AG-UI
The open event-based protocol connecting AI agents to user-facing applications
About
AG-UI (Agent-User Interaction Protocol) is the open, lightweight, event-based standard for the agent-to-frontend layer — the counterpart to MCP (agent-to-tools) and A2A (agent-to-agent). It standardizes how agent state, UI intents, and user interactions stream between any agentic backend and a user-facing app over HTTP/SSE or WebSockets, so developers can ship streaming, multimodal, interactive agentic features without ad-hoc wiring. Born from CopilotKit's partnership with LangChain and CrewAI, it now lists first-party support from AWS Strands Agents, Microsoft Agent Framework, Google ADK, Mastra, Pydantic AI, Agno, LlamaIndex, AG2, and the Claude Agent SDK, with TypeScript (@ag-ui/client, @ag-ui/core) and Python (ag-ui-protocol) SDKs plus community SDKs in Kotlin, Go, Dart, Java, Rust, Ruby, C++, and .NET. Clients range from CopilotKit React components to terminal apps, React Native, and Slack/Teams via the Channels SDK.
What it can do
- **Event-based protocol:** standardized events for text streaming, tool calls, state snapshots/deltas, and generative UI over HTTP/SSE/WebSockets
- **Framework integrations:** 1st-party support for LangChain, CrewAI, AWS Strands Agents, Microsoft Agent Framework, Google ADK, Mastra, Pydantic AI, Agno, LlamaIndex, AG2, Claude Agent SDK
- **SDKs:** TypeScript (`@ag-ui/client`, `@ag-ui/core`) and Python (`ag-ui-protocol`); community SDKs for Kotlin, Go, Dart, Java, Rust, Ruby, C++, .NET
- **Clients:** CopilotKit (React), terminal apps, React Native, Slack/Microsoft Teams via Channels SDK
- **Bidirectional:** frontend can send UI intents and user interactions back into the running agent
- **Interop:** designed to complement MCP (tools) and A2A (agent coordination), not replace them
Setup
Backend (Python example)
pip install ag-ui-protocol(ornpm install @ag-ui/core @ag-ui/clientfor TypeScript)- Wrap your agent run in an AG-UI event encoder and expose it as an SSE endpoint — see the per-framework integration docs (LangChain, CrewAI, Strands, ADK, etc.) at docs.ag-ui.com
- Emit the standard event stream: run started, text deltas, tool calls, state snapshots/patches, generative-UI renders
Frontend
- Connect with
@ag-ui/client(or CopilotKit's React components, which speak AG-UI natively) - Subscribe to the event stream to render streaming text, tool-call progress, and agent-driven UI updates
- Try the interactive demos at dojo.ag-ui.com before integrating
Prerequisites
An existing agent backend (any supported framework) and a frontend app. Familiarity with SSE/event-stream consumption. No account or API key needed — the protocol is open.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: 16,481 GitHub stars (log-scaled vs catalog peers); blended with ~7,378,422 monthly package downloads; last push 0d ago; 63 releases/12mo; 465 open issues. Manual review: npm @ag-ui/react does NOT exist (React via CopilotKit); 'In Progress' integrations unverified. License: MIT.
★ 16,481 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.5/10
- Maintenance · recency, releases, issue load
- 8.9/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · 16,481 GitHub stars (log-scaled vs catalog peers)
- · blended with ~7,378,422 monthly package downloads
- · last push 0d ago; 63 releases/12mo; 465 open issues
- · manual review: npm @ag-ui/react does NOT exist (React via CopilotKit); 'In Progress' integrations unverified
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.