This file provides instructions for AI coding agents (GitHub Copilot, Claude Code, OpenAI Codex, etc.) working in this repository.

Feast is an open source feature store for machine learning that helps ML platform teams manage features consistently for training and serving.

Feast (Feature Store) is a Python-based project that provides:

• Offline Store: Process historical data for batch scoring or model training
• Online Store: Power real-time predictions with low-latency features
• Feature Server: Serve pre-computed features online
• Point-in-time correctness: Prevent data leakage during model training
• Data infrastructure abstraction: Decouple ML from data infrastructure

# Install development dependencies
make install-python-dependencies-dev

# Install minimal dependencies
make install-python-dependencies-minimal

# Format Python code
make format-python

# Lint Python code
make lint-python

# Type check
cd sdk/python && python -m mypy feast

# Run unit tests
make test-python-unit

# Run integration tests (local)
make test-python-integration-local

# Run integration tests (CI)
make test-python-integration

# Run all Python tests
make test-python-universal

# Compile Python protobuf files
make compile-protos-python

# Compile all protos
make protos

# Build Go code
make build-go

# Test Go code
make test-go

# Format Go code
make format-go

# Lint Go code
make lint-go

# Build all Docker images
make build-docker

# Build feature server Docker image
make build-feature-server-docker

# Build Sphinx documentation
make build-sphinx

# Build templates
make build-templates

# Build Helm docs
make build-helm-docs

feast/
├── sdk/python/          # Python SDK and core implementation
├── go/                 # Go implementation
├── ui/                 # Web UI
├── docs/               # Documentation
├── examples/           # Example projects
├── infra/              # Infrastructure and deployment
│   ├── charts/         # Helm charts
│   └── feast-operator/ # Kubernetes operator
└── protos/             # Protocol buffer definitions

• Languages: Python (primary), Go
• Dependencies: pandas, pyarrow, SQLAlchemy, FastAPI, protobuf
• Data Sources: BigQuery, Snowflake, Redshift, Parquet, Postgres, Spark
• Online Stores: Redis, DynamoDB, Bigtable, Snowflake, SQLite, Postgres
• Offline Stores: BigQuery, Snowflake, Redshift, Spark, Dask, DuckDB
• Cloud Providers: AWS, GCP, Azure

The project uses pytest for Python testing with extensive integration test suites for different data sources and stores.

• Use type hints on all Python function signatures
• Follow existing patterns in the module you are modifying
• PR titles must follow semantic conventions: feat:, fix:, ci:, chore:, docs:
• Sign off commits with git commit -s (DCO requirement)
• Uses ruff for Python linting and formatting
• Go uses standard gofmt

Protocol buffers are used for data serialization and gRPC APIs. Recompile protos after making changes to .proto files.

Feast supports Python and Go SDKs. Changes to core functionality may require updates across both languages.

• Blog posts must be placed in /infra/website/docs/blog/ — do NOT create blog posts under docs/blog/ or any other location.
• Blog post files must include YAML frontmatter with title, description, date, and authors fields, following the format of existing posts in that directory.
• All other reference documentation goes under docs/.

1. Follow the contribution guide
2. Set up your development environment
3. Run relevant tests before submitting PRs
4. Ensure code passes linting and type checking