evalguard

Official Node.js/TypeScript SDK for the EvalGuard API -- evaluate, red-team, and guard LLM applications programmatically.

Installation

npm install evalguard

Quick Start

import { EvalGuard } from "evalguard";

const client = new EvalGuard({ apiKey: "eg_live_..." });

// Run an evaluation
const evalResult = await client.eval({
  name: "qa-check",
  projectId: "my-project",
  model: "gpt-4o",
  prompt: "Answer: {{input}}",
  cases: [
    { input: "What is 2+2?", expectedOutput: "4" },
  ],
  scorers: ["exact-match", "contains"],
});
console.log(`Eval ID: ${evalResult.id}`);

// Run a security scan
const scan = await client.securityScan({
  projectId: "my-project",
  model: "gpt-4o",
  prompt: "You are a helpful assistant.",
  attackTypes: ["prompt-injection", "jailbreak", "data-extraction"],
});
console.log(`Scan ID: ${scan.id}`);

// Get eval results
const run = await client.getEvalRun(evalResult.id);
console.log(`Status: ${run.status}, Score: ${run.score}`);

// Send trace data
await client.trace({
  projectId: "my-project",
  sessionId: "session-123",
  steps: [
    { type: "llm", input: "Hello", output: "Hi there!", duration: 450 },
  ],
});

Configuration

const client = new EvalGuard({
  apiKey: "eg_live_...",
  baseUrl: "https://your-self-hosted-instance.com/api/v1", // optional
});

Methods

| Method | Description | |---|---| | client.eval(params) | Run an evaluation with scorers and test cases | | client.getEvalRun(id) | Fetch results of a specific eval run | | client.securityScan(params) | Run a red-team security scan against a model | | client.trace(params) | Send agent/LLM trace data for monitoring |

TypeScript

The SDK exports all types from @evalguard/core for full type safety:

import type { EvalGuardConfig } from "evalguard";

Documentation

Full documentation at evalguard.ai/docs/sdk.

License

MIT -- see LICENSE for details.