A desktop AI assistant for geoscience data processing
English · 简体中文
🚀 Download · ✨ Skill Bazaar · 🌳 Upstream OpenCode
Tackle complex geoscience data processing through conversation — empowering anyone to analyze the planet 🌏...
| Capability | Scope | Examples |
|---|---|---|
| 🗺️ QGIS | All hundreds of algorithms in QGIS Processing | Spatial analysis, vector/raster batch processing, format conversion, accessibility analysis... |
| 🛰️ Google Earth Engine | Full GEE Python API — any remote sensing task you can express | Temporal compositing, classification, change detection, land surface temperature, image download... |
| 🐍 Python | Run arbitrary Python scripts in an isolated env, full scientific computing stack | Geospatial processing, thematic mapping, deep learning... |
| ✨ Skill | On-demand capability packs that let GeoCode grow as you need | Pull community-built packs anytime, build the geoscience ecosystem together... |
No complex setup required. Before installing GeoCode, only two things need to be on your machine:
- QGIS — desktop GIS app, GeoCode calls its algorithms
- A Python environment manager (recommend Miniconda) — isolates Python dependencies so GeoCode's scripts don't interfere with each other
Tip
Give GeoCode's agent a dedicated Python environment (create a fresh one with Conda / Mamba). The agent will install, uninstall, and upgrade Python packages on its own as it works — a dedicated env keeps your other projects clean and helps the agent run more reliably.
GeoCode is currently in public preview as the 0.9.x series. Grab the latest build from Releases. Early adopters and feedback are welcome.
| Version | Released | Changes |
|---|---|---|
v0.9.2 Latest |
2026-05-07 | Fixed Compatibility issues with certain QGIS versionsFixed Settings window stutter on certain Windows machines |
| v0.9.1 | 2026-05-05 | Fixed Google Earth Engine display issue on Windows |
| v0.9.0 | 2026-05-04 | Initial First public preview |
Tip
This repo primarily hosts GeoCode's Skill store; each version's source code is published alongside its release packages.
After installing QGIS and your environment manager, launch GeoCode and type /set in the input box to enter the setup wizard. Just follow the prompts to the end.
Tip
The free models bundled with OpenCode are limited in capability and not recommended for real work. For a noticeably better experience, plug in your own DeepSeek v4 API key, or sign in with a ChatGPT Plus subscription.
Skills are a key piece of how smoothly the agent works. GeoCode ships a built-in management UI for them — browse, install, enable, disable, all click-driven, no manual file shuffling.
Tip
The Skill catalog is small in both quantity and polish in these early days. As more users and contributors join, the bazaar will fill up with higher-quality skills.
All GeoCode Skills live under contributions/ — anyone can open a Pull Request to add a new one. The standard layout of a Skill package:
contributions/<your-skill-id>/
├── manifest/
│ ├── README.md # Human-facing description of the Skill
│ └── meta.json # Metadata: version / description / author / tags ...
└── skill/
├── SKILL.md # Required, the core prompt (LLM-facing, with frontmatter)
└── <dir>/ # Optional, any name and any nesting — references, templates, scripts, etc.
Submission flow:
- Fork this repo and create your Skill directory under
contributions/ - Fill in
manifest/andskill/following the layout above - Open a Pull Request — CI auto-validates schema, version, and structure
- Once reviewed, the maintainer merges and publishes
Tip
Skills are loaded into the agent's prompt and directly affect its behavior. Every PR is reviewed manually before merging to screen for malicious instructions or unsafe operations.
GeoCode benefited from guidance on design and technical decisions from the following research groups:
- China University of Geosciences (Wuhan) · UrbanComp Lab
- LIESMARS, Wuhan University · Urban Spatial Intelligence Research Group