GIS ANALYST
| JAX · SPATIAL ANALYSIS · NATURAL LANGUAGE |
Spatial questions,
answered in plain language
Jax for GIS Analysis. Ask spatial questions in natural language; get answers grounded in the Codex, validated against the Ontology, and rendered on every surface — table, map, graph.
AGENT 02G · GIS ANALYST
spatial analysis,at the speedof asking.
For thirty years GIS analysis has been a credential. Project files, layer dialogs, coordinate-system quirks, sprawling toolboxes — the kind of knowledge that lives in the head of one analyst and leaves with them on Friday. The work itself is rarely the bottleneck. Translating the question into the workflow is. Jax collapses that translation. Ask the question in your own language; the agent composes the same eight primitives an analyst would otherwise stitch together by hand, against the Codex and the Ontology you already maintain.
And this is what that changes:
Analysis becomes a question, not a workflow.
Six primitives.
Every analysis Jax runs.
The agent has no special-purpose GIS tool. Every spatial question — from a single buffer to a full network optimization — resolves into composed calls against the same six primitives. The composition is the analysis, and the composition is auditable end-to-end.
- T1[ PRIMITIVE ]
Read
Pull any feature, layer, or attribute table out of the Codex. The agent reaches for the same files your maps render from — no export, no copy, no shadow dataset.
- T2[ PRIMITIVE ]
Write
Materialize the answer back into the Codex as a new layer, summary table, or annotated selection. Every downstream view picks it up on the next read.
- T3[ PRIMITIVE ]
Glob
Find the right layers fast. Pattern-matched lookup across the project — "every census layer below tract resolution," "all parcels touching this corridor."
- T4[ PRIMITIVE ]
Grep
Search attribute content the way you would search code. Land-use codes, owner names, classification strings — matched across layers in one pass.
- T5[ PRIMITIVE ]
Run
Deterministic geometry and math: buffers, intersections, isochrones, choropleths, network distance, terrain queries. The same math an analyst would script — auditable, reproducible.
- T6[ PRIMITIVE ]
Solve
When the question is an optimization — siting, routing, allocation, coverage — the solver dispatches to MIP, VRP, CP, or simulation and returns a defensible answer with the formulation attached.
One loop.
Every question.
A spatial question lands in chat. The agent resolves it against the Codex, validates it against the Ontology, executes it through the Math Engine, and returns the answer to every surface at once — map, table, graph, report. The loop is the same for a buffer and for a continent-scale optimization.
Every query passes through the same loop. Audit trail intact.
- L1
Codex
The persistent file system. Every feature, layer, attribute, and selection.
- L2
Ontology
The typed world model. Entity classes, relationships, constraints, units.
- L3
Math Engine
Geometry, networks, terrain, optimization. Deterministic, reproducible.
Agent
question in
“Which substations are within a 15-minute drive of a flood-risk parcel?”
answer out
Layer + table + provenance. Rendered on every surface. Auditable, replayable, kept in the Codex.
What teams ask
before deploying Jax.
The Codex runs where your data has to run. Deploy it inside your tenancy, your cloud, or a sovereign region and the agent never reads outside that boundary. Imagery, parcels, and PII stay in the same place your maps already serve from. The model receives only the slices it needs to answer a given query, and every read is logged in the audit trail.
Every answer carries its working. The agent surfaces the layers it read, the constraints it checked, the geometry it ran, and the solver it dispatched — in line with the answer, not buried in a sidebar. You can replay the loop step by step, swap an input, and watch it propagate. The math is deterministic; the audit is reproducible.
There is no required vertical. The Ontology is a typed model you author against your own world — parcels and easements, transformers and feeders, drone corridors, forest stands, sensor stations. Jax reads whatever entity types and relationships you declare and reasons against them. The agent never assumes a domain that isn't in your schema.
Air-gapped and disconnected deployments are supported. The same agent loop runs on infrastructure with no outbound connectivity; the model, the solver, and the Codex all sit inside the boundary. Role-based read constraints are enforced at the Codex layer, so an analyst who cannot see a layer cannot indirectly query it through chat either.
The Codex is the surface, not the storage of last resort. Stand up connectors to Snowflake, BigQuery, Databricks, PostGIS, or an object store and the agent reads through to your authoritative tables. Writes flow back as new layers, materialized for the views that need them. Your warehouse stays the system of record; Jax stays the interface.