Today we're releasing LGND Geo, our new agentic product. Geo is a set of geospatial tools that plug into the LLM you're already using, Claude or ChatGPT, so that when you ask it a question about a place on Earth, the answer is based on what the imagery actually shows.

Ask an LLM "where are new solar farms going up in Texas?" and it will return answers based on written content (e.g. news articles, public datasets), without an image to visually verify.  That's the gap we've been working on at LGND, and Geo closes.

How it works

We deliberately did not build our own chatbot or agent. Your LLM is already good at understanding a question and figuring out the tools it needs to answer it. What it doesn't have is a way to look at the Earth, and that's what Geo provides - a new set of tools to let your LLM become an Earth expert.

Geo connects through the Model Context Protocol (MCP), the open standard LLMs use to call tools. Once you add it, you just ask your question the way you normally would. Your LLM decides which Geo tools to call, Geo does the searching, and you get back an answer along with the evidence behind it: the locations, the imagery, and what was found there. You can check its work.

Under the hood the tools run on embeddings. An embedding is a compact numeric summary of what a patch of imagery contains, and because we've already computed them, searching an entire continent for "places that look like this" takes seconds instead of an entire GIS team a week.

What's in it today

Geo gives your LLM the tools below to help answer questions using two sets of Earth Embeddings, available to everyone for free:

• Global Sentinel-2 at 10 m resolution from 2017 through April 2026 at monthly time intervals.
• Contiguous US NAIP aerial imagery at sub-meter resolution from 2020 to 2024 at annual time intervals. 

Today's tools fall into four groups. You don't need to know how to use them, because your LLM picks them on its own, but it helps to know what they can do.

Search finds places that look like something. You can describe it in words ("solar farm") or you can point at an example and ask for more places like it. It can also find places that changed from one thing into another.

Explore lets your LLM look around an area the way an analyst would. It can lay an area out as a grid ranked by how much each part changed between two periods, zoom in and out, and step a single place through time by year, quarter or month to see change and event timelines. It can also compare two places, or the same place on two dates.

Classify builds a simple classifier from a few examples of each thing you care about (water, forest, buildings, bare ground). The examples can be a handful of points on a map, or just a description. It then applies the classifier across an area, and running it again on another date shows what changed. That's how you get from "is there forest here?" to “how much of this area is forest, and how much came back since 2020?”

Check asks whether an LLM’s answer is supported by what’s observed in the imagery for a specific location so you can verify what’s actually there.

To make that concrete, here are a few questions it can answer today. Where in Arizona did open desert turn into solar farms between 2020 and 2023? Here's a lithium evaporation pond in Chile; find me others like it. How far did the 2021 lava flow on La Palma spread, and how much vegetation has come back since?

Each of these can be broken down into a handful of tool calls that your LLM will orchestrate. To that end, we're including a bundle of skills that instruct the LLM on how to work with the tools to build a labeled dataset, track a change across a region over several years, or check a set of findings before you rely on them. It's the same know-how our team uses, written down for your LLM.

Try it

Getting started takes about a minute. 

Claude Code: run this in your terminal.

claude mcp add --transport http lgnd-geo https://geo.lgnd.ai/mcp

Claude (desktop): add a custom connector and paste in https://geo.lgnd.ai/mcp.

ChatGPT: add https://geo.lgnd.ai/mcp as an MCP connector in your settings.

You will be prompted to create an account in one click and then ask about a place you know well, so you can judge the answer yourself. Try: "How much new construction has there been within five miles of my town since 2020?"

Tell us what you asked and where Geo fell short at support@lgnd.ai. That's what will shape what we build next.

What you can build

Because Geo speaks MCP, it isn't limited to a chat window. You can wire the same tools into your own agents and pipelines, like a monthly job that flags new construction near a protected area, or a workflow that turns a few labeled examples into a region-wide land cover map. For a look at what a fully agentic workflow on top of these tools can do, see our post on Earth Agent or try it here for yourself. 

What's next

This is the first release and there's a lot more coming. We'll keep adding tools and capability: more on change over time, more imagery sources, classifiers you can save and share, and commercial imagery for the cases where open data can't see enough. The goal stays the same throughout. Your LLM should be able to answer questions about the physical world grounded in evidence from imagery, not guesses.