Turns the domain knowledge of a Cortex instance into presentation decks:
a real .pptx with editable text and speaker notes, plus a vector PDF.
Slides are full-image: gpt-image-2 (via Venice.ai) paints each entire slide — typography included — from a prompt that combines the layout's composition brief, the slide's exact approved text, its artwork concept, and a deck-wide design system. The pipeline is gated: you review the outline (every word of it gets painted) for free, approve one paid sample slide for the look, and confirm a USD quote before any full generation. Painted text should be proofread; single slides regenerate cheaply.
Third sibling of cortex-videogen and cortex-trainings — same architecture DNA (resumable file-backed runner, human gates, per-artefact caching, two-channel Cortex research), different product. The deck workflow: outline confirm → style options → sample approval → full generation → assembly → surgical per-slide edits.
npm install
npx playwright install chromium
cp .env.example .env # fill VENICE_API_KEY, CORTEX_BASE_URL, CORTEX_API_KEY
npm run smoke # verify the renderer for $0
npm run dev # http://localhost:3020See docs/pipeline.md for how a deck is produced and CLAUDE.md for the repo guide.