There is probably no GIS practitioner in China who has not heard of Tianditu. It is a national-level geographic information public service platform led by the Ministry of Natural Resources. It is jointly built, shared, and operated through national, provincial, and city (county) nodes. It is no exaggeration to say that Tianditu enjoys a near-sacred status in China's GIS community. The online basemaps, place-name searches, administrative boundaries, geocoding, and route planning used in daily work all depend on Tianditu. For example, consider the large-screen basemaps commonly used in WebGIS: while various other services have increasingly started charging, as long as an enterprise completes verification on Tianditu, the call quota is generally sufficient. No wonder it is constantly taken advantage of by everyone. Recently, this cyber Bodhisattva of the GIS world released a Smart Edition, and its features are quite distinctive. Today I will walk you through it.

What Is the Smart Edition?

A screenshot makes it clear: it is basically a web version of a GIS industry agent, covering a GIS industry knowledge base + multiple GIS professional agents + SKILL tools + MCP.

It reintegrates Tianditu's functions based on AI agents, bringing users a new interaction experience. I did some rough testing. Currently it includes 7 agents, 2 SKILLs (in practice there seem to be more than these two), and one MCP.

I ran the following small tests on commonly used functions.

Data Download

This is a major function for many people using Tianditu. Previously, Tianditu's data download was buried fairly deep, and users had to search carefully to find it. Now users no longer need to explore on their own; a single sentence is enough, as shown below:

After the system ran for a while, a download dialog popped up directly, and the download completed. Of course, it cannot do things it does not support. For example, if you ask it to download the administrative boundary of Wuhan East Lake High-Tech Development Zone, it will tell you that it does not have it either.

Remote Sensing Interpretation (Data Extraction)

After switching to the remote sensing basemap, draw a selection area and ask it to extract relevant data. For example, take data near China University of Geosciences Wuhan and extract buildings and roads within this area.

The current function does not support road extraction, but buildings were successfully extracted. The test download and extraction speed was quite fast; I am not sure whether the data had been precomputed. In addition, the extracted data does not seem to support direct download at present. But that did not stump me. I loaded it in QGIS as follows:

PS: Because the official service does not support this function, I will not share the method. Those who need it can try it themselves or give feedback to the official team.

Development Support

Tianditu API development and integration has always been a concern for GIS developers. Normally, you need to find the API documentation yourself. Now Tianditu officially provides the latest integration method. Of course, if you use Claude Code or Cursor, such capabilities are usually integrated directly, and users only need to configure their own key.

For development scenarios, Tianditu has also created a dedicated development agent, at:

https://ai.tianditu.gov.cn/ai/dev

For example, I entered a prompt like this: search information on the internet and draw a map of Zheng He's voyages to the Western Oceans.

It automatically searches for relevant information and draws it directly on the map. It even compiles information on the various ports and related historical context. Hmm, it really has something. The generated map can also be shared directly. The sharing address of the map I just made is:

https://ai.tianditu.gov.cn/ai/dev/share/muzdq76t-596087212f

Smart Map Styling

There is another rather meaningful feature I have to mention. I do not know how many people, like me, need to customize basemap styles when working on large-screen requirements. Take Amap, which I use most often: its free version basically cannot meet business needs for this feature, and the professional version costs 399 RMB per month.

Tianditu, however, provides a smart map styling feature that can meet similar needs, for example, a tech-dark style.

However, in my testing, it does not currently seem to expose fully custom basemap styles; it only switches among default styles. I look forward to further opening this up in the future.

Data Product Query

Tianditu currently integrates many data products (most require payment), and it now supports natural-language queries. For example, ask it to query digital elevation data products for Jiangxia District, Wuhan, and place an order.

However, I could not find where the order it placed for me went. In addition, this smart portal has many other functions, and I will not describe them one by one.

Some Questions

Besides the order issue just mentioned, I also encountered inconsistent responses. Take the following prompt as an example: "Find historical and cultural landmarks near Hankou Jiangtan." Below is the response I got the first time.

But later, when I wanted to write the article, using exactly the same prompt, it could not do anything at all.

Of course, this is one characteristic of large models: they cannot guarantee consistent responses. But two completely opposite results before and after are still a bit unusual. There are also some minor bugs. For example, after clicking an agent, clicking the history records in the lower left corner does nothing. I believe such small issues will be resolved in future iterations.

Summary

Overall, this Tianditu feature update is not simply adding an AI assistant. Instead, it uses AI technologies such as MCP, knowledge bases, and SKILLs to reconnect originally scattered capabilities such as data download, development integration, and data product queries through agents. Now a task can be initiated with one natural-language sentence, greatly lowering the barrier to use.

As a national-level public service platform, Tianditu has authoritative data sources and compliant basemaps with map approval numbers behind it. This is an advantage many commercial map AI assistants do not have. If MCP and SKILLs can be further opened up and integrated into tools such as Claude Code and Cursor, their value to developers may be even greater than web-based chat.

Overall, Tianditu's attempt at AI is commendable, and the official team will likely continue iterating. Those interested can try it out.