Your own ChatGPT-style app for $0 a month: Oracle's free server, Google's free Gemma 4 and Open WebUI, with every command and the 3 switches that make it fast

The companion to the reel. We built it for real on October 2, 2026: Oracle's Always Free server (2 Arm cores, 12 GB), Ollama, Google's Gemma 4 E2B and Open WebUI v0.11.4. First word 1.5 seconds after send, a 3-sentence answer in about 9 seconds, 4.8 GB of memory still free. INSIDE: (1) THE STACK + THE HONEST CATCH — what each free part does, and why a small model is great for drafts and emails but not as smart as ChatGPT's best. (2) STEPS 1–4 — the Oracle instance settings, the Ollama install, the exact Open WebUI docker command we ran (private: it only listens on the server), the SSH tunnel, and phone access through Tailscale. (3) THE 3 SPEED SWITCHES — out of the box our first reply took 157 seconds; turning off Gemma 4 thinking, the background helpers and Builtin Tools brought it to 1.5 seconds. (4) OUR 5-MODEL SPEED TEST on this exact server. (5) KEEP IT SAFE — never open port 3000, Oracle's idle-reclaim rule, backups and five quick fixes. (6) 10 STARTER PROMPTS. Independent guide from Hyperautomation Labs — not affiliated with OpenAI, Oracle, Google, Ollama, Open WebUI or Tailscale.

By submitting, you'll be subscribed to the Hyperautomation AI Report newsletter — three practical AI emails a week. Unsubscribe anytime.

Free. No spam. Unsubscribe anytime.