Skip to content

Sovereign AI in practice

BKS Lab

In BKS Lab we make visible how we make confidential corporate knowledge accessible, automate processes, and let specialized AI agents work together under control. The setups combine delivered solutions, our own tools, and interactive demonstrators.

From a reference project

Processing e-invoices end to end

Structured invoice data produces the posting, the document image, and a traceable history in one workflow. All three setups →

Test bench

All 34 invoices posted and archived.

Controls

Switch the accounting system off and feed in the invoices. The posting records wait in front of it, and the history records it. Switch it back on and they continue.

Query about an invoice

Enter a test case or invoice number, or click an invoice or posting record.

Invoice and history

Click an invoice in the archive or a posting record. Invoice and history appear here, without searching inboxes and folders.

Received

34

34 of 34 invoices

In the accounting system

34

Total 13.1m €

Waiting for the accounting system

0

continue once it is reachable

Invoices: the 34 public test invoices of the KoSIT XRechnung test suite (Apache-2.0). The flow abstracts a delivered reference process, system and customer names are replaced by roles. The history shows the order of steps, not measured times.

Areas of work

What we build and test

Seven areas of work show how we make knowledge usable, automate processes, and bring AI systems into operation under control.

See how we approach this under Our approach.

Interactive

Two setups to try yourself

A simulation makes agent handoffs visible. In the A2A chat, you talk to a publicly reachable BKS agent.

  • 01 · SimulationStart the agent flow →

    Control the pace and stations and watch how tasks pass between specialized modules.

  • 02 · Live agentAsk the BKS agent in the A2A chat →

    Ask general questions about BKS-Lab and experience standardized communication over the A2A protocol.

    Your message is sent to the agent. Please do not enter any confidential information.

The lab overview does not process any input. Only the marked A2A chat sends your message to the BKS agent.

Open Source · MIT

BKS open-bridge: context, rules, and tools for AI agents

BKS open-bridge standardizes how AI agents receive project knowledge, rules, and tools. Instead of starting each task without context, they work inside a clearly bounded environment with traceable responsibilities and safeguards.

We run BKS-Lab on BKS open-bridge ourselves and develop the framework publicly under the MIT license. It is not tied to a specific AI model or a single interface.

BKS-Lab runs on it every day · Publicly verifiable · Open source

Isometric illustration: a fanned stack of index cards rests on a flat base, with a small module hovering above it, reading from the stack through thin circuit traces.Isometric illustration: a fanned stack of index cards rests on a flat base, with a small module hovering above it, reading from the stack through thin circuit traces.

Where does a first, clearly bounded pilot pay off?

Bring a process or a document collection. Together we assess whether sovereign AI provides meaningful support and what a solid starting point could look like.

Schedule a conversation