OPENQUOK - an Agent-native social scheduler

Problem. AI tools can generate more social content than a person can safely publish. Closed schedulers are browser-only, so agents cannot queue posts, and teams still copy-paste between ChatGPT and Buffer-style dashboards. That creates volume without control: wrong captions, wrong accounts, and posts that go live without review.

What it does:
- Connect social channels (OAuth) and group them by brand or client.
- Draft, schedule, and cross-post from a web composer with calendar and kanban review.
- Let agents schedule via @openquok/auto-cli, MCP tools, or the public API (@openquok/node-sdk).
- Media library, photo editor, analytics, recurring posts, templates, and workspace isolation (channels, tokens, and MCP endpoints per workspace).
- Self-host with Docker Compose as an alternative to closed SaaS schedulers.

Architecture:
- Web — SvelteKit 5 app (marketing, docs, authenticated workspace).
- Backend API — Express REST for the dashboard and /api/v1/public/* (API-key clients).
- Orchestrator — BullMQ workers on Redis for scheduled publish, OAuth token refresh, and notification email.
- Clients — CLI, Node SDK, and MCP against the same public API. CLI login uses a device-flow auth server.