GenSX is a simple TypeScript framework for building complex LLM applications. It's a workflow engine designed for building agents, chatbots, and long-running workflows.
- 🎯 Pure Functions: Components are pure TypeScript functions that are easily testable, reusable, and sharable
- 🌴 Natural Composition: Building workflows is as simple as composing functions together
- 🔒 Type-safe: Full TypeScript support with no DSLs or special syntax - just standard language features
- 🚀 Built for Scale: Start simple and evolve to complex patterns like agents and reflection without changing your programming model
- 📊 Automatic Tracing: Real-time tracing of all component inputs/outputs, tool calls, and LLM calls making debugging and observability easy
- ☁️ One-Click Deployment: Deploy workflows as REST APIs with a single command, optimized for long-running LLM workloads up to 60 minutes
- 💾 Built-in Storage: Zero-config blob storage, SQL databases, and vector search for building stateful agents and workflows
Check out the documentation to learn more about building LLM applications with GenSX.
Most LLM frameworks are graph oriented--you express your workflow with nodes, edges, and a global state object. GenSX takes a different approach--you compose your workflow with components, and GenSX handles the execution for you.
Components in GenSX look a lot like functions. You create them by passing in a function and a name to gensx.Component(), a higher order function::
import * as gensx from "@gensx/core";
import { openai } from "@ai-sdk/openai";
import { generateText } from "@gensx/vercel-ai";
// input interface
interface WriteDraftInput {
research: string[];
prompt: string;
}
// components are pure functions that are reusable by default
const WriteDraft = gensx.Component(
"WriteDraft",
async ({ prompt, research }: WriteDraftInput) => {
const systemMessage = `You're an expert technical writer.
Use the information when responding to users: ${research}`;
const result = await generateText({
messages: [
{
role: "system",
content: systemMessage,
},
{
role: "user",
content: `Write a blog post about ${prompt}`,
},
],
model: openai("gpt-4.1-mini"),
});
return result.text;
},
);Components can be composed together to create more complex agents and workflows:
import * as gensx from "@gensx/core";
import { OpenAIProvider } from "gensx/openai";
import { Research, WriteDraft, EditDraft } from "./writeBlog";
interface WriteBlogInput {
title: string;
description: string;
}
const WriteBlog = gensx.Workflow(
"WriteBlog",
async ({ title, description }: WriteBlogInput) => {
const queries = await GenerateQueries({
title,
description,
});
const research = await ResearchBlog({ queries });
const draft = await WriteDraft({ title, context: research });
const final = await EditDraft({ title, content: draft });
return final;
},
);
const result = await WriteBlog({
title: "How AI broke modern infra",
description: "Long-running workflows require a new approach to infra",
});Check out the Quickstart Guide to build your first workflow in just a few minutes.
This repo contains a number of examples to help you get up and running with GenSX.
To run an example:
cd examples/<example-name>
pnpm install
pnpm start| Example | Description |
|---|---|
| 🔄 Reflection | Shows how to use a self-reflection pattern with GenSX |
| 🦾 Anthropic Examples | Examples showing how to use @gensx/anthropic |
| 🧠 OpenAI Examples | Examples showing how to use @gensx/openai |
| 🌊 Vercel AI SDK Examples | Examples showing how to use @gensx/vercel-ai |
| Example | Description |
|---|---|
| 🔍 Hacker News Analyzer | Analyzes HN posts and generates summaries and trends using Paul Graham's writing style |
| ✍️ Blog Writer | Generates blogs through an end-to-end workflow including topic research and content creation |
| 🔬 Deep Research | Generates a report from a prompt after researching and summarizing a list of research papers |
| 💻 Computer Use | Demonstrates how to use the OpenAI computer use tool with GenSX |
| 🗄️ Text to SQL | Shows how to use database storage to translate natural language to SQL queries |
| 🔎 RAG | Demonstrates retrieval augmented generation using vector search storage |
| 💬 Chat Memory | Shows how to build a chat application with persistent chat history using blob storage |
This monorepo contains GenSX, its related packages, examples, and documentation. You can find more detailed instructions in CONTRIBUTING.md.
packages/- Published packagesexamples/- Example applications and use caseswebsite/- GenSX website