oc-smart-agent-hub_skill

Overview

This skill is a multi-provider agent model assignment system that routes AI tasks to the best available model across cloud vendors and local deployments. It supports automatic discovery, zero-code YAML configuration, cost-aware routing, and failover to fallback models. The goal is to simplify using many model providers while optimizing performance and cost.

How this skill works

The system scans configured cloud vendors (OpenAI, Anthropic, Alibaba Cloud, Baidu, Zhipu AI, etc.) and local model services (Ollama, LM Studio, vLLM) to build an inventory of available models. Policy-driven YAML rules determine routing by task type, latency, cost, and provider health. At runtime the hub selects the optimal model, performs failover on errors, and can be reconfigured without code changes.

When to use it

• You're operating across multiple model providers and need centralized routing
• You run a mix of cloud-hosted and local models and want unified selection
• You need automated failover and high availability for model inference
• You want to optimize cost by selecting more economical providers per task
• You prefer declarative configuration (YAML) without changing code

Best practices

• Maintain a concise models.yaml with clear tags for task capabilities, cost, and latency
• Run regular scans to keep the inventory up to date for local services
• Define fallback chains per task type to guarantee continuity on provider outages
• Use cost and latency weights in routing policies to balance budget and performance
• Version control your YAML configs and test changes in staging before production

Example use cases

• Route chat and retrieval tasks to high-throughput cloud models while using local models for private or low-latency needs
• Automatically shift expensive large models to cheaper alternatives for batch processing to control costs
• Detect a provider outage and switch to a local vLLM instance for uninterrupted service
• Combine Anthropic or OpenAI for instruction-following while using Ollama for custom fine-tuned models
• Scan the network to discover newly started local model services and immediately add them to the routing pool