Hermes + DeepSeek 4 + MiniMax 2.7: Multi-Model Coding on a ZimaBoard

Can two LLMs actually coordinate to build real software with minimal human involvement? In this experiment, I set up Hermes on a ZimaBoard 2 to orchestrate DeepSeek 4 and MiniMax 2.7 — and then assigned them two very different coding tasks to see how far they could go on their own. This is not a tutorial on "how to build software with AI." It's an honest look at what happens when you let multiple models coordinate with each other and try to stay out of their way. 🛠 Setup used in this video: ZimaBoard 2  (https://bit.ly/4cWj6Tj) Hermes agent (https://cutt.ly/mtMfYM3x) MiniMax 2.7 (https://cutt.ly/ZtMfUoMq) DeepSeek 4 (https://cutt.ly/btMfOqNe) ------------ 💡 Special Zimaboard 2 discount for my channel's community! 🎁 Discount Coupon: ZeroToMVP15 ($15 off) 🛒 Get it here: https://bit.ly/4cWj6Tj ------------- ⏱ Chapters: 00:00 Intro 00:42 Why I run agents on a ZimaBoard, not my laptop 01:27 Setup: Hermes, DeepSeek 4, MiniMax 2.7 via Opencode 04:22 Task 1: PDF + HTML export for the Markdown editor 05:42 Testing the export feature 07:32 Task 2: building an RSS aggregator in Nim 10:25 Reviewing the generated Nim project 12:00 A 700 KB binary — why I love Nim 12:20 Final thoughts 💬 Let me know in the comments: should I publish the source code of the Nim RSS aggregator? And what would you want two coordinated LLMs to build next? 👍 If you enjoyed this experiment, leave a like and subscribe — more local AI, self-hosted agents, and single-board computing experiments coming soon. #LocalAI #Hermes #DeepSeek #MiniMax #ZimaBoard #Nim