which-llm_skill

Overview

This skill provides deterministic decision-ranking to recommend a single LLM model and an ordered fallback plan based on a natural-language goal and clear constraints like budget and minimum quality. It integrates an autonomous USDC payment flow and an outcome-reporting mechanism that issues credit tokens (discounts) for future calls. Use it to get repeatable, auditable model choices while controlling spending with a dedicated wallet.

How this skill works

Send a decision request (goal + constraints) to the optimize endpoint and receive a recommended_model plus a fallback_plan. When the API returns a 402 payment-required response, the skill pays autonomously from your dedicated wallet and retries the request. After you execute the chosen model, report the outcome to the outcome endpoint to receive an X-Credit-Token that reduces future costs.

When to use it

• You need a single, ranked model recommendation with a clear fallback plan
• You want to optimize model selection under strict budget or quality constraints
• You prefer deterministic, auditable decisions for production workflows
• You can provide a dedicated wallet for low-cost autonomous payments
• You want to earn discounts by reporting real-world outcomes

Best practices

• Create and use a dedicated EVM wallet with limited USDC and native gas tokens—never reuse your main wallet
• Verify the receiver payment address from at least two independent sources before enabling autonomous payments
• Start with minimal funding ($1–$5 USDC and $1–$5 gas) to validate operation before increasing balance
• Rotate the dedicated wallet periodically and store credentials in a secure secrets manager
• Monitor all transactions on a block explorer and report discrepancies immediately

Example use cases

• Select the cheapest LLM that meets a minimum accuracy threshold for classification tasks
• Produce a ranked fallback plan for production inference if the primary model is unavailable
• Optimize model choice for specific tasks like summarization, extraction, or coding under a fixed budget
• Automate low-cost decision requests for high-frequency experiments while earning credits by reporting results
• Integrate into CI pipelines to choose models deterministically for test workloads