v2.0.0 — Astra bundle pathway

The Astra bundle pathway is the new primary Math-To-Manim distribution. math-to-manim, math-to-manim-astra and m2m run GPT-6 Astra through the official Codex SDK, using Codex ChatGPT login. Astra develops the brief, mathematics, storyboard and Manim scene, with independent evidence audits. Jev reviews are advisory by default and can be explicitly disabled or gated.

Download math-to-manim-astra-bundle-2.0.0.zip for the installable wheel and source package, locked Codex runtime manifests, the complete A Frontier for the Zeros film, retained scene and audits, rendered evidence, episode artwork and SHA-256 checksums. The film runs 221.03 seconds at 720p/30 fps.

Installation and bundle guide · Quasi-Riemann episode · Standalone movie

Install Python 3.10+, Node.js 18+ and npm, with Manim's native dependencies, FFmpeg and LaTeX for rendering. After extracting the bundle and activating a Python environment:

python -m pip install "./packages/math_to_manim-2.0.0-py3-none-any.whl[astra]"
math-to-manim setup
math-to-manim login
math-to-manim doctor --review-mode off
math-to-manim run "Explain a mathematical idea through 3D geometry" -q m --review-mode off

setup installs the pinned Codex SDK and CLI 0.156.1 in a user cache, so the wheel works outside a repository checkout. Run evidence is written to runs/astra/ in the working directory. Jev-off authoring retains Astra audits and is labeled astra_only; rendering a saved scene makes zero model/API calls and is labeled not_reviewed. Existing explicit provider commands remain available.

The included film preserves its exact authored scene and original source audits. The completed cloud manifest, candidate/source/movie hashes, metadata and full MP4 decode were checked. The manuscript reports a strict zero-free half-plane Re(s) > 7/8; the full Riemann hypothesis remains open, and this educational project does not independently certify the complete proof or Lean artifacts.

Validation includes offline tests on Python 3.10 and 3.12, clean wheel installation outside the checkout on Windows and Linux, MCP tool registration, actual SDK bridge import, Twine metadata checks and source-bound bundle checksums. These installation checks make no model calls.

Python package index status

The GitHub bundle, wheel and source package are published and their uploaded hashes match the completed cloud build. PyPI publication is pending: the upload was rejected with invalid-publisher because no matching trusted publisher is registered.

To enable the Python index channel, register a pending Trusted Publisher in the owning PyPI account with these values:

Setting	Value
PyPI project	math-to-manim
GitHub owner	HarleyCoops
Repository	Math-To-Manim
Workflow filename	publish.yml
Environment	pypi

Then rerun the failed publish job in this release workflow. The bundle can already be installed directly from GitHub using the instructions above.