AI SDK & Agent Framework for React: what 8 makers' products chose
Among 8 makers' products built with React that we know the AI SDK & agent framework of, LangChain leads: 4 of them (50%). Next come Mastra (3) and AI SDK (1).
Broad framework or thin SDK. Broad frameworks come with integrations for loaders, vector stores and memory, but add layers of abstraction you have to learn and debug. Thin SDKs give you a typed agent loop close to the raw API and leave the rest to you. An agent harness goes further the other way: the whole loop — file access, shell, context management — comes ready-made, for agents that work inside a workspace. A visual builder sits outside your code entirely: you assemble the agent, its tools and its retrieval on a canvas and call the result over an API, trading control and version history in your repo for speed and edits non-developers can make.
What makers on React chose
| Option | React products | All makers |
|---|---|---|
| 4 (50%) | 56% | |
| 3 (38%) | 20% | |
| 1 (13%) | 4% |
Live makers' products with a source for both: that they're built with React, and what they use for AI SDK & agent framework. A product using two options counts in both; not market share. "All makers" is the option's share among every maker's product that made this decision.
Which to pick, by situation
- Your app is TypeScript and you want agents without adding a Python service
AI SDK 1 here
Mastra 3 here - You need many prebuilt connectors for models, tools and data sources
LangChain 4 here
LangChain
Broad framework4 React products here (50%).
llms.txtBest forThe largest set of ready-made integrations for models, tools and data sources, in Python and JavaScript.
Trade-offLayers of abstraction make debugging harder, and the API has changed often.
“LangSmith has been a lifesaver for testing and debugging our prompt flows. It gave us the visibility we needed to make Reflex AI Builder feel smooth and reliable. Super helpful when working with complex LLM chains.”
- One interface across many model providers, so switching models, prompts and chains is cheap. 3 sources
- A large ecosystem of document loaders, chunkers, retrievers and tool integrations for RAG and agents. 3 sources
- Frequent breaking API changes force rewrites, and coding agents trained on older versions produce messy code for it. 3 sources
- Layers of abstraction make debugging slow, and many developers find calling model APIs or lighter libraries simpler. 4 sources
“Helped us deploy and test agents”
AI SDK & Agent FrameworkMaker says so · source ↗“The ultimate enabler for in-browser, private RAG. This tool is the secret sauce that allowed me to generate high-quality vector embeddings directly in the browser via WebAssembly. It completely eliminated the need to spin up expensive cloud servers or compromise user privacy. If you are building local-first edge AI or hybrid search engines, this is the absolute gold standard you need in your tech stack.”
AI SDK & Agent FrameworkMaker says so · source ↗“LangSmith has been a lifesaver for testing and debugging our prompt flows. It gave us the visibility we needed to make Reflex AI Builder feel smooth and reliable. Super helpful when working with complex LLM chains.”
AI SDK & Agent FrameworkMaker says so · source ↗“The best framework for inferencing LLM models.”
AI SDK & Agent FrameworkMaker says so · source ↗
Mastra
TypeScript framework3 React products here (38%).
llms.txtCLIBest forTypeScript and Next.js apps that want agents, workflows, memory and RAG without a Python service.
Trade-offYounger ecosystem, and some enterprise features sit under a separate source-available license.
- TypeScript-first with type safety, so it fits teams whose product stack is already TypeScript. 4 sources
- Workflows, agents, memory, RAG and evals come in one package, replacing half-built in-house frameworks. 4 sources
- Workflows with rule-based, non-LLM branching feel clunky, pushing some to write their own. 1 source
- The Elastic v2 license bars offering it as a hosted service, so it is not open source in the usual sense. 1 source
Also chosen, and not seen yet
Fewer React products here: AI SDK (1).
Real options for AI SDK & agent framework not seen among these React products yet: LlamaIndex, OpenAI Agents SDK, Pydantic AI, Claude Agent SDK, LangGraph.
Who switches, and to what
Public GitHub pull requests whose title says "X to Y", by developers in general, not only on React.
- LangChain → AI SDK 4 pull requests
- AI SDK → LangChain 3 pull requests
Getting AI SDK & agent framework right
Write your first agent as a plain loop around the model's tool-calling API — it's under a hundred lines and teaches you what a framework would hide. Reach for a framework when you need the pieces it saves you from: retries, streaming, tracing, multi-step workflows. Choose by language first; a TypeScript app should not grow a Python service just for an agent.
- Reaching for a multi-agent setup before one agent with good tools works; more agents mean more tokens and more ways to fail.
- Letting the agent loop run with no step or cost cap, so a confused model calls tools until it hits the provider's rate limit.
- Adopting a broad framework's abstractions for a single model call, then fighting them when you need to see the actual prompt.
The rest of the React stack
What makers' React products chose for their other decisions, the most common pick for each:
- Hosting: Vercel 68 of 147
- Transactional Email: SendGrid 64 of 109
- Payments: Stripe 76 of 87
- Database: PostgreSQL 32 of 73
- Web Analytics: Google Analytics 54 of 66
- LLM API: OpenAI API 23 of 38
- Product Analytics: PostHog 17 of 37
- Lifecycle Email: Loops 10 of 19
Questions
What do most React apps use for AI SDK & agent framework?
Among 8 makers' products built with React that we know the AI SDK & agent framework of, LangChain leads: 4 of them (50%). Next come Mastra (3) and AI SDK (1).
How this is counted: a product is on React when a live-product source says so (the maker, its website, a customer story or a subprocessor list), and the same for its AI SDK & agent framework. How evidence is collected.