What problem does it solve? Deploying and operating large language models on a Rainbond platform involves model discovery, ModelScope downloads, GPU/CPU capacity planning, parameter tuning, and failure diagnostics, all of which are error-prone when done manually. ## Core Features & Use Cases - Model Discovery & Download: Search the built-in catalog or the official ModelScope OpenAPI, then download models with verified completion states and idempotent retry handling. - Instance Lifecycle Management: Create, start, stop, and delete AI Engine instances with startup-safety parameter planning, GPU provider checks (NVIDIA Device Plugin, HAMi shared/whole), and CPU-first deployment paths. - Diagnostics & Tuning: Diagnose Creating/Failed deployments from events and logs, and tune throughput, latency, KV cache, and concurrency using evidence-based parameter decisions. - Use Case: Ask the agent to deploy Qwen3-8B for a knowledge-base Q&A service; it resolves the workspace, checks GPU/CPU capacity, plans safe context and concurrency parameters, creates the instance, and verifies health and model registration. ## Quick Start Use the rainbond-ai-assistant skill to deploy the Qwen/Qwen3-8B model on CPU in my workspace and verify the instance is running and registered.
