AgentsKit vs LangChain.js, Vercel AI SDK & assistant-ui
An honest comparison of AgentsKit with LangChain.js, the Vercel AI SDK, assistant-ui, and Mastra — bundle size, runtime, contracts, lock-in, and when to pick each.
People evaluating AgentsKit almost always ask the same question: how is this different from what I already use? This page answers that directly and honestly. Each tool below is good at something — the goal is to help you pick, not to win an argument.
#TL;DR
| AgentsKit | LangChain.js | Vercel AI SDK | assistant-ui | Mastra | |
|---|---|---|---|---|---|
| Core size | ~10 KB, zero deps | Large, heavy dep tree | Small | UI-only | Medium |
| Chat UI | 7 frameworks, headless | None | React-first | React, 50+ components | None |
| Autonomous runtime | Yes (ReAct, planning, multi-agent) | Yes (LangGraph) | No | No | Yes |
| Provider lock-in | None (contract adapters) | Abstraction-heavy | SDK-coupled | N/A | Framework-coupled |
| Substitutable parts | 6 formal contracts | Implicit | Partial | N/A | Partial |
| License | MIT | MIT | Apache-2.0 | MIT | Elastic/Apache |
#vs LangChain.js
LangChain.js is the most feature-complete option and has the largest ecosystem. The trade-off is weight and leaky abstractions: a real agent pulls in a deep dependency tree, and the abstractions surface at nearly every layer.
AgentsKit keeps the core at ~10 KB with zero dependencies and exposes six explicit contracts (Adapter, Tool, Skill, Memory, Retriever, Runtime). You compose what you need; nothing is implicit.
Pick LangChain.js if you want one library that already has an integration for everything and you accept the bundle cost. Pick AgentsKit if you want small, contracted, swappable parts and plain JavaScript end to end.
#vs Vercel AI SDK
The Vercel AI SDK is an excellent streaming chat SDK. It is intentionally not an agent runtime — no built-in ReAct loop, planning, multi-agent orchestration, skills, or pluggable memory contracts.
AgentsKit ships a standalone runtime (@agentskit/runtime) plus the chat layer, and interoperates with the Vercel AI SDK through an adapter — you can keep it and add a runtime on top.
Pick Vercel AI SDK if you only need streaming chat in React/Next. Pick AgentsKit when the chat box becomes an agent.
#vs assistant-ui
assistant-ui is a strong React component library — 50+ components — but it is UI only and has no opinion about runtime, tools, memory, or how to compose them.
AgentsKit components are headless (data-ak-* attributes, no hardcoded styles) and span React, Vue, Svelte, Solid, Angular, React Native, and Ink, backed by the same core.
Pick assistant-ui for batteries-included React chat UI. Pick AgentsKit for headless UI across frameworks plus the runtime behind it.
#vs Mastra
Mastra is a capable TypeScript agent framework with workflows and runtime. It is framework-shaped: you adopt its structure. AgentsKit is kit-shaped: install one package, grow into the full stack, no required project layout.
Pick Mastra if you want a batteries-included framework. Pick AgentsKit if you want substitutable packages and zero lock-in.
#When AgentsKit is the wrong choice
- You need one off-the-shelf integration that only LangChain has today, and bundle size does not matter.
- You only need streaming chat in React and will never add a runtime — the Vercel AI SDK is less code.
- You want a fully prescribed framework that makes every decision for you.