Software advice
Best AI GIS Software in 2026
AI GIS software is a geospatial platform that stores, maps and analyzes location data, and adds AI in the form of natural-language questions, imagery detection or agents that run spatial tasks. It differs from a plain map widget, which only draws tiles, and from a business intelligence dashboard, which treats location as one more column. Analysts, planners and developers use it to answer where-questions.
Quick answer
Felt ranks first because every plan, seat price and AI credit allowance is printed. Wherobots fits teams that run spatial SQL at scale, and ArcGIS Online fits organizations that need Esri's data and certifications. Disy Cadenza names the least a buyer can compare.
A GIS buyer in 2026 chooses between very different things sold under one label: a desktop-heavy suite, a cloud map editor, a spatial database, a tile API and a satellite data marketplace. AI is layered on each of them in a different way, so the useful comparison is what the AI touches and what it costs to run. Prices are the first filter, because most enterprise vendors show nothing, while a few print every tier. This ranking weighs printed prices, stated certifications and hosting, how the vendor describes AI use of your data, and which kind of user each product serves. We read vendor pages in November 2026 and ran no spatial benchmark, so speed and accuracy claims belong to the vendors.
The 11 best AI GIS software in 2026, ranked
There is no score. Vendors can pay for more visibility on this website, for example a higher place in a ranking or more room on a page. Commercial relationships can also affect which products we feature and how prominently. How we rank.
| # | Tool | Best for | Free plan | Entry price |
|---|---|---|---|---|
| 1 | Felt | Analyst teams that publish shareable maps and want plain-language questions without desktop GIS | Yes | $199 per month billed yearly for Team |
| 2 | Wherobots | Data engineers running spatial SQL, Python and imagery pipelines on large datasets | Trial | $95 per month usage minimum on Professional |
| 3 | ArcGIS Online | Public-sector and enterprise GIS teams that need Esri data, desktop tools and field apps | Trial | Quoted |
| 4 | FME | Teams that move and clean spatial data between hundreds of systems without writing code | No | Quoted |
| 5 | CARTO | Data teams that keep location data in BigQuery, Snowflake or Databricks and want maps on top | Trial | Pay As You Go, metered monthly |
| 6 | Mapbox | Developers who embed maps, search and routing in a product and want usage-based billing | Yes | Pay as you go, with free monthly allowances |
| 7 | MapTiler | Swiss and EU-leaning developers who want hosted vector tiles, geocoding and an open-source SDK | Yes | $30 per month for Flex |
| 8 | UP42 | Teams that buy archive or tasked satellite and aerial imagery and stream it into a GIS | No | 10,000 credits (EUR 100) minimum |
| 9 | Picterra | Sustainability and sourcing teams that need geospatial evidence on deforestation and EUDR compliance | No | Quoted |
| 10 | Disy Cadenza | German public-sector and utility teams analyzing attribute data and geodata together in dashboards | No | Quoted |
| 11 | QGIS | Analysts and agencies who want a free desktop GIS with server publishing and a plugin ecosystem | Yes | Free of charge when downloaded from QGIS.org |
- 1
Felt ranks first because the pricing page prints every tier, the seat price and the AI credit allowance, with no overage charges. A simple AI question uses about 5 credits, so cost is predictable. Imports cover common GIS formats and a QGIS plugin exists. The trade-off is depth, since it is a map workspace, not a full analysis suite, and its terms leave AI training unaddressed.
Strong on
- Free, Team, Business and Enterprise prices are all printed, with per-seat and add-on costs
- Felt AI credits are capped per plan, so usage stops instead of billing overages
- Imports Shapefile, GeoPackage, GeoTIFF, KML and more, and connects to ArcGIS and WMS/WFS
Weak on
- Free plan cannot upload data, so real work starts at $199 per month on Team
- Terms do not say whether customer data is used for AI training
Skip it if: Governments or utilities that need desktop editing, enterprise geodatabases and a long-established GIS vendor relationship. Free allows creating maps but no uploads. Team adds uploads and 250 AI credits, Business adds MCP.
- 2
Wherobots ranks second because compute rates, region differences and plan minimums are printed, and it lists an EU region in Ireland. It extends Apache Sedona, so existing code runs unchanged, and an MCP server lets agents drive jobs. The limits are audience and data terms: it targets engineers, not map makers, and its privacy policy is silent on AI training.
Strong on
- Spatial Unit rates are printed per region, with a 14-day trial or $95 in credits
- MCP server and VS Code extension let outside assistants run queries and jobs
- Reads GeoParquet, STAC, Zarr and Delta tables and links to Databricks and Snowflake catalogs
Weak on
- RasterFlow runs only in US West and is labeled both public and private preview
- Privacy policy names no hosting provider or location and does not address AI training
Skip it if: Cartographers and planners who want to drag layers onto a map and publish it without writing SQL or Python. Compute is billed per Spatial Unit at $1.50 in US regions, $1.67 in Ireland. Innovation starts at $3,000 per month.
- 3
ArcGIS Online ranks third. Its overview states FedRAMP Moderate and ISO 27001 certifications, a 21-day trial and built-in AI assistants with pretrained deep learning models, which gives buyers more to verify than most suites. The price stays unshown, and the AI assistants differ by app and user type, so what you receive depends on the license you buy.
Strong on
- States FedRAMP Moderate and ISO 27001 certifications, with a 21-day free trial
- Connects to Snowflake, BigQuery, Databricks, Azure and AWS, and works with ArcGIS Pro
- Public map views and data transfer carry no charge
Weak on
- No per-user price is shown, and each license is for one named person
- AI assistant access depends on which apps your user type includes
Skip it if: Small teams that only need to publish a few maps and want a printed monthly price before talking to anyone. Annual named-user licenses in six types, no per-user figure shown. Creator, Professional and Professional Plus include 500 credits.
- 4
FME ranks fourth because its security and AI data statements are among the most concrete in this group: ISO 27001, SOC 2 Type 2 and a written position that AI Assist input does not train models. It is an integration tool for spatial data, so it complements a GIS instead of replacing one, and prices come only by quote.
Strong on
- States ISO 27001 and SOC 2 Type 2, and publishes an AI Assist security whitepaper
- Privacy policy says AI Assist input is not used to train or fine-tune models
- FME Flow Hosted lets customers pick AWS regions in the EU, Canada, US or Australia
Weak on
- No list price, and subscriptions are annual with maintenance renewing each year
- AI Assist helps with authoring such as SQL and regex, not spatial questions
Skip it if: Analysts who want a map canvas and plain-language analysis, since FME builds data pipelines and does not draw finished maps. Annual subscriptions sized by organization, deployment and engine capacity. No list price is published.
- 5
CARTO ranks fifth. Its privacy notice states that neither CARTO nor its subprocessors train AI models on customer content, a rare public line, and a 30-day trial covers every component. The pricing page prints tier names but no dollar figure, and no SOC 2 or ISO claim appeared on the pages read, so the security evidence is thinner than the data statement.
Strong on
- Privacy notice says customer content does not train AI models
- Runs on Google Cloud, AWS, Azure, Snowflake, Databricks and Oracle, with data kept in the warehouse
- 30-day trial includes demo datasets and a CARTO Data Warehouse instance
Weak on
- No prices are listed, and committed tiers need a quote
- Pay As You Go is SaaS only, and self-hosting starts at the Enterprise tier
Skip it if: Teams without a cloud data warehouse, since the platform assumes your location data already lives in one. Pay As You Go is metered with a 12-month contract. Committed tiers are paid annually, by quote.
- 6
Mapbox ranks sixth because it is a developer platform, not a GIS, though its free allowances are printed and it documents MCP servers and agent skills for AI coding tools. Its product privacy policy permits training on de-identified personal data, and the retention and hosting details found were dated. It is excellent for embedded maps and thin for analysis.
Strong on
- Per-API free allowances are printed, such as 100,000 geocoding requests a month
- MCP servers cover geocoding, routing, search, style management and documentation
- Self-hosted Atlas option exists for on-premise or private cloud deployments
Weak on
- Product privacy policy allows training AI on personal data after de-identification
- Business support costs 10 percent of monthly spend, with a $500 minimum
Skip it if: Analysts who need spatial joins, field data collection or a finished dashboard, since Mapbox provides building blocks and not an analysis workspace. Free tiers include 50,000 GL JS map loads and 100,000 routing requests. Support starts at $50 monthly.
- 7
MapTiler ranks seventh because its plans, quotas and overage rates are all printed, and its privacy policy says infrastructure providers hold ISO/IEC 27001 certification. It is a Swiss company with an open-source SDK. The pages read mention no AI features, so it belongs here as a mapping data layer, not an AI product, and a buyer who wants AI features should compare it only as a basemap source.
Strong on
- Free, Flex and Custom plans print quotas, with overage such as $0.15 per 1,000 API requests
- Open-source JavaScript SDK plus WMTS, geocoding and 3D GeoSplats support
- Privacy policy names an ISO/IEC 27001 certified infrastructure provider
Weak on
- No AI features were described on the pages read
- Policy does not name its hosting provider or locations explicitly
Skip it if: Teams shopping for AI analysis or natural-language mapping, because the product is tiles, geocoding and styling. Free covers 100,000 API requests, Flex adds overage billing. Custom plans add a 99.9% SLA.
- 8
UP42 ranks eighth. The credit rate, minimum purchase, bulk bonus and expiry are printed, which most data sellers omit, and the company is a Berlin GmbH. Its pages describe processing but name no AI features, and the privacy policy names no hosting provider. It supplies data and does not replace a GIS, so budget separately for the software that displays and analyzes what you buy.
Strong on
- Credit rate of 100 per euro, 10,000 minimum and 5 percent bonus above 1,000,000 are printed
- Professional includes API and SDK access, and Enterprise adds WMTS streaming into GIS software
- Privacy policy names UP42 GmbH in Berlin as controller under GDPR
Weak on
- Credits expire 24 months after purchase, so unspent prepayment is lost
- Pages read name no AI features and no hosting provider
Skip it if: Anyone who needs a mapping or analysis interface, since UP42 sells imagery access and processing, not a map workspace. Credits cost 100 per euro and expire after 24 months. Price depends on area, resolution and provider.
- 9
Picterra ranks ninth. Its home page describes GeoAI insights for crop yield, carbon, deforestation and EUDR compliance, with a Lausanne address. It shows no plans, no hosting location and no certifications on the pages read, so a buyer has little to check before a sales call. It suits a narrow compliance use well, but only after you confirm plans and hosting directly with the vendor.
Strong on
- Names concrete use cases: deforestation, EUDR compliance, regenerative agriculture and crop yield
- Lists Forge, Tracer and Accelerate as separate technology components
- Swiss address in Lausanne appears on the home page
Weak on
- No prices, hosting location or certifications appear on the pages read
- Integrations are linked but not listed, so connectors need confirming
Skip it if: General mapping or spatial SQL work, because the product is built for environmental monitoring and supply-chain evidence. The pages read show only a Talk to us call to action, with no plans or prices.
- 10
Disy Cadenza ranks tenth because the pages read give the least to verify: no price, no deployment model, no hosting statement and only a general mention of AI and machine learning for non-specialists. What it does show is concrete, such as scheduled reports and embedding in portals. Disy is a GmbH in Karlsruhe, which suits buyers who want a German supplier.
Strong on
- Dashboards combine maps, charts and tables, with scheduled reports by time or event
- Self-service discovery accepts Excel files alongside specialist-system data
- Vendor is a registered GmbH in Karlsruhe, Germany
Weak on
- No price, deployment option or hosting location appears on the product page
- AI is mentioned only in general terms, with no named features
Skip it if: Buyers who need printed pricing, a self-serve trial or a named list of AI features before they speak to a vendor. No price or deployment model is stated. Interested buyers go through a contact form.
- 11
QGIS ranks last in an AI-focused list because the pages read name no AI feature at all, and no commercial support tier or privacy statement was found. As a plain GIS it is the strongest free option here: cartography, digitizing, geoprocessing, QGIS Server for OGC services and community plugins, under the GPL on Windows, macOS, Linux, Android and iOS.
Strong on
- Free under the GNU GPL, with no license fee stated for the desktop software
- QGIS Server publishes WMS, WMTS, WFS and WCS services
- Runs on Windows, macOS, Linux, Android and iOS with plugins listed on the overview page
Weak on
- The pages read name no AI features, so no AI capability can be claimed
- No commercial support tier is named; credit goes to volunteers and contributing organizations
- The privacy page returned HTTP 404, so no vendor privacy statement was found
Skip it if: Teams that need vendor-backed AI feature extraction, a hosted service or a contractual support and privacy commitment. The download page says the software is and always will be free from QGIS.org under the GNU GPL.
How these AI GIS software were judged
What can the AI in a GIS actually do?
The AI claims fall into three groups, and they are not interchangeable. Felt answers plain-language questions and builds maps from them, with a simple question costing about 5 credits. ArcGIS Online lists built-in assistants and pretrained deep learning models. Wherobots and Mapbox expose MCP servers so an outside assistant such as Claude can call their tools. Picterra applies GeoAI to imagery for compliance and deforestation checks. FME limits AI Assist to authoring help such as SQL and regular expressions. Pick the group that matches your task, then test it on one real dataset of your own, because none of these vendors publishes an accuracy figure you can compare.
How are GIS platforms priced?
Pricing models differ more than the numbers do. As of November 2026, Felt prints Team at $199 a month billed yearly and Business at $599, with Enterprise from $1,250. MapTiler lists Flex at $30 a month, and Wherobots Professional carries a $95 monthly usage minimum. UP42 sells credits at 100 for one euro with a 10,000-credit minimum. Mapbox bills by usage above free allowances such as 50,000 GL JS map loads. ArcGIS Online prices annual user types without showing a figure, and FME, CARTO, Picterra and Disy publish no list price on the pages we read.
Is my location data used to train AI models?
Few vendors answer this on a public page. Safe Software states that FME AI Assist input data does not train or fine-tune any model, including those of its processor Azure OpenAI. CARTO's privacy notice says neither it nor its subprocessors use customer content to train AI models. Mapbox says it may train models on personal data after de-identification, and customer data falls under its DPA. Felt's terms allow aggregated de-identified data for improving products without naming AI. For Wherobots, UP42, MapTiler and Disy, the policy was not checked for this beyond the pages read, and ArcGIS points to a Trust Center whose detail was not read.
How hard is it to move a GIS to another platform?
Open formats lower the cost of leaving, and proprietary logic raises it. Felt accepts GeoJSON, GeoPackage, Shapefile and GeoTIFF uploads, and Wherobots reads GeoParquet, STAC and Zarr and runs existing Apache Sedona code unchanged. CARTO keeps data in your own warehouse, so the map layer is the only piece you rebuild. ArcGIS projects accumulate styles, credits and user types that do not travel. UP42 credits expire 24 months after purchase, which turns unspent prepayment into a loss. Before signing, export one finished map or workflow and confirm it opens in QGIS or another tool you already own.
Which GIS software fits my team?
Fit follows who builds and who only looks. A small analyst group publishing maps for others suits Felt, where viewers do not take paid seats on the pages we read. A data engineering team in Snowflake or Databricks suits CARTO or Wherobots. A city or utility with field crews and a license budget suits ArcGIS Online or Disy Cadenza, the latter from a German vendor. Developers adding maps to a product need Mapbox or MapTiler, not a GIS. Teams that only need imagery pick UP42 or Picterra. Match the product class first, then compare price and data terms inside it.
How to choose an AI GIS platform in three steps
- Name the job before the vendor. Write down whether the output is a published map, a spatial query inside a warehouse, a tile service for an app, a batch of satellite imagery or a detection over a region. Each of these belongs to a different product class in this ranking, and comparing a map editor with a tile API on price alone gives meaningless numbers. List the data you already hold, the formats it sits in and who will edit versus only view. That list removes half the candidates before any demo is booked.
- Test the AI on a question you already know the answer to. Take one dataset and a question whose answer you have already calculated, such as customers within a ten-minute drive of a site. Ask each shortlisted product in plain language, and compare the result with yours, noting the credits or compute it spent. Felt's credit counter and Wherobots' spatial unit rate make this cost visible. A product that returns a plausible map with the wrong boundary is worse than one that refuses, so record misses as carefully as hits, then rerun the same question a week later to see whether answers are stable.
- Get the data terms and exit path in writing. Ask each vendor whether customer content trains any model, where the data is hosted and what is deleted after cancellation. Safe Software and CARTO state their position publicly, so use those statements as the benchmark for the rest. Then export one finished map or workflow in an open format and open it elsewhere. If a vendor cannot show an export of layers, styles and scripts, count the rebuild effort as part of the first-year cost, and ask whether prepaid balances survive cancellation.
Frequently asked questions
What is the best AI GIS software in 2026?
Felt ranks first here because its plan prices, seat costs and AI credit allowances are printed on its pricing page. Wherobots is the pick for teams running spatial SQL at volume, and ArcGIS Online suits organizations that need Esri's data catalog and stated FedRAMP Moderate and ISO 27001 certifications. The right choice depends on whether you map, query, or supply imagery.
Is there a free AI GIS platform?
Felt's Free plan allows 3 seats, 25 viewers and 10,000 map views a month, but no data uploads and no AI. MapTiler's Free plan and Mapbox's free allowances suit prototypes, and Wherobots offers a 14-day trial with $95 in credits. QGIS is free and open source under GPLv2+, but its main page names no AI features.
Can a GIS platform work with Claude or ChatGPT through MCP?
Yes, in several products. Wherobots lists an MCP server for agents including Claude, Cursor and ChatGPT. Mapbox documents MCP servers for geocoding, routing, search and developer tasks. Felt includes MCP access on Business and Enterprise, and FME Flow offers an MCP integration layer plus an MCPCaller transformer. Check which plan unlocks it, because Felt's lower tiers do not include it.
How much does ArcGIS Online cost per user?
Esri does not print a per-user price on the ArcGIS Online buying page. It sells six annual user types: Creator, Professional, Professional Plus, Mobile Worker, Contributor and Viewer, each tied to one named person. Creator, Professional and Professional Plus include 500 credits, and Viewer includes none. A quote from Esri or a reseller gives the actual figure.
Which GIS software keeps data inside my own warehouse?
CARTO is built around that model. Its pages describe analysis running in Google BigQuery, Amazon Redshift, Azure, Snowflake, Databricks or Oracle, with the platform acting as a layer on top. Wherobots also reads Databricks Unity Catalog and Snowflake Horizon Catalog tables. Felt Business adds one database or bucket source, and Enterprise allows unlimited sources.
Which GIS vendors are based in Europe?
Disy Informationssysteme is a GmbH in Karlsruhe, Germany, with its Cadenza product. UP42 GmbH is in Berlin, Germany, and sells satellite and aerial imagery. MapTiler AG is registered in Zug and Picterra's address is Lausanne, both in Switzerland. Hosting location is a separate question, and only some of these vendors state it.
Do I pay for viewers in a GIS platform?
It depends on the product. Felt includes viewers per plan, 25 on Free and Team, 100 on Business and unlimited on Enterprise, with extra seats at $49 a month billed yearly. Esri says public map views carry no charge, while Viewer is a named user type for private content. Developer platforms such as Mapbox bill by map loads instead of people.
Is satellite imagery included in GIS software?
Usually not as a flat feature. UP42 sells archive and tasked imagery priced by area, resolution and provider, using credits at 100 per euro. Picterra analyzes imagery you bring or connect, and Cesium and Mapbox supply basemaps and 3D data. Felt and CARTO map imagery you load, and the data fee is a separate line from the platform fee.