Question 1

Is ABIS another LLM wrapper?

Accepted Answer

No. ABIS is a context layer and a knowledge graph that any agent can connect to over MCP. Bring your own model — or run none at all and use ABIS as a governed BI tool.

Question 2

How do I connect it with Claude Desktop?

Accepted Answer

Since ABIS runs locally, setup is the same as any other MCP server. After you receive the desktop app, a getting started page walks you through connecting Claude Desktop with clear step-by-step instructions.

Question 3

Why does the ABIS context graph matter?

Accepted Answer

Because an agent without a graph is guessing. The graph gives every agent a versioned map of your data estate: which tables exist, how metrics are defined, which joins are valid, which columns are PII, which writes the policy refuses, and what downstream artefacts depend on what. That single map enables preflight simulation of changes, real driver analysis across domains, and grounded answers that don't hallucinate — all on the same substrate the EU AI Act asks you to evidence.

Question 4

What is preflight simulation — and why does it save iterations?

Accepted Answer

Before an agent through the MCP server modifies a Power BI semantic model, a dbt model, or any node upstream of your dashboards, ABIS walks the graph of downstream dependencies and reports exactly what would break or change. No more 'apply, refresh, see what crashes, undo'. Each loop you avoid is a Tableau refresh you don't pay for and a few thousand tokens you don't burn.

Question 5

Can ABIS answer 'how do I grow revenue by 8%'?

Accepted Answer

Yes — because the graph encodes how revenue actually depends on price, mix, churn, marketing spend, and the entities behind each. Ask the agent a driver question and it traverses those dependencies rather than guessing from a single chart. You get a causal chain across the estate, not a single dashboard cell.

Question 6

Why do AI agents hallucinate less with a context graph?

Accepted Answer

Because the agent reads structure first and data second. The graph tells it which tables exist, which metrics are signed and reviewed, where PII lives, where it is allowed to write, and which writes the policy refuses outright. Hallucination collapses when the model picks from real nodes instead of inventing table names. And because compliance rules live on the same graph, the answer that grounds the question is the same one that enforces the rule.

Question 7

Does my data leave the desktop?

Accepted Answer

Only if you tell it to. The default deployment runs the agent, graph, and MCP server locally. No telemetry, no phone-home. SBOMs are published for every release.

Question 8

Can I migrate from Tableau, Looker, or dbt?

Accepted Answer

Yes. The migrate tool reads your existing semantic model and produces a draft graph for a steward to review. Lineage and joins are preserved; metric definitions are diffed line-by-line.

Question 9

How does ABIS handle high-risk AI systems under the Act?

Accepted Answer

Every read and model change is signed and attributed. Auditor mode exports an Annex IV-shaped report covering risk management, data governance, and transparency.

Question 10

Why made in NL?

Accepted Answer

Because data sovereignty is a stance, not a marketing line. Amsterdam is our base; the Netherlands is becoming a European AI hub and we want to be part of it.

Knowledge, governed. Built for Europe.

Knowledge, governed.
Built for Europe.