Changelog
SDK 0.2.0
October 12, 2026
Rank a short list by your own criterion. The new experimental rank function orders 2–10 JSON items using the existing model. It returns original item indices, relative scores, comparison count, model identity, and token usage.
For example, give Zils several possible support actions and ask which best addresses the customer's issue. Your application can use the full order or choose the first item under its own policy.
npm install @zils/sdk@^0.2.0Read the ranking guide for runnable TypeScript and Python examples. Noul, Choice, and Score remain available; ranking combines Choice comparisons in the SDK without adding a new API question type or training a new model.
| Surface | Available in this release |
|---|---|
| JavaScript/TypeScript SDK | client.rank in the public npm package |
| Python SDK | Sync and async client.rank; source and release wheel require repository access |
| npm CLI | 0.2.0 bundles the matching Python SDK; no ranking command |
| MCP | Existing decision tools; no ranking tool |
| Playground | Ranking guide link; no interactive ranking demo yet |
Ranking is experimental. Scores are relative strengths, not confidence percentages. Every pair is compared in both orders: three items require six judgments and ten require 90. Existing usage limits and billing apply. This release does not establish better accuracy on your task.
The SDK overview covers installation and credentials. The JavaScript/TypeScript package is available on npm.