Learnings and patterns for future agents working on this project.
Sign in to like and favorite skills
Learnings and patterns for future agents working on this project.
PRIMARY LANGUAGES: Python
USAGE CONTEXT: Python is the sole programming language used for this project. It's a Python package (neo4j-graphrag) designed to facilitate Neo4j's GraphRAG (Graph Retrieval-Augmented Generation) features. The codebase contains:
src/neo4j_graphrag/tests/docs/Project uses uv for dependency management (pyproject.toml), pytest for testing, mypy for type checking, and ruff for linting. No other programming languages are present in the active codebase.
BUILD SYSTEMS: Poetry, Hatchling, Sphinx, Make BUILD COMMANDS: uv run pytest tests/unit/, uv run ruff format --check ., uv run ruff check . --fix, uv run mypy ., make html (in docs/) BUNDLING: Hatchling (wheel builder)
TESTING FRAMEWORKS: pytest, coverage
TEST COMMANDS:
poetry run pytest tests/unit/uv run pytest tests/unituv run pytest tests/e2euv run coverage run -m pytest tests/unit && uv run coverage report --fail-under=90TEST ORGANIZATION:
tests/unit/tests/e2e/unittest.mockE2E TESTING:
tests/e2e/ARCHITECTURE PATTERN: Modular Multi-Provider RAG Framework with Neo4j-Native Knowledge Graph Integration
DIRECTORY STRUCTURE: Layered package organization with provider abstraction -
src/neo4j_graphrag/DESIGN PATTERNS: Template Method (Retriever base with abstract get_search_results), Strategy Pattern (multiple retriever/LLM implementations), Provider Pattern (pluggable LLM/embedding/vector DB adapters), Adapter/Bridge Pattern (external DB wrappers), Component/DAG Pattern (experimental pipeline with configurable components), Factory-like Configuration (AbstractPipelineConfig instantiation from dicts), Observer Pattern (EventNotifier callbacks).
DATABASE: Neo4j-native architecture with three query categories - vector search via
db.index.vector.queryNodes()db.index.fulltext.queryNodes()apoc.meta.data()apoc.schema.nodes()API DESIGN: RAG pipeline follows init→compose→search flow with polymorphic
search(query_text, top_k, filters, retriever_config)RetrieverResultLLMInterfaceLLMInterfaceV2invoke(input, message_history, system_instruction)DEPLOYMENT STRATEGY: PyPI package distribution with GitHub Actions CI/CD pipeline
CONTAINERIZATION: Not containerized. Project designed as a Python library, not a containerized application. E2E tests use Docker Compose for service dependencies (Neo4j, Weaviate, Qdrant) but the package itself is distributed as a pure Python wheel.
CI/CD: GitHub Actions with multi-stage pipeline:
HOSTING: PyPI (Python Package Index) - distributed as
package. End-users install via neo4j-graphrag
. No server-side hosting required. Examples provided for integration into user applications.pip install neo4j-graphrag[optional-extras]
ENVIRONMENT MANAGEMENT:
Key characteristics:
This AGENTS.md was generated using agent-based project discovery.