Next-gen AI
platform

Artificial intelligence you need not be
afraid to let into live operation.

Baroque seagull seen from below with its wings spread

Marietta

The whole platform: it brings together some seven tools for working with all kinds of data sources.

Marie

An AI bot and chat interface that operates a whole range of clever tools under the bonnet.

How does the whole thing work?

Marie is not just a chatbot. It is a whole architecture, so take it point by point.

Architecture
Client application
Methodology
Security
Data under control

Server Tools get_incident send_email create_task update_record run_report Storage upload_file get_file delete_file Scheduler schedule_job cancel_job run_job Audit get_log get_audit_trail Client Web render_page live_typing Reasoning show_thinking confidence_score Streaming stream_tokens partial_render Notifications email_alert sms_alert push_alert

Architecture

When you close the browser, Marie keeps running: it schedules and launches tasks by the calendar and handles incidents. It never sleeps.

You get the overview in the client application: it shows clearly what it is working on right now. Every step stays traceable in the audit and the activity log.

The client application

However powerful a tool is, it is no good unless it can be easily controlled and watched, and unless you see at a glance what it is doing and what it is working on.

That is why we build Marie's graphical interface with the same care as what runs beneath it: it should be understandable at first glance, not flashy.

Tabs
Reasoning
Tool calls in real time
Answers typed out live
Conversation history
Source citations
Audit log of actions
Attachments and files
Permissions by role

A glowing colourful sphere, the symbol of Marie

A methodology for writing modern ERP systems

The trouble with most AI tools is that somebody bolted them onto a system that never expected them, and that was that.

Marie has adapters that let it understand the context of your system and your business domain perfectly. It moves around in it the way your people do, only it never sleeps and is incomparably faster.

No

AI bolted on from outside

The chatbot stands beside the system. It answers nicely, but it cannot see your data and does nothing in it.
Yes

Marie as part of the system

Adapters hand her your entities, permissions and processes. She moves through the system like one of your people, only faster.

Security

Letting AI roam freely through your data and change it is a recipe for disaster. Before Marie changes anything, it generates a clear plan and describes it all, and every step can be taken back.

For less critical stretches you can switch on bypass permission mode, in which Marie works entirely on its own. No data is ever lost and no harm is ever done.

Plan proposal Read-only

Marie describes what she wants to do and why
Wait

Marie stands still until someone approves the plan.
Checkpoint

You approve the plan, adjust it, or turn it down
Data action Write

The steps run and go into the audit trail
Roll back

Every step can be taken back

Your data under control

Marie (server, database, adapters and audit alike) runs at your place. Only what is necessary for the given task travels to the model, not your whole database.

On top of that the model runs in a mode where it does not train on your data and nothing is stored anywhere, and we can keep the operation inside the EU. Full control and GDPR with no compromises.

Stays with you
The server, the database, the adapters and the audit trail run in your infrastructure.
Only context reaches the model
The minimum the task needs, not your whole database.
Sees only what you may see
Marie works with the permissions of whoever is asking. Data outside their role stays out of her reach.