What the platform can do

Real-time streaming answers in Hungarian and English, with light and dark themes. Every request passes through the protection pipeline.

Conversations and knowledge can be organised into projects, and moved by drag and drop.

The system names conversations on its own, and they can be organised with tags.

A sent question can be edited and resubmitted; the answer can be regenerated or stopped.

Formatting, code highlighting and formulas. Longer texts can be edited in place on the canvas panel.

A whole conversation can be downloaded as a branded PDF or Word document.

Long conversations scroll smoothly, and the system caches repeated content intelligently.

A quick search over conversations and actions, opened with a single keyboard shortcut.

A side help panel explains what each page does, in Hungarian and English.

Every control has a short, plain explanation, in the language of the interface.

The interface also works as an installable web app (PWA), fully localised.

The assistant finds answers in your uploaded documents and marks the source it used.

Hovering a citation shows a short extract of the source straight away.

PDF, Word, Excel and PowerPoint; extracted text is screened against the protection rules immediately.

Text in scanned documents is recognised (OCR), making their content searchable too.

An image is only sent if verified free of sensitive data; otherwise only the masked text goes.

The user can see exactly the image the system would send. A human checkpoint.

For complex questions it researches the internal knowledge base over several rounds.

Documents can be organised into thematic collections, uploading whole folders at once.

It profiles uploaded spreadsheets and asks specifically about the relevant details.

The assistant reaches only what the user is already entitled to, with their own permissions.

Calendar reading and approval-gated event handling directly from the conversation.

An installable agent on your own file server, honouring live NTFS permissions, with no inbound port.

Read-only queries; modifying operations require separate approval.

New systems connect through a uniform descriptor, enabled per tenant.

The assistant works from Teams too; protection and DLP rules apply in full.

The system also sends notifications on its own, for example about a finished report.

AI assistance for e-mail directly in Outlook, tied to the corporate sign-in.

The assistant is built into Word, Excel and PowerPoint.

Users and groups are created and removed in sync with the corporate directory.

Sensitive data is pseudonymised before the request reaches the model, and restored in the answer.

Configurable rules decide what may leave: block, warn or mask.

Every conversation is encrypted with its own key; nobody else can technically read it.

Every sensitive field is encrypted; master keys live in a hardware key store, rotated automatically.

Corporate single sign-on with multi-factor authentication and role-based permissions.

Irreversible actions require a fresh, repeated corporate sign-in.

The CamouDatAI operator cannot disable or rewrite the client's administrator account.

What counts as sensitive data cannot be changed by the CamouDatAI operator.

Issuing the local agent identity and public API keys requires client administrator rights.

The private key needed for decryption lives in a separate, offline key store.

Only the CamouDatAI operator can set the backup destination, never the service user.

Whoever writes a rule proposal cannot approve their own; expedited rights expire.

Passwords, keys and tokens left in files are detected and quarantined.

The system watches for unusual usage patterns and flags anything suspicious.

It proposes rule refinements from the logs, always subject to human approval.

Every significant event enters an immutable, hash-chained log, sealed daily.

A NIS2- and ISO-oriented report and evidence pack exports in a single operation.

Retention is configurable, deletion is certified, and a scheduled job removes expired entries.

Who accepted which version and when: a server-side trail that cannot be forged.

Granted for at most four hours, with every use logged.

If the deletion log is missing after a restore, the system flags it loudly and blocks the data.

Scheduled management reports, in the responsible person's mailbox, sent by a human.

The system forecasts expected AI cost and sets budgets and alerts.

A full management interface with every change logged; one switch stops outbound traffic.

Admins can follow in real time how a request moves through the protection pipeline.

Multi-step tasks form declarative chains that can run automatically with management approval.

Automation is built by clicking, with no code; the flow can decide and branch.

A user submits a request and an admin approves it; who asked and who allowed stays visible.

Results save to SharePoint, post to Teams or become an Outlook draft, all screened.

It turns Teams meeting transcripts into structured notes and a task list.

The interface can carry your own name, logo and colours.

It works with several AI providers, adaptable to data residency requirements.

Automatic backup and restore, continuous monitoring, restricted outbound traffic.

If the last backup did not complete, the dashboard says which step it stopped at.

Public API key scope is editable from the admin interface, with a readable audit trail.

The system caps concurrent heavy work; hitting the cap never releases data.

The whole environment deploys from declarative, versioned code, auditably.