{"name":"Kolossus","version":"0.1.0","description":"Multi-Agent LLM Platform with LoRA and Auto-Training","model":"Qwen/Qwen3-8B-AWQ","features":["OpenAI-compatible API","Smart routing: auto-classifies queries (fast/think/RAG)","Thinking mode: step-by-step reasoning for complex tasks","Speculative decoding: 2-3x faster inference","RAG: knowledge base per agent for factual accuracy","Multimodal: text, image, audio, video endpoints","Tool calling: Hermes-format function calling","Multi-agent support with LoRA","Auto-training from feedback"]}