Skip to content

Lean PyPI install: pip install graphier + graphier setup, trusted publishing - #14

Merged
Joshuaakaspace merged 1 commit into
mainfrom
claude/semantica-repo-evaluation-essnux
Aug 16, 2026
Merged

Joshuaakaspace merged 1 commit into
mainfrom
claude/semantica-repo-evaluation-essnux

Conversation

@Joshuaakaspace

Copy link
Copy Markdown
Collaborator

Follows the merged #13. Makes pip install graphier a real, lean distribution channel — and removes a launch-day trap.

The trap

graphier declared semantica as a hard dependency, and semantica's own dependency list pulls torch + transformers + spacy (multi-gigabyte) — none of which Graphier's deterministic path ever imports. A first-time pip install graphier downloading gigabytes of ML frameworks would have been the top complaint in every launch thread.

The fix

  • Lean dependencies: graphier now depends only on what it actually uses (fastapi, uvicorn, pypdf, python-multipart, numpy, pandas, scipy, networkx). Full Semantica moves to the optional graphier[ml] extra.
  • graphier setup: new CLI entry point installs the engine package alone (pip install --no-deps semantica — quick, no ML downloads). Launching without the engine prints the one command to run instead of a traceback; graphier-mcp gets the same guard.
pip install graphier && graphier setup
GRAPHIER_VAULT=~/notes graphier --demo
  • PyPI trusted publishing: the release workflow gains a publish-pypi job using OIDC (pypa/gh-action-pypi-publish) — no token or secret to manage. Gated on the PYPI_PUBLISH repo variable; the one-time pypi.org setup (pending publisher for this repo, workflow release.yml, environment pypi) is documented step-by-step in docs/launch-checklist.md.
  • README gains the pip quick start ahead of Docker; CHANGELOG updated.

Why PyPI and not npm: Graphier's runtime is Python — the frontend is already built and bundled inside the wheel by the release workflow, so a single pip install ships the whole app. npm would only make sense for a JS-native runtime, which this isn't.

Testing

4 new tests (55 total): engine detection, the missing-engine message + exit code, setup invoking the lean pip command, and the MCP guard. Entry points verified end-to-end locally: graphier setup ran the engine install and both console scripts resolve.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QVbxD1gd3gsC6zXKE6L3ya


Generated by Claude Code

'pip install graphier' previously declared semantica as a hard
dependency — whose own dependency list pulls torch/transformers/spacy
(multi-gigabyte) that Graphier's deterministic path never imports.

- dependencies are now the lean set Graphier actually uses (fastapi,
  uvicorn, pypdf, python-multipart, numpy, pandas, scipy, networkx);
  semantica moves to the optional [ml] extra.
- New graphier.cli entry points: 'graphier setup' installs the engine
  package alone (pip install --no-deps semantica); launching without it
  prints the one command to run instead of a traceback. graphier-mcp
  gets the same guard.
- Release workflow gains a publish-pypi job using PyPI trusted
  publishing (OIDC, no token) — gated on the PYPI_PUBLISH repo variable
  and documented step-by-step in docs/launch-checklist.md.
- README gains the pip quick start; CHANGELOG updated.
- 4 new tests (55 total).

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QVbxD1gd3gsC6zXKE6L3ya
@Joshuaakaspace
Joshuaakaspace merged commit 9cf6858 into main Aug 16, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants