model-selector_skill

Overview

This skill provides an intelligent routing layer that chooses the best LLM tier for each query based on semantic intent and cost-efficiency. It classifies requests into Elite, Balanced, or Basic tiers and returns a suggested model before the agent makes the main LLM call. The orchestrator is optimized for common providers and adapts routing signals over time using user history.

How this skill works

The orchestrator computes embeddings for incoming queries and compares them to intent profiles to determine complexity and required capability. It then balances predicted quality needs against cost estimates to select Elite, Balanced, or Basic tiers and returns the suggested model identifier. Rolling adjustments refine keyword and intent thresholds from recent usage to improve future routing decisions.

When to use it

• When you need to minimize LLM spend while preserving quality for complex tasks.
• When an agent handles mixed workloads that vary from simple lookups to deep reasoning.
• When you want to centralize model selection logic across multi-provider environments.
• When rapid experimentation with model tiers is required without changing agent code.

Best practices

• Call get_optimal_model (or analyze_and_route) as the first tool before any heavy LLM invocation.
• Log routing decisions and outcomes to continuously tune the rolling adjustment logic.
• Define clear intent examples for your domain to improve semantic classification accuracy.
• Set cost and latency weightings according to your product priorities (budget vs. response quality).
• Validate routing for edge cases by running A/B checks comparing routed versus always-Elite responses.

Example use cases

• Route long-form technical architecture design prompts to Elite while sending simple formatting tasks to Basic.
• Optimize a customer support agent that can escalate complex issue summaries to Balanced or Elite automatically.
• Lower costs for high-volume telemetry parsing by routing pattern-matching and extraction to Basic.
• Support multi-provider setups by mapping tiers to preferred vendor models and failing over if a provider is unavailable.