AG-UI
An open protocol that standardizes how AI agents connect to user-facing apps.
- Stars
- 17k
- Forks
- 1.6k
- Licence
- MIT
- Language
- TypeScript
- Last update
- 5 days ago
- Type
- Framework
What it is
AG-UI is an open, lightweight, event-based protocol that defines how AI agent backends send events to a user interface and accept simple inputs in return. The README says it works over any event transport such as SSE or WebSockets, and it ships with a reference HTTP implementation and a default connector. Developers use it to bring agents into frontend applications, such as chat, shared state, and human-in-the-loop workflows.
- Stream agent responses to a frontend chat interface in real time
- Keep agent state and UI state synchronized in both directions
- Let an agent trigger frontend tools or render generative UI from structured messages
- Add human-in-the-loop approval steps to an agent workflow
Great for
- Developers adding an AI agent to a web or in-app user interface
- Teams using a supported agent framework such as LangChain, CrewAI, or Google ADK who want a standard frontend connection
- Builders who want real-time streaming chat and shared state between agent and UI
Not ideal for
- People who only want a ready-made chat product without writing code, since the README describes a protocol and developer tooling
- Projects that need tool access for agents alone, since the README positions MCP for that role rather than AG-UI
How to get it
- 1
Install with a package manager
Create a new AG-UI application
npx create-ag-ui-app my-agent-app - 2
Run it
- Run the npx create-ag-ui-app command with a project name
- Follow the Quickstart for building AG-UI-powered applications linked in the README
- Explore example demos on the AG-UI Dojo site
- 3
Get the most out of it
- Use the AG-UI Dojo demos to see shared state and generative UI in action
- Pick a supported framework integration that matches your existing agent backend
- Read the specs linked in the README for event type details
Downloads
Strengths
- Built on a small set of about 16 standard event types, per the README
- Transport-flexible: the README says it works with SSE, WebSockets, webhooks, and others
- Has a broad list of supported frameworks and an MIT license
Trade-offs
- The README is truncated and does not give full API details, so integration effort is hard to judge from it alone
- Several listed integrations are marked In Progress or rely on partner documentation hosted elsewhere
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| ag-ui (this) | 17k | Framework | intermediate | 5 days ago | Developers adding an AI agent to a web or in-app user interface |
| CopilotKit | 38.5k | Library | intermediate | 4 days ago | Developers adding an AI assistant or agent interface to a React or Next.js app |
| assistant-ui | 12.9k | Library | intermediate | 4 days ago | React or Next.js developers adding an AI chat panel to an existing app |
| deer-flow | 84.9k | AI model / agent | advanced | 4 days ago | Developers who want to experiment with multi-agent workflows on their own infrastructure |
| langchain | 149k | Framework | intermediate | 4 days ago | Python developers building LLM-powered applications |
| ECC | 295k | Dev tool | intermediate | 4 days ago | Developers who already use Claude Code and want a shared workflow for planning, testing and review |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.