What is NeuraCities? The AI GIS platform that runs the analysis, not just the chat

The problem NeuraCities solves

Most spatial questions are simple to say and slow to answer. "Which parcels near a substation are outside the floodplain?" is one sentence. Answering it the traditional way means finding a parcel layer, a substation layer and a flood hazard layer, downloading each one, fixing projections, clipping everything to the study area, running a buffer, a spatial join and an overlay, then styling a map and writing up the result.

That work usually happens across five or more tools: a GIS desktop application, spreadsheets, a web browser full of open data portals, Google Maps and often a Python notebook. The analysis itself might take twenty minutes. Finding and cleaning the data can take days, and the knowledge of how it was done lives in one analyst's head.

NeuraCities collapses that into one place. The data search, the processing, the map, the chart and the written answer all happen inside a single conversation, and the conversation itself becomes the record of how the answer was produced.

What you can ask it

Anything that has a "where" in it. A few real examples of the kind of questions teams run:

• "Find parcels in Houston within 1,500 m of a substation and 500 m of a fiber line." (a data center siting screen)
• "Which commercial zones in Salt Lake City fall inside the FEMA 100-year floodplain?"
• "Map every building permit issued in Boston in 2023 and show where activity is concentrated."
• "Which census tracts in San Francisco have the highest affordable housing need?"
• "How many Charlotte bus stops have no shelter, broken down by council district?"
• "Where in Phoenix do residents live more than a ten minute walk from a park?"

You can be as precise or as loose as you like. A precise question ("within 500 m") is run exactly as written. A looser one ("good sites for a grocery store") gets a checklist first: NeuraCities states which criteria it will use and which data it needs, so you can see its interpretation before any results come back.

What happens after you press enter

Behind the chat box is a small team of specialised agents coordinated by an orchestrator. Each one has a narrow job and a fixed set of tools.

The important design choice: the language model plans, the GIS engine calculates. The model decides that a 3 mile buffer is needed. A GIS library draws the buffer. Areas, distances, counts and statistics come from tested geospatial code (GeoPandas, Shapely, rasterio, OSMnx, scikit-learn and PySheds), never from the model estimating a number. That is what makes the output usable in real work.

For the full walkthrough of this pipeline, read How NeuraCities turns a plain-English question into a real GIS analysis.

What you get back

Every answer is a set of working artifacts, not just a paragraph:

• An interactive map with each dataset as its own layer, a legend, light, dark, satellite, 3D and Google basemaps, Street View, and a time slider when the data has dates.
• Charts and tables built from the analysed data. Tables export to Excel and charts to PNG, PDF or CSV.
• A written answer with the actual counts and findings, plus a list of the data sources used and the steps taken.
• Reports that can be edited in place and exported as PDF.
• Dashboards: one click turns a conversation into an editable, long-form page with the maps and charts embedded, which can be shared by link.
• Print-ready map layouts from the export canvas: multi-page decks in landscape or portrait, exported as PNG, JPG, SVG or PDF.
• GIS files of any layer, downloadable as Shapefile or CSV for use in ArcGIS, QGIS or anything else.

neuracities.com

Working with your own data

Public data only gets you so far. Most real projects combine it with something private: a client's parcel list, a utility's asset inventory, a survey spreadsheet or a stack of PDFs.

• Uploads: shapefiles, GeoJSON, CSV and Excel, GeoTIFF rasters, File Geodatabases, PDFs, Word documents and images. Spreadsheets with addresses can be geocoded, and tables can be joined to boundaries by a shared key.
• My Database: every file you upload is saved to your own persistent database, so you can reuse it in any future conversation without uploading it again. On Pro, a team can share one database.
• Connectors: connect a PostgreSQL database and NeuraCities can run read-only queries against it live, in every message. Google Drive and OneDrive connectors let you pull files straight from your cloud storage.

More detail in Your data and your team in NeuraCities.

When you want to do it yourself: GIS Studio

Sometimes you know exactly which operation you want and would rather click than type. GIS Studio is a manual workspace inside NeuraCities with a catalog of around fifty operations: buffer and multi-ring buffer, clip, intersect, union, difference, dissolve, spatial joins, hex and point grids, kernel density, hot spot analysis, DBSCAN and k-means clustering, Voronoi polygons, IDW interpolation, weighted overlay, Moran's I, raster zonal statistics, slope, hillshade, reclassification, viewshed, network service areas and shortest paths.

Lighter operations run directly in your browser without using any of your plan's usage. Heavier ones run on the NeuraCities backend. You can switch between asking and clicking on the same data at any point.

How you can trust the result

The fair criticism of AI in GIS is that it can produce a confident map from the wrong data. NeuraCities is built to make that visible instead of hiding it:

• Every layer keeps its source: the publisher, the dataset and the link it came from. Derived layers point back to the source they were built from.
• Every step is listed in the conversation as it runs: which datasets were found, which were loaded, which GIS operation ran on what.
• Requirements are checked. The checklist declared at the start is compared against what was actually delivered, and anything missing is stated in the answer rather than quietly dropped.
• You can inspect any layer: its columns, summary statistics and rows, directly on the map.
• Every operation is logged on the backend, which is how we audit and improve the agents.

Like any analysis, results are only as good as the underlying data. Open data portals vary in how current and complete they are, which is exactly why NeuraCities shows you the source and date of each layer rather than blending them into an unexplained answer.

Sharing and collaboration

Spatial work is rarely for the person who made it. NeuraCities is built for handing results to people who do not use GIS:

• Share links to a conversation. Recipients open the same maps, charts and answer in their browser and can explore the map themselves.
• Invite by email, with a roster of who has access.
• Comment pins that anchor to an exact spot on a map (in 2D and 3D), or to a passage in a report, with threads and resolve.
• Workflows: save a prompt as a reusable template with {{variables}} such as a city or a distance, and rerun it for the next project with new inputs.

The NeuraCities product family

The GIS platform described above is the core product. The same agent approach powers purpose-built applications for specific industries:

Plans and pricing

Current details are always on the pricing page.

How NeuraCities fits next to ArcGIS, QGIS and other tools

NeuraCities is not trying to replace every desktop GIS workflow. Cartographers producing a statewide map series, or teams running enterprise geodatabases, will keep using ArcGIS or QGIS for that work, and NeuraCities exports Shapefiles and CSVs so results move into those tools easily.

Where NeuraCities is different is the part of the job that comes before and after the analysis: finding data you did not already have, running the multi-step analysis from a description of the goal, and delivering something a non-GIS stakeholder can open and understand. We wrote a longer, even-handed comparison in AI GIS software in 2026: how to choose.

Frequently asked questions