Review of the AI Agent for QGIS plugin by TerraLab

[music] Hello, welcome back on my channel. In today's video, I'm going to show you a new plug-in from Terab. It's called the AI Agent plugin. In the past, I've evaluated several agent plugins, but this one definitely is much better than the others. Instead of clicking through menus, searching for data sets, or chaining tools together manually, the AI agent lets you describe what you want in plain language. And it does the work directly inside your CQIS project. It reads your layers, your projections, your map extent, and then it plans the steps, finds the data, runs the tools, styles the layers, and even checks the results before showing it to you. In this video, I'll walk you through how the plug-in works, how to send your first message, and how AI agent can handle real GIS tasks. What I also like very much of the plug-in is that every action is reversible. The plug-in creates a checkpoint for each message, so you're always in control, and with built-in access to dozen of open data sources, you can go from an empty project to a fully styled map with a single prompt. So let's dive in and explore how AI agent brings natural language GIS directly into QGIS. Installing the plug-in is very easy. Just go to the plugins manager there. Search for Terab and you will see all their plugins. In this video, we'll discuss the AI agent by Terab plug-in. Just install it. After installing, you'll find the panel there. Sign in. Just follow the steps in the browser to create an account or sign in with your existing account. At the top of the panel, you can find icons to start a new chat, to check the chat history, or to open the settings dialogue. In the settings dialogue, you can find the connections to data sources that the agent can use, such as open street map or even national data sources like POC in the Netherlands. It will give you an explanation of what it can use and how it will credit the data. Here you see it for open street map. In the settings dialogue, you can also write your name and give some extra personal information that can help also on how it should respond. Here you can also find keyboard shortcuts and links to tutorials and information about your plan. Let's give our first prompt. I'll ask to add a satellite base map and zoom into Rotterdam with a bit of a speed up here in the recording. I see a very nice result. It advises me to install the quick map services plugin to have more base maps. So, uh, let's do that. And there we go. Let's close that panel. A really nice feature is that we can expand every action and see what tools it's using from QGIS and what it's exactly doing. At the bottom of the prompt dialogue, there are settings to tweak the algorithm. By default, it's set to medium, which plans, but we can switch it to low, which is included in the free plan. It will not make a planning of its tasks. And we can switch it to high. Here we keep it at medium. We can also change the AI permissions for several tasks. It can ask you permission or you can choose for autopilot. Here we keep it as balanced between the two cases. Let's give a new prompt. I would like to add the municipality boundary polygon of Rotterdam using the paid data source and give it a red outline. And here it starts processing and it easily finds the polygon and displays it. A very nice feature is that we can roll back to previous states by using the undo button. After each prompt, it stores a break point. So here you can see where we are currently and go back to the start of the chat. Let's restore this before our last prompt. And here we see the project at that state. When I go to the undo button again, I can go back to the last run. And there we see the last state that was saved. Let's start a new chat for a new task. I would like it to make a heat map of the restaurants and cafes in the center of Rotterdam. And here you see that it correctly locates the center of Rotterdam. and it starts downloading restaurants and cafes from overture maps and it created this heat map adds a styling and gives us the requested result. Now I'm going to challenge the agent with more advanced tasks. I have a new project and selected some river segments and I want to create a 100 m buffer around the river segments. It starts executing the task. It use a processing algorithm and then I need to remove some layers and it found some problems. So it needs to fall back on some Python code for which I need to give permission. And here's the result. If I zoom in, I find the buffers around the selected rivers. However, it didn't dissolve the buffer. So our next prompt is to dissolve these buffers. And there it goes. Let's check the result. And here we see the dissolve buffer. Next, I would like to see a hill shade. So I ask if it can render a hill shade. And it calculates a hill shade. It's not a rendered hill shade. As you can see here, it's just a single band gray, but uh it works fine. It has some transparency. Let's now improve the visualization of the rur channels. Here they have a field with straa orders and I ask it to render the channels with the interpolated line renderer. This however doesn't work out as I thought because it doesn't know the renderer. Let's ask specifically to use that sub renderer. And there it doesn't know how to use that. Let's now see if it can apply the inverted polygon shape burst fill to the catchment polygon. Unfortunately, it also doesn't know how to do that. Then I ask explicitly if it can use blending instead of transparency for the hill shade and to use multiply. Also there it can't find this. The catchment boundary polygon here was derived from a ruster and shows a staircasing. So I'll ask if it can smooth this boundary. It's a bit hard to compare with a fill color. So, it ask it to give it a red outline. If I zoom in, I can compare the two lines. And I see indeed that there was some smoothing. So, this works correctly. Let's ask if it can clip the elevation model with the catchment boundary polygon. And of course, keeping the styling here, it did a great job, but now it needs to fix the hillshade. So, I ask it to create a hillshade for the clipped DEM and try to use the hillshade renderer. And there we see that it did work. Let's ask if it can blend the hillshade with the DEM using multiply. And there we see that it did work. Now I only want to see the clipped layers. So I'm going to ask to remove the unclipped ones. I don't see any difference. Let's see what happened here. The layers have been removed, but apparently the canvas hasn't refreshed, but it did the job well. Now I would like to add some towns and cities to the catchment map. and I ask if it can do it with open street map using a certain key and value combination. It says that the area is too large. I first uh tried to just use the map canvas, but that's not enough. So, I asked to split it into 25 smaller tiles that was suggested. And there it goes. for every tile and it's finding some points in some of the tiles and in the end it's merged and it removes the separate tile results. So that's a great example of automization and seems to work well. It would be nice if these points are nicely styled and have a label. So I ask a specific styling. It uses the name field. And to prevent that I have to do a lot of clicking, I switch to the autopilot mode. And there it fetches all the towns. It styles and labels it. And here we see the nice result. Cities and towns now have the same symbol and that is not very clear. So I ask if it can show a hierarchy and this is then the result which looks good. Just to test some raster tools I'll ask to calculate the slope and aspect of the elevation and style it. That seems to work well. So, that's already a nice map. Now, I would like to see decorators with a north arrow, scale bar, title, acknowledgements. It looks like it's going to create a print layout. That's not exactly what I asked, but it's nice that it can do that. So, let's see what it comes up with. So, it says that it created it, and I can find it in my print layout. And this is the result. It's a good try. I would make it portrait rather than landscape and also some elements are overlapping with each other which is not great for readability and I would place things a little bit different but let's see what we can tweak here. So let's ask the agent if it can change the layout to portrait instead of landscape and also make sure that elements don't overlap. So here it goes again. Well, let's uh check the result. And there it is. I would personally uh zoom the map a little bit so it fits better. The paper, the elements are not overlapping anymore, but it's not brilliant. But at least this is a good try and a starting point to further um correct your map. In this video, I've evaluated the AI agent plugin from Terab. I think it's a very good plugin and one of the best that I've seen uh for agents in QGIS. Seems to be specifically trained for QGIS functionality and can even make print layouts for many workflows. It works very nice and it can download data automatically. Um I think the print layout uh it's nice that it can generate it but it uh doesn't come close with styling also and with uh designing the print layout to uh what human creativity will do but it's a good starting point to further tweak these results with some human intelligence. So for sure I'm going to follow further progress of this plug-in and u I'm happy to read your comments. See you next time. >> [music]