OpenClaw's "Unknown Model" Error — How One Missing JSON Entry Broke My AI Assistant for 4 Hours

🧵 I chased a phantom through two config files, three API keys, and 47 SSH sessions. The fix was one line of JSON. 🤖 What's OpenClaw? Before I dive in — if you haven't heard of OpenClaw , it's an open-source AI agent framework that lets you run persistent AI assistants on your own server. Think of it as your self-hosted ChatGPT, but with memory, personality, tools, scheduled tasks, and multi-channel support (Telegram, Discord, WhatsApp, etc.). You configure which LLM models power your agents (GPT-4, Gemini, Claude, or any model via OpenRouter ), and OpenClaw handles the orchestration — routing messages, managing sessions, executing tools, and maintaining long-term memory across conversations. I run my personal AI assistant on an EC2 instance using OpenClaw. It handles my emails, calendar, reminders, and general chat. The model I'd been using for weeks: stepfun/step-3.5-flash:free via OpenRouter — a solid, free, 250K-context model. Until one Saturday morning, when it just… stopped. 🔇 The