Skip to content

Releases: mayflower/sc-graph-rag

v2.0.0 — uv / Python 3.12 / graphrag 3.x

22 Apr 16:00
b7c86a4

Choose a tag to compare

Breaking migration of the project's tooling and runtime stack. v1.0.0 is preserved for reference.

BREAKING CHANGES

  • Dependency manager: `pipenv` → `uv`. `Pipfile` / `Pipfile.lock` removed; `pyproject.toml` + `uv.lock` added.
  • Python: 3.10 → `>=3.12,<3.14` (uv provisions 3.12 automatically on `uv sync`).
  • graphrag: 2.7.2 → `>=3.0,<4` (new config schema with `completion_models` / `embedding_models`; defaults moved to `gpt-4.1` and `text-embedding-3-large`).

Existing clones will not pick this up incrementally — re-clone or do a fresh `uv sync` against the new lockfile.

Highlights

  • Resolves all 25 Dependabot alerts that were open against v1.0.0's `Pipfile.lock`.
  • Fixes the `gpt-4-turbo-preview` 404 that made fresh v1.0.0 checkouts unusable out of the box.
  • `pipeline.py` updated for graphrag 3.x: `QUERY` is now positional (was `--query`); also switched from `shell=True` string to an argv list.

Upgrade

See `README.md`. Summary: install uv, `uv sync`, `uv run graphrag init --root ./` (back up any existing `.env` first — `graphrag init` clobbers it), set `GRAPHRAG_API_KEY` in `.env`, `direnv allow`, `uv run python main.py`.

Full changelog: v1.0.0...v2.0.0

v1.0.0 — pipenv / Python 3.10 / graphrag 2.x

22 Apr 15:52

Choose a tag to compare

Initial versioned release of the GraphRAG example implementation, preserved as a reference point before migrating to the modern tooling stack.

Stack

  • Dependency management: pipenv (Pipfile / Pipfile.lock)
  • Python: 3.10
  • graphrag: 2.x (graphrag init scaffolds settings.yaml with the deprecated gpt-4-turbo-preview default)
  • Indexing and querying via the graphrag CLI against OpenAI

Known issues

  • OpenAI has retired gpt-4-turbo-preview; fresh checkouts fail with a 404 on the first index step until the model in settings.yaml is manually swapped.
  • 25 Dependabot alerts open against Pipfile.lock as of this release.

Both issues are addressed in v2.0.0.