Skip the graph: LLM workflows as React-style components

What it does GenSX is a TypeScript framework for building LLM applications—agents, chatbots, and long-running workflows—by composing pure functions it calls components. You write standard async TypeScript functions and wrap them with gensx.Component() or gensx.Workflow(), and the framework handles execution, automatic tracing of all inputs and LLM calls, and one-click deployment as REST APIs. It also bundles zero-config blob storage, SQL databases, and vector search for stateful workflows.

The interesting bit Instead of the usual graph-oriented abstractions—nodes, edges, and global state bags—GenSX borrows React’s composition model for the backend. You nest and reuse async components like UI elements, and the framework implicitly builds the execution graph for you. It is a neat trick: the mental model stays familiar to frontend engineers, but everything runs as an optimized, long-running backend workflow.

Key highlights

• Full TypeScript support with no DSL or special syntax; just ordinary language features
• Automatic tracing of component I/O, tool calls, and LLM invocations for built-in observability
• One-command deployment to REST APIs optimized for jobs up to 60 minutes
• Built-in storage primitives: blob storage, SQL, and vector search
• Integrations for OpenAI, Anthropic, and the Vercel AI SDK

Verdict TypeScript developers who find graph-based agent frameworks tedious should try this; if your stack is Python-first or you need heavy data-science tooling, it probably will not fit.