Running OpenClaw on Nebius

Live ClawCamp workshop recording. From zero to a live AI agent on Nebius — Token Factory setup, three deploy paths (npm / Docker / Nebius Serverless), and a hands-on build wiring an agent to Kimi K2.5, Llama 3.3 70B, and DeepSeek-R1.

Watch the recording on Zoom →

What's covered

• Why orchestration (CPU) and inference (GPU) should be separate
• Choosing a model: Kimi K2.5, Llama 3.3 70B, DeepSeek-R1, GLM-5, Qwen3
• Connecting OpenClaw to Token Factory in three minutes (CLI or config file)
• Three deploy methods compared:
  • npm install for local dev
  • Docker for isolated/repeatable
  • Nebius Serverless for prod auto-scaling
• JSON-mode tool calling — no prompt-engineering hacks
• Cost breakdown (light/medium/heavy usage vs. running your own H100)
• Security checklist before going live

Prerequisites

• Laptop with browser, Node.js 18+
• A Nebius account (free trial works)
• Token Factory API key — grab it in the workshop, or ahead of time at console.nebius.com

Follow-up curriculum

• Deploy via Web Console (20 min)
• Deploy with Install Scripts (15 min)
• Deployment Patterns & Troubleshooting (45 min)
• Private Agents for sensitive data (55 min)
• Agent Lifecycle & Fine-Tuning (70 min)
• Robotics with Solo CLI (75 min)

Full curriculum at claw.camp/curriculum.