The gap we kept hitting: research-grade AI capabilities exist, but putting them into a reliable, inspectable, controllable production workflow requires gluing together too many tools. We built Heym to address this. It's a self-hosted, source-available AI workflow automation platform. Visual canvas for building multi-agent pipelines, built-in vector store management for retrieval-augmented workflows, human-in-the-loop review checkpoints, full LLM execution traces, and an MCP Server to expose any workflow as a callable tool for AI assistants. The execution engine builds a DAG from the workflow graph and runs independent nodes concurrently. Agent nodes have automatic context compression so long-running agents don't silently fail as context grows. Everything runs on your own infrastructure via Docker Compose. Source available :) https://github.com/heymrun/heym

Hey everyone, For the past several months we've been building something that came out of a frustration most of you probably know well. We love n8n for what it does. But the moment a workflow needed real agent behavior, document retrieval, approval checkpoints, and full observability in the same place, we kept fighting the platform instead of building. It wasn't designed for that and it's not a criticism, just a different problem. So we built Heym. Self-hosted, source-available, visual canvas for AI-native workflows. Multi-agent orchestration, built-in knowledge retrieval, human review checkpoints, MCP support. Launched today on Product Hunt and the repo just went public. We're two engineers and genuinely curious: for those of you building real AI workflows with agents and orchestration, what's the part that breaks down most often in your current stack? github.com/heymrun/heym

I'm excited to share our second-generation n8n workflow generator model! After releasing the Qwen2.5-Coder-14B model three days ago, we've taken a massive leap forward with the Qwen3-Coder-30B-A3B-n8n-Workflow-Generator - a Mixture of Experts (MoE) architecture that brings both superior quality and incredible speed. Blog: https://n8nbuilder.dev/blog/qwen3-coder-30b-a3b-n8n-workflow-generator-model 💡 What Makes This Special? • 30B total parameters with only ~3.5B active per token • MoE architecture for smarter expert routing • 75-80 tokens/second on Mac M4 (MLX Q4) • Complete workflows generated in ~15 seconds • Better quality than dense models, faster than you'd expect Why did we move from Dense (14B) to MoE (30B)? Simple - MoE uses specialized "expert" networks that only activate when needed. Think of it as having 30B parameters worth of knowledge, but only using 3.5B at a time. This means: ✅ 30B model quality ✅ 3.5B model speed ✅ Best of both worlds 🛠️ Technical Details: • Base: Qwen3-Coder-30B-A3B-Instruct • Fine-tuned with QLoRA on 2,308+ workflow templates • 8192 token context window • Available in Transformers, MLX Q4, and LoRA formats 📊 Performance on Mac M4 Pro (64GB): • Inference: 75-80 tok/s • Complex multi-node workflows • AI agent integrations • Structured data extraction • API workflow generation The model handles everything from simple RSS monitoring to complex AI agent workflows with multiple decision points. 💬 Available Now: • HuggingFace: https://huggingface.co/mbakgun/qwen3-coder-30b-a3b-n8n-workflow-generator Would love to hear what workflows you build with this! The jump from dense to MoE has been a game-changer for us.

Before the year ends, I released my first fine-tuned model built specifically for the needs of the n8n automation community and fellow “flowgrammers”. Using QLoRA, Qwen2.5-Coder-14B was trained on more than 2,500 real-world workflow templates to create a simple, fluent AI that actually speaks the n8n language. Blog: https://n8nbuilder.dev/blog/qwen2-coder-14b-n8n-workflow-generator-model Why fine-tune for n8n/MCP-style automation? Even with great MCP/tool/protocol setups, you often end up with a lot of tool calls, brittle prompts, and constant back-and-forth. Fine-tuning lets you compress “how to use tools + how to format outputs” directly into the model, so you spend fewer tokens on instructions, reduce context pressure, and cut latency by avoiding endless round-trips to a remote LLM. On top of that, n8n has its own quirks like node wiring, expression language, and edge cases that a generic model is not aware of—training the model on these patterns makes it far more likely to produce workflows you can drop into the canvas and run. Try it on Hugging Face: https://huggingface.co/mbakgun/Qwen2.5-Coder-14B-n8n-Workflow-Generator Wishing everyone an amazing and productive New Year in advance 🎉

Hi Folks, I wanted to share my latest project, n8n Builder, built for anyone looking to automate with n8n faster and smarter. - Describe your workflow in natural language – our AI takes care of the build. - Chrome extension that opens with Ctrl+M, letting you generate, visualize, and export workflows to n8n in seconds. - Pay-as-you-go pricing: No subscriptions, pay only for successful workflow generations. Credits never expire. - Zero n8n API key needed! Just describe what you want, review the result, and export directly. - Supports freelancers, teams, and developers – speed up client delivery, automate lead handling, sync with CRMs, and more. - Works with Gmail, Slack, Google Sheets, databases, and more. - Advanced workflows in seconds – see real examples of sales automations, expense processing, user-generated content creation and more. - Priority support and error-free guarantees – only pay for successful runs, get quick help by email if needed. Check it out and let me know what you think! Happy to answer questions or show how you can optimize your n8n workflows with AI. https://n8nbuilder.dev/