Skip to Content
Introduction

LaunchGate Documentation

Automated quality gating for AI applications.

LaunchGate stands between your prompt changes, model swaps, or pipeline updates and your users — running evaluations and blocking deployments when quality drops below defined thresholds.

How it works

  1. Define what good looks like — Create eval suites with test cases and scoring criteria
  2. Run evals on every change — Trigger evaluations from your SDK, CLI, or CI/CD pipeline
  3. Gate deployment on results — Automatically pass or fail based on your quality thresholds
  4. Ship with confidence — Deploy knowing your AI outputs meet your standards

Core loop

Code change → LaunchGate eval → Cleared for launch ✓ → Deploy
                                 Launch aborted ✗    → Fix & retry

Choose your integration

IntegrationBest for
SDKProgrammatic eval runs from your application code
CLIRunning evals from the terminal or shell scripts
GitHub ActionAutomated PR gating in GitHub CI/CD
REST APIDirect HTTP integration with any platform

Quick example

import { LaunchGate } from "@launchgate/sdk";
 
const lg = new LaunchGate({
  apiKey: process.env.LAUNCHGATE_API_KEY,
});
 
const result = await lg.run("rag-faithfulness", {
  input: {
    context: "The Eiffel Tower was built in 1889.",
    query: "When was the Eiffel Tower built?",
  },
  output: "The Eiffel Tower was built in 1889.",
});
 
console.log(result.status);   // "cleared"
console.log(result.passRate); // 1.0

Key concepts

  • Projects — Top-level grouping for your eval suites
  • Eval Suites — A collection of test cases with a pass threshold
  • Eval Cases — Individual checks with inputs, scorers, and thresholds
  • Scorers — Five evaluation functions: exact_match, regex, json_schema, contains, and llm_judge
  • Runs — Each execution of a suite, with immutable results
  • BYOK Keys — Bring your own LLM provider keys for AI-powered scoring
Last updated on
Quick Start