Quick Start

Two repos, one workflow

Repo	What it is	Who it's for
`geo-agent`	Core library — map, chat, agent, tools. Loaded from CDN.	Library contributors; this docs site.
`geo-agent-template`	Starter template. Three files to edit, then deploy.	App builders — start here.

To build a new map app: use the template. You never touch the library repo.

Start from the template

Go to boettiger-lab/geo-agent-template and click Use this template → Create a new repository. You get a ready-to-deploy repo with the three files you need. Edit them for your dataset and deploy.

Live demo: https://boettiger-lab.github.io/geo-agent-template/

Your app is just three files

index.html          ← HTML shell — loads core JS/CSS from CDN
layers-input.json   ← which datasets to show, how to connect to the LLM
system-prompt.md    ← personality and guidelines for the AI assistant

No JavaScript to write. The core library (map, chat, agent, tools) loads from CDN.

index.html

Pick a CDN version and paste it in:

html

<!DOCTYPE html>
<html>
<head>
  <meta charset="utf-8">
  <title>My Map App</title>
  <link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/boettiger-lab/geo-agent@v2.5.0/app/style.css">
  <link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/boettiger-lab/geo-agent@v2.5.0/app/chat.css">
</head>
<body>
  <script type="module"
    src="https://cdn.jsdelivr.net/gh/boettiger-lab/geo-agent@v2.5.0/app/main.js">
  </script>
</body>
</html>

CDN reference	When to use
`@v2.5.0`	Production — pinned, immutable
`@main`	Staging/dev — always latest

layers-input.json

Minimal example:

json

{
  "catalog": "https://s3-west.nrp-nautilus.io/public-data/stac/catalog.json",
  "collections": ["cpad-2025b"],
  "llm": {
    "user_provided": true,
    "default_endpoint": "https://openrouter.ai/api/v1",
    "models": [
      { "value": "anthropic/claude-sonnet-4", "label": "Claude Sonnet" },
      { "value": "google/gemini-2.5-flash", "label": "Gemini Flash" }
    ]
  }
}

See Configuration Reference for all fields.

system-prompt.md

Plain Markdown instructions for the AI assistant. Tell it what data it's looking at, what kinds of questions it should help with, and any style preferences.

markdown

You are a helpful assistant for exploring California's protected lands.
You can show or hide map layers, filter by attributes, and run SQL queries
to answer quantitative questions about the data.

Local development

bash

# From your app folder (index.html + layers-input.json + system-prompt.md)
python -m http.server 8000
# Open http://localhost:8000
# Enter your API key in the ⚙ settings panel

Next steps

Configuration Reference — all layers-input.json fields
Deployment — GitHub Pages, Hugging Face Spaces, Kubernetes
Agent Loop — how the LLM tool-use loop works internally

Quick Start ​

Two repos, one workflow ​

Start from the template ​

Your app is just three files ​

index.html ​

layers-input.json ​

system-prompt.md ​

Local development ​

Next steps ​