Cartographer

On any data or AI program inside a large enterprise, the first quarter is spent hunting, not building: finding which team owns which system, filing access tickets, working out which of dozens of near-identical tables actually holds the data. The map lives in people’s heads. The access lives behind approval queues.

A leader runs one tool. It scans the reachable estate across cloud, on-premises and laptop, and reads each source’s structure, not its records. A small on-device model interprets what each source means and how sources relate. Where it cannot reach, it drafts a precise access request naming the exact address, port, protocol and metadata scope.

Estate map and report

For the sponsor. A visual, plain-language map of what data exists and how it connects, with an explicit “what we could not reach” section.

Schema graph

For the technical team. A connected, confidence-scored graph plus a retrieval index that downstream agents and engineers can query. This is the substrate generative AI initiatives need.

Access checklist

Each blocked source becomes an approval-ready line item, exported into your own ticketing and governance. You approve; Cartographer never grants.

• Metadata-only egress, architecturally enforced. No payload data leaves. Interpretation happens locally, on-device.
• Runs local or in your cloud. A signed binary, fully offline. Or a container and Terraform inside your boundary.
• Authorized-scope probing only. Documented rules of engagement, full audit, mapped to NIST 800-53.

1. THE FIRST STEP

   Pick one business unit that spans a few layers: a claims workflow touching a warehouse, a FHIR API and a mainframe extract. Read-only, in an agreed scope, map reviewed together. Low risk, nothing leaves your environment.