Your agent decides how each page should be — density, device, material — then renders every slide as one designed image.

PPT-Zen makes full-image PPT decks with your AI: every page is one designed image, not a text-box template. You say one sentence; it decides how much each page holds, what it looks like and in which material, then assembles the finished .pptx.
Every AI PPT tool turns content into pages. PPT-Zen open-sources the part nobody else does: for each page — how much should it hold, what should it look like, and on what grounds? The finished deck is copyable; the chain of decisions is not.
You bring an agent and an image model. The skill brings the judgment.
One command per runtime — Claude Code, Codex, Cursor, Windsurf, Hermes, OpenClaw.
./install.sh auto # detects your runtimes; or: claude / hermes / ... --global
In your agent:
"Make me a 10-page pitch deck about <project>, Portolan style."
Plan → one image per page → proofread → assembled .pptx.
python3 scripts/assemble_pptx.py \ slides/ deck.pptx
Relayboard: a fictional 10-page pitch generated from one sentence, in one material — with the judgment log for every page. The finished deck.pptx and all ten JPGs ship in the repo under examples/ — clone and open them, no key needed.
Every style ships a reproducible prompt formula. Contribute your own with a folder and a PR.
One command per runtime. The repo is the product — the site just shows you around.
git clone https://github.com/rocsgh/ppt-zen cd ppt-zen && ./install.sh auto # detects your runtimes; or: claude / hermes / ... --global