GeoPilot is an AI-powered geospatial analysis assistant plugin for QGIS that lets you control QGIS through natural language. Tell GeoPilot what you want to do in plain Chinese or English โ it will automatically generate and execute QGIS Python code to process data, run analyses, and create publication-ready figures.
๐ฅ Key Innovation: Unlike other AI plugins that only chat, GeoPilot actually executes code in QGIS. Your maps, buffers, NDVI, and classifications happen in real-time โ no copy-pasting required.
- Natural Language โ QGIS Operations: Describe tasks in plain language, GeoPilot generates Python code for you
- Reviewed Code Execution: AI-generated Python is shown in chat; by default you confirm before it runs (auto-run is opt-in)
- 18 AI Providers: OpenAI, DeepSeek, Moonshot, Qwen, Zhipu, Yi, Mistral, Cohere, Perplexity, xAI Grok, Together AI, Fireworks AI, Groq, Anthropic Claude, Google Gemini, Ollama (Local), Baidu ERNIE, iFlytek Spark
- Smart Context Awareness: Auto-detects current layers, CRS, fields, and available algorithms to inform the AI
- Safe-by-default networking: SSL certificate verification, HTTP error handling, and a blocklist of dangerous builtins (os.system, subprocess, shutil.rmtree, eval)
| Category | Capabilities |
|---|---|
| Vector Analysis | Buffer, Clip, Union, Intersect, Dissolve, Spatial Join, Merge, Reproject, Field Calculator |
| Raster Analysis | Slope, Aspect, Hillshade, Contour, Raster Calculator, Reclassify, Polygonize |
| Remote Sensing | 28+ spectral indices (NDVI, EVI, SAVI, NDWI, MNDWI, NBR, etc.), Land Cover Classification, Change Detection |
| Spatial Statistics | Hotspot Analysis (Getis-Ord Gi*), Moran's I, LISA, Kernel Density, Voronoi |
| Network Analysis | Shortest Path, Service Area |
| SCI Figures | Study Area Map, Land Cover, Accuracy Assessment, Change Detection, Spatial Pattern, Graphical Abstract |
- Chinese & English interface and AI responses
- All Chinese AI providers natively supported (DeepSeek, Moonshot, Qwen, Zhipu, Baidu, Spark, Yi)
- QGIS 3.30+ (tested on 3.44.2)
- Internet connection (for cloud AI providers)
- API key for your chosen AI provider (except Ollama which runs locally)
1. Download GeoPilot-v1.1.0.zip from GitHub Releases
2. Open QGIS โ Plugins โ Manage and Install Plugins
3. Click "Install from ZIP" โ Select the downloaded file
4. Restart QGIS
5. Enable GeoPilot in Plugins โ Installed tab# Clone
git clone https://github.com/xingguangYan/GeoPilot.git
# Copy to QGIS plugins directory (Windows)
xcopy /E /I GeoPilot "%APPDATA%/QGIS/QGIS3/profiles/default/python/plugins/GeoPilot/"Click the GeoPilot icon ๐ in the QGIS toolbar, or go to Plugins โ GeoPilot โ GeoPilot Chat.
Click Provider Settings (โ) to expand settings:
Provider: DeepSeek # Select your preferred AI
Model: deepseek-chat # Select model version
API Key: sk-... # Enter your API key
Base URL: (leave default) # Most providers have pre-configured endpointsJust type what you want in natural language. GeoPilot will:
- Understand your request and gather the current QGIS project context
- Generate QGIS Python code and show it in the chat
- Ask you to confirm before running the code (uncheck "Auto-execute" in Provider Settings to skip)
- Execute the confirmed code and show results (layers, output, or errors) in the chat
| Category | Example |
|---|---|
| Buffer | ๅฏน้่ทฏๅพๅฑๅ100็ฑณ็ผๅฒๅบ / "Buffer the roads layer by 100 meters" |
| Clip | ็จ็ ็ฉถๅบ่พน็่ฃๅชๆๆๅพๅฑ / "Clip all layers by the study area boundary" |
| NDVI | ่ฎก็ฎๅฝๅๅฝฑๅ็NDVIๅนถๆพ็คบ / "Calculate NDVI from the current image" |
| Classification | ็จ้ๆบๆฃฎๆๅ5็ฑปๅๅฐๅฉ็จๅ็ฑป / "Run Random Forest classification with 5 classes" |
| Change Detection | ๆฃๆต2015ๅฐ2020็ๅๅธๆฉๅผ / "Detect urban expansion from 2015 to 2020" |
| Hotspot | ๅๆ็ฏ็ฝชๆฐๆฎ็็ญ็นๅบๅ / "Find crime hotspots in the point layer" |
| Export Figure | ็ๆ็ ็ฉถๅบ็คบๆๅพ๏ผๅ
ๅซๆฏไพๅฐบๆๅ้ / "Create a study area map with scale bar and north arrow" |
| Batch | ๅฏนๆฏไธช็ข้ๅพๅฑๆๅญๆฎต่ๅ / "Dissolve each vector layer by its name field" |
| Provider | Get API Key | Best For |
|---|---|---|
| OpenAI | platform.openai.com | General purpose, best quality |
| DeepSeek | platform.deepseek.com | Cost-effective, Chinese support |
| Moonshot/Kimi | platform.moonshot.cn | Chinese, long context |
| Qwen (Tongyi) | bailian.console.aliyun.com | Chinese, Alibaba ecosystem |
| Zhipu GLM | open.bigmodel.cn | Chinese, code generation |
| Yi (01.AI) | platform.01.ai | Chinese, fast responses |
| Anthropic Claude | console.anthropic.com | Long context, analysis |
| Google Gemini | aistudio.google.com | Free tier available |
| Ollama (Local) | No key needed (local) | Offline use, free |
| Provider | Models |
|---|---|
| OpenAI | gpt-4o, gpt-4o-mini, gpt-4-turbo, o1, o1-mini, o3-mini, gpt-4.1 |
| DeepSeek | deepseek-chat, deepseek-reasoner, deepseek-v3, deepseek-r1, deepseek-v4-pro, deepseek-v4-flash |
| Moonshot | moonshot-v1-8k, moonshot-v1-32k, moonshot-v1-128k, moonshot-v1-auto |
| Qwen | qwen-max, qwen-plus, qwen-turbo, qwen-long, qwen2.5-72b-instruct |
| Zhipu | glm-4-plus, glm-4, glm-4-flash, glm-4-air, glm-4v-plus |
| Yi | yi-lightning, yi-medium, yi-large, yi-vision, yi-large-turbo |
| Mistral | mistral-large-latest, mistral-small-latest, codestral-latest |
| Anthropic | claude-sonnet-4-20250514, claude-3-5-sonnet-20241022, claude-3-5-haiku |
| gemini-2.5-pro-exp-03-25, gemini-2.0-flash, gemini-1.5-pro | |
| Ollama | llama3.3, llama3.1, mistral, qwen2.5, deepseek-r1, phi4 |
| Baidu | ernie-4.0-8k, ernie-3.5-8k, ernie-speed, ernie-lite |
| Spark | 4.0Ultra, 4.0, 3.5, 3.0 |
| xAI Grok | grok-beta, grok-2, grok-2-vision, grok-3 |
| Perplexity | sonar-pro, sonar, sonar-reasoning |
| Groq | llama-3.3-70b-versatile, mixtral-8x7b, gemma2-9b-it |
| Together AI | meta-llama-3.3-70b, deepseek-ai/DeepSeek-R1 |
| Fireworks AI | llama-v3p3-70b, qwen2p5-72b, deepseek-r1 |
| Cohere | command-r-plus, command-r7-12-2024, command-a-03-2025 |
GeoPilot features a modern dark-theme chat interface:
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ GeoPilot - AI Geospatial Assistant [? Help] โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ โ Provider Settings (collapsible) โ
โ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โ
โ โ Provider: [DeepSeek โผ] Model: [.. โผ]โ โ
โ โ API Key: [โขโขโขโขโขโขโขโขโขโขโขโขโขโขโขโข] โ โ
โ โ Base URL: [______________] โ โ
โ โ System Prompt: [______________] โ โ
โ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ ๐ฌ Welcome to GeoPilot! โ
โ ๐ Your AI Geospatial Assistant โ
โ โ
โ 1. โ Click Provider Settings to configure โ
โ 2. ๐ Type your task in natural language โ
โ 3. ๐ค GeoPilot executes code automatically โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ [Describe your task here... ] [โถ Send] โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
Click ? Help to reveal quick-prompt buttons for common tasks:
- ๐ Study Area Analysis
- ๐ฟ Vegetation Analysis (NDVI)
- ๐๏ธ Land Cover Classification
- ๐ Change Detection
- ๐ Spatial Pattern Analysis
- ๐ Export Research Report
GeoPilot/
โโโ __init__.py # Plugin entry point (classFactory)
โโโ geopilot.py # Main plugin class (menu, toolbar)
โโโ geopilot_dialog.py # Chat dialog + auto-execution engine
โโโ metadata.txt # QGIS plugin metadata
โโโ README.md # This file
โโโ LICENSE # MIT License
โโโ Makefile # Build automation
โ
โโโ providers/ # ๐ง AI Provider System (18 providers)
โ โโโ __init__.py # Provider registry (all 18 registered)
โ โโโ base.py # BaseProvider, register_provider()
โ โโโ openai_compat.py # OpenAI + 13 compatible providers
โ โโโ anthropic_provider.py # Anthropic Claude
โ โโโ google_provider.py # Google Gemini
โ โโโ baidu_provider.py # Baidu ERNIE (OAuth)
โ โโโ spark_provider.py # iFlytek Spark
โ โโโ ollama_provider.py # Local Ollama
โ
โโโ scripts/ # โ๏ธ GeoAI Analysis Engine
โ โโโ geoai_data_manager.py # Data I/O (Shapefile, GeoJSON, GPKG, etc.)
โ โโโ geoai_vector_analysis.py # Buffer, Clip, Union, Spatial Join
โ โโโ geoai_raster_analysis.py # GDAL, SAGA, GRASS raster ops
โ โโโ geoai_remote_sensing.py # 28 spectral indices, 8 classifiers
โ โโโ geoai_sci_figure.py # Figure 1-8, Graphical Abstract, Poster
โ โโโ geoai_paper_agent.py # Journal recommendation, Cover Letter
โ โโโ geoai_pipeline.py # 6 pre-built analysis workflows
โ โโโ geoai_report.py # Research report generator
โ โโโ geoai_qgis_bootstrap.py # QGIS environment initialization
โ โโโ geoai_gee_bridge.py # Google Earth Engine integration
โ โโโ geoai_env_setup.py # Environment detection
โ โโโ geoai.py # CLI interface
โ
โโโ icons/ # ๐จ Plugin icons
โโโ icon.png
โโโ icon.svg
The core innovation of GeoPilot is the review-then-execute pipeline:
flowchart LR
A[User: "Buffer by 100m"] --> B[AI generates Python code]
B --> C[Code extracted from response]
C --> D{Auto-execute enabled?}
D -- No (default) --> E[User confirms in dialog]
D -- Yes --> F[Dangerous-call guard]
E --> F
F --> G[QGIS exec runs code in restrained namespace]
G --> H[Layer created, results shown]
- Context Gathering:
build_qgis_context()collects all layer info, CRS, fields, and available algorithms - AI Processing: The LLM generates QGIS Python code based on context and user request
- Code Extraction: fenced ```python blocks are pulled from the AI response
- Safety Gate: If auto-execute is off (default), a dialog asks you to confirm; a regex guard blocks obvious destructive calls (
os.system,subprocess,shutil.rmtree,eval) - Execution: Code runs via
exec()with a curated QGIS namespace (iface, processing, QgsProject, etc.) and print output captured - Feedback: Results (new layers, processing output, errors) are displayed in the chat and logged to the QGIS message log
โ ๏ธ Security note: Code execution is not sandboxed at the Python level. Only auto-execute code from providers you trust, and keep review-confirmation enabled unless you understand the risk.
The analysis engine in scripts/ provides programmatic access to all GeoPilot capabilities:
| Module | Purpose | Key Functions |
|---|---|---|
geoai_vector_analysis |
Vector GIS operations | buffer_layer(), clip_layers(), spatial_join() |
geoai_raster_analysis |
Raster processing | calculate_slope(), reclassify_raster(), contour() |
geoai_remote_sensing |
Remote sensing | calculate_ndvi(), classify_landcover(), detect_change() |
geoai_sci_figure |
SCI figures | create_study_area_map(), create_classification_figure() |
geoai_paper_agent |
Journal matching | recommend_journal(), generate_cover_letter() |
geoai_pipeline |
Workflows | run_forest_disturbance(), run_urban_expansion() |
geoai_report |
Reports | generate_research_report(), generate_thesis_outline() |
geoai_qgis_bootstrap |
QGIS init | init_qgis() |
geoai_gee_bridge |
GEE integration | init_ee(), gee_export() |
make install # Install to QGIS profile
make zip # Create distributable ZIP
make clean # Remove __pycache__from .base import BaseProvider, register_provider
class MyProvider(BaseProvider):
ENV_KEY = "MY_API_KEY"
DEFAULT_URL = "https://api.example.com/v1"
def chat(self, messages, system_prompt=None, temperature=0.7, max_tokens=4096):
# Implement API call
return response_text
register_provider('myprovider', MyProvider,
display_name='My Provider',
models=['model-1', 'model-2'],
env_key='MY_API_KEY',
default_url='https://api.example.com/v1')Q: Does GeoPilot actually execute code in QGIS?
A: Yes! Unlike chat-only plugins, GeoPilot runs AI-generated Python code directly in QGIS using exec(). Results appear as new layers, processed data, and visual feedback.
Q: Do I need internet?
A: For cloud AI providers, yes. For offline use, install Ollama locally with models like llama3.1 or qwen2.5.
Q: Which provider is best for geospatial analysis? A: DeepSeek, Qwen, and Claude generally produce the best QGIS Python code. OpenAI GPT-4o is also excellent.
Q: Can I use it without an API key? A: Yes! Use the Ollama provider with locally installed models โ completely free and offline.
Q: How do I get API keys? A: See the Getting API Keys table above. Most providers offer free credits for new users.
MIT License โ see LICENSE.
@software{geopilot2026,
author = {GeoPilot Team},
title = {GeoPilot: AI-Powered Geospatial Analysis Assistant for QGIS},
year = {2026},
url = {https://github.com/xingguangYan/GeoPilot}
}โญ If you find GeoPilot useful, please star the repo!