Engineering insights and product updates from the LunaOS team.
Today we're launching LunaOS on Product Hunt. 140+ commands that compose into pipelines — code, deploy, design, create music, generate videos, publish everywhere. Powered by 33 MCP servers. Run them from the CLI, dashboard, or API. Start free with unlimited commands, upgrade to Pro for managed keys and team features.
LunaOS runs entirely on Cloudflare Workers, D1, KV, and Vectorize. No traditional servers. In this post we explain our architecture: how we handle SSE streaming from Workers, manage RAG with Vectorize embeddings, and keep cold starts under 5ms. We also cover our 200-line-per-file rule and how it keeps the codebase maintainable.
Most AI tools lack context about your specific codebase. LunaOS solves this with built-in RAG: connect your GitHub repo, and every agent automatically receives relevant code context before generating results. We use Cloudflare Vectorize with BGE embeddings for fast, accurate semantic search across your entire repository.
A single agent is useful. A chain of agents is powerful. In this tutorial, we walk through creating a "Full Review" chain that runs code-review, security-audit, and test-generator in sequence — passing context between each step. You can build custom chains in the Studio IDE or define them in YAML.