes back as itself. The model still reads the whole document, so the answer is just as good.
Nordvest Pharma
Mette Lindqvist
DKK 42m
Project Aurora
Hallberg & Roe
Client A
Person 1
Amount 1
Project 1
Counterparty A
Names, clients, people, projects, and numbers turn into placeholders before the request leaves Hebno.
It reads the whole document and answers on that. Only the names are different, so the quality holds.
Real names go back on inside your workspace. The draft reads exactly the way you wrote it.
It stays in your organisation's workspace. Names, clients, and numbers turn into placeholders before it leaves, so the model never sees the real ones. We send it over a direct API connection to the model you enabled, the reply comes back for review with the real names restored, and we do not use it to train a model. Usage is recorded against the person, team, and client on the conversation.
A prompt
Written in your workspace
Stored with the conversation.
Anonymised on the way out
Names become placeholders first.
Sent to the model you enabled
A direct API call, placeholders only.
Draft comes back for review
Real names restored, sources attached.
Not used to train a model
A connected source stays in that system, including custom connectors your firm sets up. Hebno can search it, or you can pick a file, using the permissions the account already has. What is used in the conversation is stored there. Someone who cannot open the file in the source system does not get it by asking an assistant.
Files you drag into chat go to your organisation's private storage, encrypted at rest. They stay on the conversation so the history still makes sense. They follow the same deletion terms as the rest of your workspace data.
The reply (a draft, an email, a note) is stored on that conversation in your organisation's workspace. We do not use it to train a model. It stays there until someone on your team copies or sends it, or the conversation is deleted. Then it follows the same deletion terms as the rest of the workspace.
A prompt
Written in your workspace
Stored with the conversation.
Anonymised on the way out
Names become placeholders first.
Sent to the model you enabled
A direct API call, placeholders only.
Draft comes back for review
Real names restored, sources attached.
Not used to train a model
No, and we do not claim to. We use the major model providers you switch on. What makes the work private is what we send them: every name that identifies your firm, your clients, the people, and the projects is replaced with a placeholder before the request leaves Hebno, and put back inside your workspace before anyone reads the answer. The provider does the reasoning while the identities stay with you.
The text of the work, with placeholders where the identifying names were, sent over a direct API connection under a no-training agreement. The list that maps a placeholder back to a real name stays in your workspace, so nobody on the provider's side can work out who the work is about.
People who work on the platform are bound by confidentiality. Access for support is limited to what is needed to fix a problem. We do not browse customer work as a matter of course.
You can export data using product features. We delete customer personal data within 30 days of termination unless the law requires us to keep it. Backup copies may persist for up to 90 days before they are overwritten.
Yes. Accepting the terms incorporates the online DPA for business use. Enterprise customers can request a signed version with custom annexes. Write to support@hebno.com.
Email support@hebno.com. If we confirm a personal data breach that affects your data, we notify you without undue delay, and within 72 hours where feasible.
A short walkthrough of Hebno, shaped around the kind of work your team does. Bring your questions.