About Zenancy

Zenancy is an agent over Australia's open data. You ask a question about a place, an asset or a portfolio. The agent works out which public datasets answer it, joins them, and shows the method.

Why we exist

Australian governments and regulators publish a large amount of data: addresses, cadastral parcels, hazard mapping, climate projections, land use, soil, water entitlements, emissions, company and carbon registers. Most of it is free and most of it is maintained.

Almost none of it is joined. Each agency publishes into its own portal, on its own schedule, in its own schema, with no key that links to any other agency's data. A question such as "which of these sites is on bushfire-prone land, and which are held by the same entity" is answerable in principle. In practice it takes months of assembly work, so it is rarely asked.

That assembly is the product. We ingest each portal, harmonise it, and attach it to a small number of spines so it can be joined. The land spine comes first: parcel, address, coordinate and H3 cell. The entity spine, covering ABN, ACN and directorships, comes next. On those spines, a join across portals becomes one query.

Verticals

The first vertical we shipped is AASB S2 climate disclosure. Mandatory climate reporting gave several thousand entities a physical risk question whose data is public and fragmented. Agriculture, water and carbon follow. Procurement and the entity graph, market and infrastructure data, and renewables siting follow the entity spine.

Each vertical uses the same lake and the same agent with a different expertise pack. That is why the spines are built first.

Principles

Traceability. Every number cites its dataset, vintage and licence. A result that cannot be defended in front of an audit committee is not a result.

Gaps are findings. Australian coverage is uneven, and hazard mapping is the most uneven. We report unassessed records as unassessed. We do not substitute an average.

Open data, openly cited. We do not build proprietary models. The value is in the assembly, the method and the interface, not in withholding the inputs.

The stack

An agent with a sandboxed Python and DuckDB environment, a parquet data lake on Cloudflare R2, parcel-level maps from vector tiles, and Python pipelines that ingest and harmonise each open dataset. Rows never enter the model's context. Results are handled by reference, so summaries are computed rather than estimated by the model.

Status

Private preview. The land spine is live and design partners are working on the first vertical. If you have a question that spans several Australian portals with nothing to join them on, request access.