Documentation

Everything you need to understand how ModelGenius validates a model switch and how to capture the traffic snapshots a validation runs on. This section will grow — here is what's available today.

Doc

Validation methodology — a primer

How we decide whether a model switch is safe: replay-based validation, success criteria defined up front, and statistical confidence you can inspect.

SDK · GitHub

Logger setup — snapshot capture SDK

Open-source Python SDK that captures LLM call/response snapshots from your app (OpenAI, Anthropic, or LiteLLM) and validates them locally — no network calls, kill switch included. Install with pip install modelgenius-sdk.

Reference

Snapshot format specification

The modelgenius.snapshot/v1 JSONL format: required and optional fields, provider request shapes, and how snapshots group into workflows.