Descriptions:
Everyone says local AI is free — but is it really? In this video I share my honest experience replacing my Jetson Nano setup with the Beelink SER10 Max, running OpenClaw and Hermes locally with a Qwen 3.5 9B model.
I break down the 7 decisions you need to make before going local, compare local vs cloud speed, and show you exactly how I set up a hybrid system that saves money without sacrificing performance.
00:00 Local Agents Aren’t Free
01:08 Seven Stack Decisions
05:07 Where Agents Run
06:48 Cloud vs Local vs 24/7
07:42 Picking Local Hardware
09:03 Choosing the Right Model
10:49 Find Models with LM Studio
12:36 LLM Fit Speed Estimates
13:42 Runtimes Ollama vs Llama.cpp
14:35 Speed Tokens and Latency
16:57 Privacy and Hybrid Strategy
18:01 Hands On Setup Preview
18:29 Beelink SER10 Max Specs
19:12 Unboxing the Mini PC
21:14 Ports and What’s Included
22:08 OpenClaw Ubuntu Guide
22:41 Linux First Boot Setup
23:06 Local Llama CPP Test
23:51 OpenClaw Onboarding Config
24:53 Web UI and Local Chat
25:40 Why You Need Tailscale
28:04 Install Tailscale on Linux
30:16 Autostart and SSH Gotchas
32:22 Tailscale on Mac SSH
34:58 iPhone SSH via Tailscale
36:20 Install Hermes Agent
37:24 Connect Hermes to Local Model
38:29 Local vs Cloud Speed Test
42:09 Final Verdict and Tradeoffs
🔗 LINKS MENTIONED
🖥️ Beelink SER10 Max (affiliate): https://beelink.sjv.io/VONRrk
🛠️ LM Studio: https://lmstudio.ai
🔒 Tailscale: https://tailscale.com
💻 Warp Terminal: https://app.warp.dev/referral/539GR4
📊 LLM Fit: https://github.com/AlexsJones/llmfit
📱 Termius (iPhone SSH): https://termius.com/index.html
🔀 OpenRouter: https://openrouter.ai
🤝 JOIN THE COMMUNITY
https://substack.com/@rumjahn
#localai #OpenClaw #HermesAgent #BeelinkSER10 #AIAutomation
