AI NewsDev toolsAnnouncement

AG-UI, an open protocol for AI agents, ranks 4th on Trendshift

AG-UI is an open protocol for AI agents to send events to a user-facing app. It has 18,080 stars.

AI News

Editorial2 min read

LinkedInX
GitHub social card for the ag-ui-protocol/ag-ui repository

Image: GitHub

Why it mattersA team building an agent app this week can pick AG-UI as the message format. That is one less thing to design and one less thing to change later.

A developer building an agent app has to decide how the agent talks to the user's screen. The agent sends partial answers, tool calls, state changes and asks for input. The frontend shows all of that. Most teams invent a message format for this and change it twice. AG-UI is a shared one.

AG-UI is an open, event-based protocol for AI agents to connect to user-facing applications. It is ranked fourth on Trendshift's daily list with 245 mentions, against 18,080 stars on GitHub under the MIT licence. The project is maintained by the ag-ui-protocol organisation and was born from CopilotKit's partnership with LangChain and CrewAI. It has 5,467 commits and 1,838 forks.

What it does

The protocol defines events the agent sends to the app. These include streaming chat, state updates between the agent and the frontend, and generative UI. They also include tool-output streaming, human-in-the-loop interrupts, and files, images and audio. The frontend subscribes to the stream and renders each event.

That saves a team two decisions. First, which messages the agent and the app exchange. Second, how to add a new event later without breaking either side. A reader can see the event list and build either half on its own.

Who ships integrations

The README names first-party integrations for LangChain, CrewAI, and Microsoft Agent Framework. Google's two frameworks are also listed: Google ADK and Google Antigravity. The AWS side covers Strands Agents and Amazon Bedrock AgentCore. A second group names Mastra, Pydantic AI, Agno, LlamaIndex and AG2. Community integrations cover Claude Agent SDKs, Langroid and an in-progress OpenAI Agents SDK.

SDKs are listed for Python, JavaScript, Kotlin, Golang, Dart, Java, Rust, Ruby, C++ and .NET, with Nim in progress. Clients include CopilotKit, React Native, OpenBot, and bots for Slack and Teams.

Recent releases

The last three releases were small. The @ag-ui/langgraph package reached 0.0.46. The notes add support for LangGraph's native Interrupt.response_schema field in langgraph >= 1.2.12. They also unwrap serialised tool outputs and skip tool-end outputs with no tool call id. Each RUN_STARTED event now declares its protocol version.

The @ag-ui/claude-agent-sdk package reached 0.0.9 and fixed clone() of ClaudeAgentAdapter, which previously lacked state maps. Frontend tool parameter schemas now convert from JSON Schema recursively. The @ag-ui/a2ui-middleware package reached 0.0.15. It preserves recovery state after a stream completes and keeps action context out of the public transcript.

What a reader does with this

A team starting an agent app this week picks AG-UI as the message format. The agent code on one side and the frontend on the other both read it. The integration for one of the listed frameworks comes for free. A team that already has a message format can map it to AG-UI events. That is work, and it keeps the option to swap in a different agent framework later without rewriting the client.

The protocol does not dictate how the agent is built or where it runs. It says only what goes over the wire. A shared vocabulary for agent events exists before each team invents its own.

Source

ag-ui-protocol/ag-ui on GitHub.

SourceGitHub

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project