Security and trust
Where your data goes, who can see it, and what we never do with it.
Plain answers for security and IT reviewers. What leaves your workspace, who can see what, what qbrin will not do, and how to run it on your own infrastructure.
- No training on your data
- GDPR-aligned
- Self-hosting on Enterprise
Where your data goes
Follow your data, step by step.
This is the path from your sources to a cited answer. Two of the steps send text to a model. We say which, and what.
- 01Your sources
You connect mail, files, chat and tickets. qbrin reads them and keeps each item’s permissions.
- 02Your workspace
Text is split into passages and indexed. Every record carries your workspace ID.
- 03Access check
Before a search runs, it is pinned to your workspace and, where permission checks are on, to what the person asking can open.
- 04Model
The question and the content found to answer it go to the model. Passages are marked as untrusted data.
- 05Cited answer
Each claim is checked against its sources: Verified, Rejected or Need more evidence.
What goes to a model
- When someone asks: the question and the content found to answer it go to the language model, such as matching passages or your compressed knowledge map.
- When content is indexed: passage text goes to the embedding model.
- Both endpoints are settings. On the hosted service the language model belongs to a model provider. On a deployment you run, you can point both at hosts you control.
How it is protected
- Connections to qbrin use TLS 1.2 or 1.3.
- Connector credentials and agent secrets are encrypted with AES-256-GCM.
- Passwords are hashed with scrypt. Session and API tokens are stored only as hashes, so a database leak does not hand over working sessions or API keys.
Who can see it
qbrin is built to follow the access you already set.
qbrin copies each source’s permissions when it reads the content. When permission checks are on for your workspace, it applies them to every question: a person gets answers, citations and results only from what they could already open in the original tool.
| Source | Who can see it in qbrin |
|---|---|
| Only the mailbox owner. | |
| Files in a drive | The people the file is shared with: a named person, your whole domain, or anyone, as the file’s own sharing says. |
| Calendar events | The owner and the listed attendees. |
| Chat channels | Public channels: members of your workspace. Private channels: not shown. |
| WhatsApp conversations | The people in the conversation, and the business number. |
| Anything marked public | Members of your workspace only. |
These are the rules qbrin applies when permission checks are on for a workspace.
Checked before the search, not after.
qbrin narrows what is searched before any model reads a word. A filter on a finished answer would come too late.
One workspace, one boundary.
Every record carries your workspace ID. Every search is pinned to it, on the passage and on its document.
Consoles for teams and clients.
Inside a workspace, a console sees only its own documents unless you share them. If qbrin cannot confirm a console, it searches nothing.
What we never do
Three things we do not do.
These are written in our Terms of Service and Privacy Policy, not only here.
Will qbrin train on our data?
No. qbrin does not use your data, files, messages or questions to train or improve general-purpose or third-party AI models.
Will qbrin sell it?
No. qbrin does not sell, rent or trade your data. It is processed only to provide the service to you.
Are our logins stored in the clear?
No. Connector credentials are encrypted. Passwords are hashed. Session and API tokens are stored only as hashes.
- Google user data is not read by people at qbrin, except with your consent, for security, to follow the law, or when it is aggregated and anonymized.
- Google user data is never used for advertising.
Self-hosting
Run it where your data already lives.
On the Enterprise plan, qbrin can run local-first on your own infrastructure, inside your private cloud boundary.
- Point the language and embedding models at hosts you run. What is sent to a model then stays on your side.
- Custom data residency is part of the Enterprise plan.
- The parts are few: an API service, a worker, Postgres, Redis and your model endpoints.
- API serviceTakes questions and checks who is asking.
- Background workerReads sources, splits passages, builds the graph.
- Postgres with pgvectorContent, passages and vectors for your workspace.
- RedisJob queues, rate limits and budgets.
- Model endpointsLanguage and embedding models. Hosted, or yours.
Security for the AI you adopt
Give your AI the task, not your company.
Every AI tool you adopt asks for wide access. qbrin puts three controls between your company and those tools.
Know
A list of every AI tool, agent and API key that can act, with what each can reach.
See KnowEarly accessScope
Each AI gets only the data its job needs. qbrin narrows what passes through qbrin. A tool that keeps its own direct grant to a source is not narrowed.
See ScopeLiveCheck
Go, hold or ask a person before an AI acts. Try it in the public sandbox. It runs in shadow mode, so nothing is enforced.
See CheckQuestions
Questions reviewers ask.
Does qbrin train AI models on our data?
No. qbrin does not use your data, files, messages or questions to train or improve general-purpose or third-party AI models, and it does not sell your data. This is written into the Terms of Service. To answer a question, qbrin sends the question and the content it found to a language model endpoint. On a deployment you run, you can point that endpoint at a host you control.
Can we run qbrin on our own infrastructure or in our VPC?
Yes, on the Enterprise plan. qbrin can run local-first on your own infrastructure, inside your private cloud boundary. You can also point the language and embedding models at hosts you run, so what is sent to a model stays on your side. There is no public self-hosting guide yet, so talk to us about your setup.
Is qbrin GDPR compliant?
qbrin is GDPR-aligned. For your data, qbrin acts as a processor and follows your instructions. A Data Processing Addendum is available on request. qbrin has no published security certifications yet, and we do not claim any.
Can we choose where our data is stored?
Custom data residency is part of the Enterprise plan. Tell us the region you need when you talk to us.
Who can see what in qbrin?
qbrin is built to follow the access you already set. It records each source’s permissions when it reads content. With permission checks on, people get answers only from what they could already open in the original tool. Permission checks are a per-workspace setting, so confirm it is on for yours before you connect real data.
How are credentials and passwords protected?
Connector credentials and agent secrets are encrypted with AES-256-GCM. Passwords are hashed with scrypt. Session and API tokens are stored only as hashes. Connections to qbrin use TLS 1.2 or 1.3.
How do we delete our data?
Disconnect a source to stop syncing, then delete the indexed content from the admin portal or with the account and data deletion flow. Indexed content and sync data are erased within 30 days.
How do I report a security problem?
Report it privately through the security section of the support page, not in public. We look into every report and keep you updated.
Bring your hardest questions.
Bring your security reviewer to a 20-minute walkthrough and ask anything.