opengeolab-geocopilot 0.2.0

GeoCopilot: a Codex-powered geospatial research agent for JupyterLab

GeoCopilot for JupyterLab

GeoCopilot is a Codex-powered geospatial research agent embedded in the right side of JupyterLab. The Python distribution installs all three parts together:

• the prebuilt JupyterLab 4 frontend extension;
• the Jupyter Server backend and notebook execution tools;
• the bundled OpenGMS platform skill, activated only by an explicit user request.

Install

GeoCopilot requires Python 3.9 or later, JupyterLab 4, Node.js, and the Codex CLI.

npm install -g @openai/codex
python -m pip install opengeolab-geocopilot

Configure the provider before starting JupyterLab:

export OPENAI_API_KEY="your-api-key"
export JUPYTER_AGENT_CODEX_MODEL="gpt-5-mini"
# Optional for an OpenAI-compatible gateway:
export JUPYTER_AGENT_CODEX_BASE_URL="https://example.com/v1"

jupyter lab

The package automatically enables its Jupyter Server Extension. Verify the installation with:

jupyter labextension list
jupyter server extension list
opengms-notebook --help

Install In A Docker Image

For an existing JupyterLab 4 image, install the Codex CLI and this package during the image build:

RUN npm install -g @openai/codex \
 && python -m pip install --no-cache-dir opengeolab-geocopilot==0.2.0

API keys and provider URLs should be injected at container startup rather than stored in the image. When OpenGMS model or data-method services are used, set NODEJS_BACKEND_URL to the OpenGeoLab backend address reachable from the container.

Build A Release

Install release tooling and run the checked release builder:

python -m pip install -e ".[release]"
./scripts/build-release.sh

The wheel and source archive are written to dist/<version>/. Test an unpublished release with a fresh environment or TestPyPI before uploading it to PyPI.

Publish

Create an API token in PyPI and keep it outside the repository:

export TWINE_USERNAME=__token__
export TWINE_PASSWORD="pypi-your-token"
python -m twine upload dist/0.2.0/*

After publication, downstream systems only need the installation command shown above. A new package version must be published for every update because PyPI does not allow replacing an existing release.