fAIr connects mappers with geo-AI models they can run and adapt in their own area. It is a shared collection of models: a mapper chooses a model, runs it over an area of interest, validates what it finds, and can fine-tune it on local examples to produce a local model. fAIr is developed by the Humanitarian OpenStreetMap Team (HOT) and the people who contribute to it.
fAIr offers its outputs as open data layers on the Humanitarian Data Exchange and Hugging Face. Bringing AI-generated features into OpenStreetMap is outside fAIr's aim; see fAIr and OpenStreetMap.
The documentation is at hotosm.github.io/fAIr. Useful starting points:
- What is fAIr and How it works
- Models available today and planned
- Using fAIr
- Developer setup and Architecture
- Roadmap
The full stack needs Docker with Compose 2.23.1 or newer, Git, and at least 8 GB of RAM.
git clone https://github.com/hotosm/fAIr.git
cd fAIr
cp env_example .env
docker compose up --buildThe app and API are at http://localhost:8000, with Swagger UI at http://localhost:8000/api/docs/. The local stack signs in with AUTH_PROVIDER=dev: authorize API calls with Bearer dev-token.
The stack also runs the STAC catalog on port 8082, MinIO on 9000 and 9001, MLflow on 5000, ZenML on 8080, and Postgres on 5434. A fresh stack starts without base models; Register a base model describes how to add one. Run locally covers the services in detail, and Developer setup covers running the backend on the host.
backend/ holds the Django REST API. Its apps cover accounts, datasets, the model registry (base models and local models), trainings, predictions, feedback labels, banners, stars and saved user state. The backend orchestrates ZenML pipelines and the STAC catalog through the fair-py-ops package, stores artifacts in MinIO or S3, and runs background jobs with django_tasks. See Backend.
frontend/ holds the React single-page app, built with pnpm and Vite. See Frontend.
docs/ holds the source of the documentation site. chart/ holds the Helm chart used for production deployments. docker-compose.yml defines the full local stack, and test.py runs an end-to-end check against a running stack through the public API.
Base models live in the fAIr-models repository. A model that follows its contribution guide can be registered in fAIr as a base model; see Contributing a model.
Read the code of conduct and the contribution guide. Questions and discussion take place on the HOT Slack.
fAIr is licensed under the GNU Affero General Public License v3.0. See LICENSE. Base model, imagery and label licenses are on Licensing.
