Compare HeroUI Agent with OpenUI Cloud, Thesys, JSON Render, A2UI, Kapa AI, CopilotKit, OpenGenerativeUI, Tambo, and WebMCP.
HeroUI Agent gives product teams a hosted assistant with conversation state, tools, approvals, knowledge, monitoring, and generative UI in one embed. Some alternatives cover the whole agent experience; others solve one layer particularly well.
This comparison uses public product pages and documentation reviewed on September 20, 2026. Products change, so follow the linked sources before making a final decision.
| Option | Best fit |
|---|---|
| HeroUI Agent | A hosted web agent with the model, runtime, tools, UI, persistence, and operations handled together |
| OpenUI Cloud by Thesys | A generative UI layer for an existing AI stack, built around your React components |
| JSON Render | A schema-constrained generative UI renderer with custom catalogs and many output targets |
| A2UI | An open protocol for agents to send declarative, updatable UI to different client applications |
| Kapa AI | Knowledge retrieval and sourced answers for documentation, support, and developer education |
| CopilotKit | An open agentic frontend stack with self-hosting, multiple channels, and deep control over the agent architecture |
| OpenGenerativeUI | Open-ended React experiences using registered components or sandboxed HTML and SVG |
| Tambo | A React agent built from registered application components, with cloud and open-source deployment options |
| WebMCP | An existing website that compatible browser agents can operate while the page is open, without a dedicated agent UI |
HeroUI Agent owns the path from the user's message to the rendered response. The hosted runtime streams model output, keeps conversation state, calls server tools and MCP servers, and renders validated charts, tables, metrics, forms, and actions. The dashboard holds the system prompt, knowledge, appearance, usage, conversations, and run logs.
Your application keeps control of product-specific work. Client tools
run in the visitor's browser with the session and permissions your application already provides.
They can read page state, call authenticated APIs, navigate, or mutate records. Mark a tool with
needsApproval when the person should confirm it first.
This setup suits teams that want one supported agent product rather than assembling a model gateway, conversation store, tool protocol, chat shell, generative UI renderer, and operations dashboard.
OpenUI is the open-source language, renderer, and tooling. OpenUI Cloud, formerly the C1 API, is Thesys's managed layer around it. They belong to the same product family.
OpenUI Cloud concentrates on turning model output into streamed React interfaces. You define the component library, and its OpenAI-compatible API can sit inside an existing AI SDK, agent framework, or backend. Thesys also provides validation, model fallbacks, stateful endpoints, and observability.
HeroUI Agent supplies the surrounding agent product as well: the hosted conversation runtime, authenticated embed, client-tool bridge, approvals, knowledge, MCP connections, persistence, and monitoring. Its built-in UI catalog reduces setup, but it offers less rendering freedom today; custom Agent components are still coming.
OpenUI Cloud fits an existing agent stack where custom React components are central. HeroUI Agent fits teams that want the runtime and user-facing agent to arrive together.
JSON Render is an Apache-licensed generative UI framework from Vercel Labs. A developer defines a catalog of components, actions, and schemas; the model emits streaming JSON patches that the application renders through that catalog. Its packages cover React, React Native, Vue, Svelte, Solid, Next.js, email, PDF, video, images, terminal UIs, and other targets.
JSON Render gives developers ownership of the catalog and renderer. The application still chooses the model, prompt, conversation system, storage, authentication, and production operations. HeroUI Agent includes those layers in its hosted service and renders its own validated component catalog.
JSON Render fits teams building a custom generative UI layer across several output formats. HeroUI Agent fits teams that want an embedded assistant with the runtime, operations, and interface already connected.
A2UI stands for Agent-to-User Interface. Created by Google with community contributors, it is an open protocol and set of libraries for sending declarative, incrementally updatable UI from an agent to a client. The client maps the JSON description to trusted components in its own framework, while A2A, AG-UI, or another transport carries the message.
A2UI addresses interoperability between agents and client renderers; it does not prescribe the model runtime, conversation product, storage, monitoring, or billing. HeroUI Agent provides those services, but its generated UI is not exposed as the A2UI protocol. A2UI is in early-stage public preview: v0.9.1 is the current production release and v1.0 is a release candidate.
A2UI fits teams that need one UI message format across agents, trust boundaries, and native clients. HeroUI Agent fits teams that need a managed web agent more than a portable UI protocol.
Kapa AI specializes in ingesting unstructured knowledge and retrieving accurate, source-backed context. Teams can use its prebuilt website and support agents, or hand Kapa AI to their own agent through an API, SDK, or hosted MCP server.
HeroUI Agent includes a knowledge base, but its scope extends to application actions and generated interfaces. A support conversation can look up a policy, fetch the signed-in customer's order through a client tool, show the result as a record or table, and ask for approval before changing it.
Kapa AI is the sharper fit when documentation retrieval and support deflection drive the project. HeroUI Agent fits when sourced answers are one part of an assistant that also acts inside the product.
CopilotKit is an agentic frontend stack built around the open AG-UI protocol. It connects existing agent frameworks to web, mobile, and messaging interfaces. Teams can self-host its SDK and enterprise runtime or use Copilot Cloud, bring their own database, and render custom components.
HeroUI Agent makes more of those choices for you. The UI runs in a HeroUI-hosted iframe, the edge runtime manages conversations and model streaming, and the dashboard configures the agent. You add product context and client tools without operating an agent backend.
CopilotKit fits teams that prioritize protocol-level openness, self-hosting, multiple channels, or deep frontend control. HeroUI Agent fits teams that prefer a hosted web embed and a curated generative UI system.
OpenGenerativeUI is a CopilotKit project for open-ended visual responses. It can render registered React components and free-form HTML or SVG in sandboxed iframes. That makes it suitable for one-off visualizations and interfaces whose structure cannot be predicted from a fixed catalog.
HeroUI Agent keeps model output inside a built-in, validated catalog and adds the hosted runtime, tools, approvals, knowledge, persistence, and monitoring around it. OpenGenerativeUI fits teams that value rendering freedom and plan to assemble the surrounding agent stack. HeroUI Agent fits teams that prefer a narrower rendering surface inside a managed product.
Tambo is an open-source React toolkit for agents that render application components. Developers register a component with a description and Zod schema; Tambo selects it, generates valid props, and renders it in the conversation. It also supports browser tools, MCP, stored threads, authentication, and a self-hosted deployment.
Tambo starts with your React components. HeroUI Agent starts with a hosted, validated component catalog designed to work across its React, Next.js, and plain JavaScript integrations. HeroUI reduces the UI work needed to standardize an agent across products, while Tambo offers more control over the rendered component set today.
Tambo fits teams that need deeply custom React components or a self-hosted runtime. HeroUI Agent fits teams that prefer a managed runtime and ready-made interface.
WebMCP is a proposed browser standard for exposing a
page's tools to browser agents. A page can register JavaScript tools through document.modelContext
or turn an HTML form into a tool with attributes such as toolname and tooldescription. A
compatible browser agent can discover and call those tools while the page is open. The browser or
another service still supplies the agent runtime.
HeroUI client tools solve a related problem for the HeroUI-hosted agent. They execute in the customer page, but their schemas travel through the HeroUI embed rather than a browser-standard discovery API. HeroUI also provides the parts WebMCP leaves to the browser or another service: the model runtime, conversation UI, durable history, knowledge, server-side MCP, monitoring, and billing.
WebMCP tools are tab-bound and disappear when the user leaves the page. The proposal is currently in a Chrome origin trial and targets local workflows with a person present. HeroUI Agent supports modern Chrome, Edge, Firefox, and Safari releases through its SDK and hosted embed.
Use both when useful: keep backend systems available through MCP servers, use HeroUI Agent for the embedded assistant, and expose selected page actions through WebMCP for compatible browser agents.
HeroUI Agent is currently an invite-only beta. It is a hosted service, and generated responses use the built-in Agent UI catalog. Those constraints buy a smaller integration surface and one place to configure, observe, and support the resulting agent.
Request beta access, then follow the quickstart to add the hosted agent and your first client tool.
