Turn text into visuals. Markwise is a document editor where you highlight a passage and AI turns it into a diagram — ~90 diagram types, 26 render styles, a freeform canvas board, a presentation builder with 200+ slide layouts, and export to PNG / SVG / PDF / Word / Markdown / PPTX.

Now a fullstack app: accounts, companies with role-based access control, an application-owner admin panel, and a pluggable AI provider layer.

Select text → ✦ Visualize → pick a diagram → switch layouts live. Every type ships multiple layouts (e.g. Tree as Two sides, Indented, Top-down grid).

Editor	Canvas board
Presentation builder	Dashboard
Application-owner admin panel	A diagram, up close
AI provider keys & models — stored encrypted, set from the UI	AI request log — every prompt & response, for review
AI usage — requests, tokens & est. cost (individual / company / app-admin)

# Requires Node 18+ and PostgreSQL
createdb markwise   # or use the credentials in server/.env.example

cd server
cp .env.example .env   # fill in DATABASE_URL and AI keys
npm install
npm run migrate
npm run seed           # prints the app-owner login once
npm run dev            # → http://localhost:3000

Sign in at http://localhost:3000/login. The account surface is a single-page React app with clean URLs: the dashboard at /docs, org settings at /org/:id, and the owner panel at /admin. The editor lives at /index.html?doc=:id and shared docs open at /share/:token.

# App + PostgreSQL in one command. Optional: export CODEX_AUTH_JSON / APP_OWNER_PASSWORD first.
docker compose up --build   # → http://localhost:3000

The image runs migrations and the seed on boot (both idempotent) and restores Codex CLI credentials from the CODEX_AUTH_JSON env var (the contents of ~/.codex/auth.json after a codex login on any machine).

render.yaml is a Render blueprint: a Docker web service plus managed PostgreSQL. Create a Blueprint in the Render dashboard, point it at this repo, and set the two secrets it asks for: APP_OWNER_PASSWORD (seeded admin) and CODEX_AUTH_JSON (Codex CLI auth — rotate anytime by re-running codex login and updating the var).

• Editor — write a doc, select text, hit ✦ Visualize; pick from live diagram previews; click any element to recolor, move, scale, or rotate it (text and per-item icons alike), and tailor a visual's content to the diagram type
• Canvas — every visual as a draggable card on a pannable, zoomable board
• Present — build a deck from the document; 50 themes, 200+ slide layouts, full-screen present mode, PPTX/PDF export
• Companies & RBAC — default Owner/User roles plus custom roles with per-permission checkboxes; single-use invite links
• Admin panel — users, companies, audit log, the active-provider switch, plus per-provider API keys & models stored AES-256-GCM encrypted in the DB and managed from the UI, a full AI request log (every prompt, response, provider, model, status & latency) for review, and detailed AI usage dashboards (requests, tokens & est. cost by provider/model) at three scopes — individual, company, and platform-wide
• AI providers — Google Gemini via REST (default), OpenAI Codex SDK, Claude Code headless CLI, and Claude SDK/API (gated by a policy toggle). Keys & models are set in the admin panel (encrypted at rest) with env vars as fallback; the active provider is switchable, with automatic failover. No key? The built-in offline parser still generates diagrams deterministically.

See CLAUDE.md for architecture, conventions, and the RBAC model, and DIAGRAM-TERMINOLOGY.md for the diagram component vocabulary and layout-variant naming reference. The platform roadmap (AI control plane, enterprise identity, governance) lives in ROADMAP.md.