galileo-typescript-sdk

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Galileo TypeScript SDK

The Galileo TypeScript SDK (

@rungalileo/galileo

) provides evaluation and observability workflows for GenAI applications in Node.js and TypeScript. It supports logging LLM calls, retriever operations, tool invocations, and multi-step workflows with built-in scoring.

Additional references:

Framework Integrations — Vercel AI SDK, Mastra, LangGraph (JS), and more
Guardrail Metrics Reference — Scoring metrics available for evaluation workflows
Advanced Evaluation Patterns — Complex workflow evaluation and experiment design

Galileo TypeScript SDK（

@rungalileo/galileo

）为Node.js和TypeScript环境中的GenAI应用提供评估和可观测性工作流能力。它支持记录LLM调用、检索器操作、工具调用以及内置评分的多步骤工作流。

额外参考：

框架集成 — Vercel AI SDK、Mastra、LangGraph (JS) 等
防护规则指标参考 — 评估工作流可用的评分指标
高级评估模式 — 复杂工作流评估和实验设计

Installation

安装

bash

npm install @rungalileo/galileo

Or with yarn/pnpm:

bash

yarn add @rungalileo/galileo
pnpm add @rungalileo/galileo

bash

npm install @rungalileo/galileo

或者使用yarn/pnpm安装：

bash

yarn add @rungalileo/galileo
pnpm add @rungalileo/galileo

Quick Start

快速开始

typescript

import { GalileoObserveWorkflow } from "@rungalileo/galileo";

const workflow = new GalileoObserveWorkflow("my-observe-project");
await workflow.init();

workflow.addWorkflow({ input: "What is quantum computing?" });

workflow.addLlmStep({
  input: "What is quantum computing?",
  output: "Quantum computing uses quantum bits...",
  durationNs: 1500000000,
  model: "gpt-4o",
});

workflow.concludeWorkflow("Quantum computing uses quantum bits...");

await workflow.uploadWorkflows();

typescript

import { GalileoObserveWorkflow } from "@rungalileo/galileo";

const workflow = new GalileoObserveWorkflow("my-observe-project");
await workflow.init();

workflow.addWorkflow({ input: "What is quantum computing?" });

workflow.addLlmStep({
  input: "What is quantum computing?",
  output: "Quantum computing uses quantum bits...",
  durationNs: 1500000000,
  model: "gpt-4o",
});

workflow.concludeWorkflow("Quantum computing uses quantum bits...");

await workflow.uploadWorkflows();

Authentication

认证

Set the following environment variables in your

.env

file or shell:

bash

GALILEO_API_KEY="your-api-key"            # Required — from Galileo console
GALILEO_CONSOLE_URL="https://app.galileo.ai"  # Console URL (or self-hosted)

Alternative authentication via username/password:

bash

GALILEO_USERNAME="your-username"
GALILEO_PASSWORD="your-password"

在你的

.env

文件或shell中设置以下环境变量：

bash

GALILEO_API_KEY="your-api-key"            # 必填 — 从Galileo控制台获取
GALILEO_CONSOLE_URL="https://app.galileo.ai"  # 控制台地址（或自托管地址）

另一种用户名/密码认证方式：

bash

GALILEO_USERNAME="your-username"
GALILEO_PASSWORD="your-password"

Observability with GalileoObserveWorkflow

使用GalileoObserveWorkflow实现可观测性

Basic Workflow Logging

基础工作流日志记录

typescript

import { GalileoObserveWorkflow } from "@rungalileo/galileo";

const workflow = new GalileoObserveWorkflow("my-project");
await workflow.init();

workflow.addWorkflow({ input: "User question here" });

workflow.addLlmStep({
  input: "User question here",
  output: "LLM response text",
  durationNs: 1200000000,
  model: "gpt-4o",
});

workflow.concludeWorkflow("LLM response text");

await workflow.uploadWorkflows();

typescript

import { GalileoObserveWorkflow } from "@rungalileo/galileo";

const workflow = new GalileoObserveWorkflow("my-project");
await workflow.init();

workflow.addWorkflow({ input: "User question here" });

workflow.addLlmStep({
  input: "User question here",
  output: "LLM response text",
  durationNs: 1200000000,
  model: "gpt-4o",
});

workflow.concludeWorkflow("LLM response text");

await workflow.uploadWorkflows();

Logging Retriever Steps

记录检索器步骤

typescript

workflow.addWorkflow({ input: "What are the benefits of RAG?" });

workflow.addRetrieverStep({
  input: "What are the benefits of RAG?",
  output: ["Document 1 content", "Document 2 content"],
});

workflow.addLlmStep({
  input: "Based on the context, explain RAG benefits.",
  output: "RAG provides improved accuracy by...",
  durationNs: 2000000000,
  model: "gpt-4o",
});

workflow.concludeWorkflow("RAG provides improved accuracy by...");

typescript

workflow.addWorkflow({ input: "What are the benefits of RAG?" });

workflow.addRetrieverStep({
  input: "What are the benefits of RAG?",
  output: ["Document 1 content", "Document 2 content"],
});

workflow.addLlmStep({
  input: "Based on the context, explain RAG benefits.",
  output: "RAG provides improved accuracy by...",
  durationNs: 2000000000,
  model: "gpt-4o",
});

workflow.concludeWorkflow("RAG provides improved accuracy by...");

Logging Tool Steps

记录工具步骤

typescript

workflow.addWorkflow({ input: "Calculate 15 * 42" });

workflow.addToolStep({
  input: "15 * 42",
  output: "630",
  durationNs: 50000000,
});

workflow.addLlmStep({
  input: "The calculator returned 630. Respond to the user.",
  output: "15 multiplied by 42 equals 630.",
  durationNs: 800000000,
  model: "gpt-4o",
});

workflow.concludeWorkflow("15 multiplied by 42 equals 630.");

typescript

workflow.addWorkflow({ input: "Calculate 15 * 42" });

workflow.addToolStep({
  input: "15 * 42",
  output: "630",
  durationNs: 50000000,
});

workflow.addLlmStep({
  input: "The calculator returned 630. Respond to the user.",
  output: "15 multiplied by 42 equals 630.",
  durationNs: 800000000,
  model: "gpt-4o",
});

workflow.concludeWorkflow("15 multiplied by 42 equals 630.");

Evaluation with GalileoEvaluateWorkflow

使用GalileoEvaluateWorkflow实现评估

Running an Evaluation

运行评估

typescript

import { GalileoEvaluateWorkflow } from "@rungalileo/galileo";

const evaluateWorkflow = new GalileoEvaluateWorkflow("eval-project");
await evaluateWorkflow.init();

const testCases = [
  { input: "What is ML?", expected: "Machine learning is..." },
  { input: "Explain AI", expected: "Artificial intelligence is..." },
];

for (const testCase of testCases) {
  evaluateWorkflow.addWorkflow({ input: testCase.input });

  const response = await callYourLLM(testCase.input);

  evaluateWorkflow.addLlmStep({
    input: testCase.input,
    output: response,
    durationNs: 1000000000,
    model: "gpt-4o",
  });

  evaluateWorkflow.concludeWorkflow(response);
}

await evaluateWorkflow.uploadWorkflows({
  context_adherence: true,
  completeness: true,
  toxicity: true,
});

typescript

import { GalileoEvaluateWorkflow } from "@rungalileo/galileo";

const evaluateWorkflow = new GalileoEvaluateWorkflow("eval-project");
await evaluateWorkflow.init();

const testCases = [
  { input: "What is ML?", expected: "Machine learning is..." },
  { input: "Explain AI", expected: "Artificial intelligence is..." },
];

for (const testCase of testCases) {
  evaluateWorkflow.addWorkflow({ input: testCase.input });

  const response = await callYourLLM(testCase.input);

  evaluateWorkflow.addLlmStep({
    input: testCase.input,
    output: response,
    durationNs: 1000000000,
    model: "gpt-4o",
  });

  evaluateWorkflow.concludeWorkflow(response);
}

await evaluateWorkflow.uploadWorkflows({
  context_adherence: true,
  completeness: true,
  toxicity: true,
});

Evaluation with RAG Steps

含RAG步骤的评估

typescript

const evaluateWorkflow = new GalileoEvaluateWorkflow("rag-eval");
await evaluateWorkflow.init();

evaluateWorkflow.addWorkflow({ input: "How does photosynthesis work?" });

evaluateWorkflow.addRetrieverStep({
  input: "How does photosynthesis work?",
  output: ["Photosynthesis is the process by which plants..."],
});

evaluateWorkflow.addLlmStep({
  input: "Using the context, explain photosynthesis.",
  output: "Photosynthesis is a process used by plants...",
  durationNs: 1500000000,
  model: "gpt-4o",
});

evaluateWorkflow.concludeWorkflow("Photosynthesis is a process used by plants...");

await evaluateWorkflow.uploadWorkflows({
  context_adherence: true,
  chunk_attribution: true,
});

typescript

const evaluateWorkflow = new GalileoEvaluateWorkflow("rag-eval");
await evaluateWorkflow.init();

evaluateWorkflow.addWorkflow({ input: "How does photosynthesis work?" });

evaluateWorkflow.addRetrieverStep({
  input: "How does photosynthesis work?",
  output: ["Photosynthesis is the process by which plants..."],
});

evaluateWorkflow.addLlmStep({
  input: "Using the context, explain photosynthesis.",
  output: "Photosynthesis is a process used by plants...",
  durationNs: 1500000000,
  model: "gpt-4o",
});

evaluateWorkflow.concludeWorkflow("Photosynthesis is a process used by plants...");

await evaluateWorkflow.uploadWorkflows({
  context_adherence: true,
  chunk_attribution: true,
});

Common Patterns

常见模式

Multiple Workflows in a Single Upload

单次上传多个工作流

typescript

const workflow = new GalileoObserveWorkflow("batch-project");
await workflow.init();

const queries = ["Question 1", "Question 2", "Question 3"];

for (const query of queries) {
  workflow.addWorkflow({ input: query });

  const response = await callYourLLM(query);

  workflow.addLlmStep({
    input: query,
    output: response,
    durationNs: 1000000000,
    model: "gpt-4o",
  });

  workflow.concludeWorkflow(response);
}

await workflow.uploadWorkflows();

typescript

const workflow = new GalileoObserveWorkflow("batch-project");
await workflow.init();

const queries = ["Question 1", "Question 2", "Question 3"];

for (const query of queries) {
  workflow.addWorkflow({ input: query });

  const response = await callYourLLM(query);

  workflow.addLlmStep({
    input: query,
    output: response,
    durationNs: 1000000000,
    model: "gpt-4o",
  });

  workflow.concludeWorkflow(response);
}

await workflow.uploadWorkflows();

Nested Agent Workflows

嵌套Agent工作流

typescript

const workflow = new GalileoObserveWorkflow("agent-project");
await workflow.init();

workflow.addWorkflow({ input: "Research and summarize quantum computing" });

workflow.addToolStep({
  input: "search: quantum computing overview",
  output: "Search results...",
  durationNs: 200000000,
});

workflow.addRetrieverStep({
  input: "quantum computing",
  output: ["Doc1: Quantum bits...", "Doc2: Superposition..."],
});

workflow.addLlmStep({
  input: "Summarize the following research on quantum computing...",
  output: "Quantum computing leverages quantum mechanical phenomena...",
  durationNs: 2500000000,
  model: "gpt-4o",
});

workflow.concludeWorkflow(
  "Quantum computing leverages quantum mechanical phenomena..."
);

await workflow.uploadWorkflows();

typescript

const workflow = new GalileoObserveWorkflow("agent-project");
await workflow.init();

workflow.addWorkflow({ input: "Research and summarize quantum computing" });

workflow.addToolStep({
  input: "search: quantum computing overview",
  output: "Search results...",
  durationNs: 200000000,
});

workflow.addRetrieverStep({
  input: "quantum computing",
  output: ["Doc1: Quantum bits...", "Doc2: Superposition..."],
});

workflow.addLlmStep({
  input: "Summarize the following research on quantum computing...",
  output: "Quantum computing leverages quantum mechanical phenomena...",
  durationNs: 2500000000,
  model: "gpt-4o",
});

workflow.concludeWorkflow(
  "Quantum computing leverages quantum mechanical phenomena..."
);

await workflow.uploadWorkflows();

Best Practices

最佳实践

Always call
init()
before adding workflows — it authenticates and sets up the project.
Always call
concludeWorkflow()
with the final output before starting the next workflow or uploading.
Always call
uploadWorkflows()
at the end to send data to Galileo.
Use accurate
durationNs
values — measure actual LLM call duration in nanoseconds for meaningful latency tracking.
Set environment variables in
```
.env
```
files rather than hardcoding API keys.
Use
GalileoEvaluateWorkflow
for test/eval runs and GalileoObserveWorkflow
for production monitoring.
Pass scorer configuration to
```
uploadWorkflows()
```
in evaluate mode to get metric scores computed.

始终先调用
init()
再添加工作流 — 它会完成认证并完成项目初始化。
始终调用
concludeWorkflow()
传入最终输出，再开始下一个工作流或执行上传操作。
始终在最后调用
uploadWorkflows()
将数据发送到Galileo。
使用准确的
durationNs
值 — 实际测量LLM调用的纳秒级时长，以获得有意义的延迟追踪数据。
在
.env
文件中设置环境变量而非硬编码API密钥。
测试/评估运行使用
GalileoEvaluateWorkflow
，生产环境监控使用**
```
GalileoObserveWorkflow
```
**。
在评估模式下向
uploadWorkflows()
传入评分器配置以计算得到指标分数。

Resources

资源

Documentation: https://docs.galileo.ai
TypeScript SDK repo: https://github.com/rungalileo/galileo-js
SDK examples: https://github.com/rungalileo/sdk-examples
npm: https://www.npmjs.com/package/@rungalileo/galileo
Galileo console: https://app.galileo.ai

文档： https://docs.galileo.ai
TypeScript SDK仓库： https://github.com/rungalileo/galileo-js
SDK示例： https://github.com/rungalileo/sdk-examples
npm地址： https://www.npmjs.com/package/@rungalileo/galileo
Galileo控制台： https://app.galileo.ai