marnis.ai

Large enterprises

A private AI cloud. Across all your sites.

Large organisations need AI that scales without data leaving the building. With marnis.ai you connect Nodes into Fleets per site or department, manage rights centrally and keep every action in view.

Several sites, each with its own fleet. Every fleet has several nodes. The organisation manages rights and shared tasks.

Examples

What your Node.Bot can take on.

Each example says what runs today and what comes next.

6 examples shown

Fleets per site and department

Planned

Each site gets its Fleet, each department its Bots. You manage rights, people and approvals in one place.

Example: The Hamburg site sees its Nodes, head office sees all of them. Nobody sees personal content except the person themselves.

Today
One Node with several Bots, roles and rights.
Next
Fleets with several Nodes in the apps, then approvals for data exchange.

Spread large tasks

Idea

A large task is split into parts and spread across several Nodes that work in parallel.

Example: Reviewing a thousand contracts: each Node takes a part, and in the end there is one joint report.

Today
Within one Node, a queue shares the work by importance.
Next
An idea: spreading work across Nodes.

Phone agent with several lines

Planned

Several calls at once, spread across the Nodes of a Fleet. A note for the responsible person after every call.

Example: Monday morning, many calls: the agent picks up, asks follow-up questions and queues the call-backs by urgency.

Today
Talking to the Bot in the apps.
Next
One line per Node at the start; more lines across several Nodes in a Fleet. Company phone systems after measuring and a legal review.

Security lock for data

Idea

A Node that checks what passes back and forth between systems and stops anything unusual before it does harm.

Example: An export contains customer data that must not leave. The lock stops it and asks.

Today
Every action of your own Bots passes through a checkpoint, observing for now.
Next
An idea: the checkpoint for third-party systems too.

Analyse data and report

Idea

Read exports and spreadsheets, find anomalies and write a report that explains what is behind them.

Example: “Why did returns rise in October?” The Bot finds the cluster around one item.

Today
Questions about spreadsheets in the chat.
Next
An idea: regular reports from your systems.

Operation without internet

Planned

For sensitive areas: Nodes that work only inside the company network and send nothing out.

Example: The development department uses its Node with no internet connection at all.

Today
Without a hub, the Node stays in your own network today.
Next
A fixed intranet mode with a network block.

Honest maths

How many calls can a phone agent handle at the same time?

We are still measuring this, and we will not name a number before it has been measured.

What it depends on

  • Computing power of the NodeEvery reply in a call needs the language model for a moment. A Mac Studio with more memory can handle more of them at once than a Mac mini.
  • Queue per NodeCalls take priority over background work. The queue shares out the computing time so nobody waits for an answer.
  • More lines, more NodesIf you expect many calls at once, spread them across several Nodes in a Fleet.
  • One line per Node at the startThat is how the phone agent is planned. More calls per Node will follow only after measuring and a legal review.

Setup

What your setup could look like.

Organisation

One account, rights managed centrally.

Fleets

Per site or department.

Many Nodes

Mac mini or Mac Studio, depending on the load.

Requirements in detail

Is this for you?

Join the waitlist. We will get in touch personally and talk about which tasks your Node should get first.

Join the waitlist