{"slug": "show-hn-purple-flow-ai-first-n8n-alternative-10k-workflows-on-512mb", "title": "Show HN: Purple Flow – AI-first n8n alternative, 10k+ workflows on 512MB", "summary": "A developer known as leetheguy released Purple Flow, an open-source Elixir/Beam workflow engine positioned as an AI-first alternative to n8n, claiming 10,000 concurrent workflows executed in 3 seconds using about 330MB of RAM (listed as 512MB in the case study), versus n8n needing 2GB to run 100 equivalent workflows. Purple Flow defines workflows in pure TOML rather than a visual node editor, keeps credentials hidden from the agent, and has been used in production for almost 2 months, including a live demo built in an afternoon that lets an agent query Adventure Works data and generate reports, documents and slides. The author says the project was built to address n8n's scaling cost, limited streaming and confusing MCP support while building agentic swarms.", "body_md": "Hey all! I ran into a wall with n8n - hard and expensive to scale, limited streaming, heavy MCP is confusing for agents.\n\nThat scaling was becoming more and more critical as I was beginning to build agentic swarms. So I built my own alternative with Elixir and Beam. Been using it in production for almost 2 months.\n\nIt connects my agent to multiple databases, sandboxes, and APIs. I've even migrated over my most used n8n workflows without a hiccup.\n\nI ran some benchmarks and 10,000 concurrent workflows executed in 3 seconds and cost about 330MB of ram (listed as 512 in the case study). Equivalent workflows in n8n took 2GB of ram to run 100. It's roughly 200x the memory efficiency of the usual Node/Python stacks. And while you don't edit the workflows yourself visually, you can see and manage them through a UI, along with credentials that the agent never sees.\n\nThe workflows themselves are pure TOML. Not as easy as dragging node connectors. But the idea is that AI creates the flows and you can easily understand them.\n\nThis isn't a shaky demo. I'm using it as my daily driver to create complex multi-agent personas, run research and info gathering swarms, and whatever else my agent needs to get its job done. (it's built more than 70 so far)\n\nI have a live working demo on my website that was built by an agent using Purple Flow to connect to my server. The demo itself lets an agent using Purple Flow access live (Adventure Works) data and generate reports, documents, and slides on demand. My agent and I built it in an afternoon. It runs on a lightweight model (to save money for a public demo) with very few hiccups.\n\n- Repo: [https://github.com/leetheguy/purple_flow](https://github.com/leetheguy/purple_flow)\n- Discord, benchmarks, and demo links in the repo.\n\nComments URL: [https://news.ycombinator.com/item?id=50560170](https://news.ycombinator.com/item?id=50560170)\n\nPoints: 1\n\n# Comments: 1", "url": "https://wpnews.pro/news/show-hn-purple-flow-ai-first-n8n-alternative-10k-workflows-on-512mb", "canonical_source": "https://github.com/leetheguy/purple_flow", "published_at": "2026-11-19 13:57:14+00:00", "updated_at": "2026-11-19 14:16:30.156574+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "agent-protocols"], "entities": ["Purple Flow", "n8n", "Elixir", "Beam", "leetheguy", "Adventure Works", "GitHub"], "also_reported_by": [], "takeaway": "Purple Flow runs 10,000 concurrent workflows in 3 seconds at roughly 330MB of RAM by defining flows as TOML on Elixir/Beam instead of Node-based visual nodes, versus 2GB for 100 equivalent n8n workflows.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/show-hn-purple-flow-ai-first-n8n-alternative-10k-workflows-on-512mb", "markdown": "https://wpnews.pro/news/show-hn-purple-flow-ai-first-n8n-alternative-10k-workflows-on-512mb.md", "text": "https://wpnews.pro/news/show-hn-purple-flow-ai-first-n8n-alternative-10k-workflows-on-512mb.txt", "jsonld": "https://wpnews.pro/news/show-hn-purple-flow-ai-first-n8n-alternative-10k-workflows-on-512mb.jsonld"}}