Kreatar
Kreatar (by Neurl) is a multi-tenant content operating system for creators — draft, version, narrate, adapt, and publish from a single workspace. It exposes workspace-scoped MCP tooling so external AI agents can act on a creator's content with human oversight built in.
Kreatar is a content operating system for creators, built by Neurl, where I'm a partner and engineering lead. A creator works in a single workspace to draft, version, narrate, and adapt a piece of content, then publish it across channels — with AI assistance and agent tooling wired in from the ground up.
What it does
- Workspace-scoped multi-tenancy. Content, members, API keys, and audit logs are all isolated per workspace, with rate limiting keyed on IP, account, and subject.
- Audio narration pipeline. End-to-end text-to-speech via Deepgram with caching, progress reporting, and a clean draft/published separation, served through an embedded player.
- Adaptations & suggestions. One source piece is adapted into newsletter, LinkedIn, and X formats, with an LLM-backed suggestion queue.
- MCP tooling. Workspace-scoped API keys and MCP contract routes let external AI agents operate on a creator's content — under human oversight, not unattended.
How it's built
A pnpm monorepo: an Astro + React landing and CMS over a Hono API on Node.js, backed by MongoDB, with RabbitMQ workers for audio and import jobs and MinIO for object storage. Anthropic and OpenAI SDKs power the AI workflows. Production runs on Docker Compose behind Caddy, with a Playwright smoke suite, integration tests for the content lifecycle and MCP keys, and documented backup/restore runbooks.