v0.2.0

MANIMA v0.2.0 makes the server deployable by someone other than its author, and puts CI in front of it. No change to core/, adapters/, server.py, or the render Dockerfile's contents — this is packaging and guarding, not new tool behaviour.

Deployment

• One-command render-only deploy (make deploy): builds the pinned render image, verifies the container daemon, and proves a scene renders — needs only a container host, no GPU / vLLM / Qdrant.
• Generate path is an explicit opt-in compose profile (make generate-up), never a prerequisite for render (invariant 6).
• Single .env.example enumerating every MANIMA_* knob; pinned dependency locks per package (requirements.lock, generated on Linux/py3.12).
• Deployment smoke test asserts a scene actually reached SUCCEEDED with a retrievable artifact — verified live under rootless podman (a 4344-byte MP4 from a trivial Circle scene).

CI pipeline

• Offline test suites (manima_server + harness), ruff lint, hadolint, and openspec validate --specs --strict on every push / PR — all GPU-free.
• Render-image build verified on every relevant change; publication to ghcr.io/groscy/manima-render gated to release tags, authenticated with the built-in workflow token.
• The generate/GPU surface (Apertus/vLLM, Qdrant grounding) is declared out of CI scope, not faked green — a GPU-less runner cannot exercise it.

Published on this tag

ghcr.io/groscy/manima-render:pinned and ghcr.io/groscy/manima-render:0.18.1 (Manim CE 0.18.1).