GeoAgent

An AI agent for geospatial data analysis and visualization.

• Documentation: https://geoagent.gishub.org
• Source code: https://github.com/opengeos/GeoAgent
• PyPI: https://pypi.python.org/pypi/geoagent

Features

• Natural language interface for geospatial data workflows
• 4-agent LangGraph pipeline: Planner, Data, Analysis, Visualization
• Multi-LLM support (OpenAI, Anthropic, Google Gemini, Ollama)
• Multi-catalog STAC search (Earth Search, Planetary Computer, USGS, NASA CMR)
• Code transparency showing generated Python code at each step
• Jupyter-native with interactive MapLibre maps via leafmap
• DuckDB spatial SQL for GeoParquet and Overture Maps
• Raster analysis with xarray, rioxarray, and rasterio
• Vector operations with geopandas

Installation

pip install geoagent

Or install from conda-forge:

conda install -c conda-forge geoagent

With all optional dependencies:

pip install "geoagent[all]"

LLM Setup

GeoAgent supports multiple LLM providers. You need at least one configured to use the agent.

Supported Providers

Provider	Default Model	API Key Env Variable	Install Extra
OpenAI	gpt-4.1	OPENAI_API_KEY	(included)
Anthropic	claude-sonnet-4-5-20250929	ANTHROPIC_API_KEY	pip install "geoagent[llm]"
Google Gemini	gemini-2.5-flash	GOOGLE_API_KEY	pip install "geoagent[llm]"
Ollama (local)	llama3.1	(none needed)	pip install "geoagent[ollama]"

Setting API Keys

Set your API key as an environment variable:

# OpenAI (included by default)
export OPENAI_API_KEY="your-openai-key"

# Anthropic
export ANTHROPIC_API_KEY="your-anthropic-key"

# Google Gemini
export GOOGLE_API_KEY="your-google-key"

You can also add these to a .env file or your shell profile (~/.bashrc, ~/.zshrc).

Choosing a Provider

By default, GeoAgent auto-detects available providers by checking environment variables in order: OpenAI → Anthropic → Google → Ollama. The first available provider is used.

To specify a provider and model explicitly:

from geoagent import GeoAgent

# Use the default provider (auto-detected)
agent = GeoAgent()

# Use a specific provider
agent = GeoAgent(provider="anthropic")

# Use a specific provider and model
agent = GeoAgent(provider="openai", model="gpt-4o-mini")
agent = GeoAgent(provider="google", model="gemini-2.5-flash")

Using Ollama (Local LLMs)

To run GeoAgent with a local LLM via Ollama, no API key is needed:

# Install Ollama and pull a model
ollama pull llama3.1

# Install the Ollama extra
pip install "geoagent[ollama]"

agent = GeoAgent(provider="ollama", model="llama3.1")

Using a Custom LLM Instance

You can pass any LangChain-compatible chat model directly:

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o", temperature=0.0)
agent = GeoAgent(llm=llm)

Quick Start

from geoagent import GeoAgent

agent = GeoAgent()
result = agent.chat("Show NDVI for San Francisco in July 2024")
result.map   # displays interactive map in Jupyter
print(result.code)  # shows the generated Python code

Web UI

GeoAgent includes a Solara-based chat interface with a persistent, interactive map.

pip install "geoagent[ui]"
geoagent ui

Or run directly:

solara run geoagent/ui/pages

Features:

• Persistent map — layers accumulate across queries on the same interactive MapLibre map
• Native widget rendering — full bidirectional map interaction (zoom, pan, click)
• Chat interface — natural language queries with real-time status updates
• Provider selection — switch between OpenAI, Anthropic, Google Gemini, or Ollama
• Code transparency — toggle generated code display for reproducibility

You can also launch it programmatically:

from geoagent.ui import launch_ui

launch_ui()

Architecture

GeoAgent uses a 4-agent pipeline orchestrated by LangGraph:

1. Planner parses natural language into structured parameters
2. Data Agent searches STAC catalogs and retrieves geospatial data
3. Analysis Agent computes indices and statistics with transparent code generation
4. Visualization Agent renders results on interactive leafmap MapLibre maps

License

MIT