GeoAI
Ask the map instead of clicking it — and let AI do the map work a GIS expert would otherwise do by hand.
In plain language: filter, symbolize, query, compare — and the answer lands on the map itself. A deterministic command mode means the assistant executes named operations rather than improvising, and the whole thing — the model included — can run on your own hardware, so no prompt and no row of data leaves the building. Beyond asking, it does the map work a GIS expert would otherwise do by hand: symbology, extraction, change detection.
Asking the map — and letting it do the map work
From plain-language querying and sovereign deployment through to authoring, extraction from imagery and documents, change detection, and a governed endpoint for your own agents. Live ones open what runs today; the rest are marked Soon.
Ask Your Map
Ask in plain language: filter, symbolize, query, compare — the map answers
Deterministic command modeSovereign AI Deployment
The whole assistant on your hardware with a local model — nothing leaves
On-prem / local-LLMAI Map Authoring
Symbology, popups, labels and layouts authored from a sentence
ESRI drawingInfo / popupInfoSemantic Catalogue Search
Find the right layer by meaning, not by remembering its name
Embeddings over the catalogueAI Briefing Generator
A written situation brief from the current map, on a schedule or on demand
Reporting practiceDocument-to-Map Extraction
Permits, reports and PDFs turned into mapped features
Extraction + geocodingImagery Feature Extraction
Buildings, roads, assets and land cover extracted from imagery
STAC MLM extension · SAM-class modelsChange Detection Alerts
Tell me when something changed here — encroachment, construction, clearing
Change detectionAnomaly & Hotspot Detection
Surface the clusters and outliers a human would not scan for
Spatial statisticsSpatial RAG over Documents
Answer questions across the organisation's spatial documents and records
Retrieval-augmented generationAgent Tools Endpoint (MCP)
Give the company's AI agents a safe, governed way to do geospatial work
Model Context Protocol specificationThe map holds the answer — but only the GIS team can get it out
Every question about the data queues behind the few people who know the tools — and the cloud assistants that could answer it are exactly where your data is not allowed to go.
What it covers
On Strata, or on your existing Esri services
The differentiator. Ask ships with the platform; sovereign / local-LLM is the seam Esri's cloud AI cannot fill.
Sovereign by default
Deployed on your own servers, on-prem or in your cloud — with a local model option, so no data or prompt leaves the building.
Honest boundary
Naming what this doesn't do is what makes where it wins believable.
Where it stops
Read and explain, not author of record. In solution repos Ask stays off until round 2 — do not sell it as configured.
Build it yourself, or have us deliver it
Every solution starts from an open, AI-guided template — build it yourself for free, or have us deliver and support the finished result.
DIY — the free open template
Clone the open (MIT) repo, follow the AI-guided setup, and run it on Strata Free — on your own data, on your own servers. Yours to customize, no license.
Clone the recipe →Delivered + supported
We stand it up, customize it with AI on your data and jurisdiction, load your first layers, and support it — delivery, license and annual support on Strata Team or Unlimited.
Serve your data. Then talk to it.
Strata serves your spatial data over the ArcGIS REST protocol, then lets anyone explore it in plain language — on-prem and sovereign, at a fraction of Enterprise cost. Keep your Esri licenses.