Vibe Index
🎯

ag-ui-protocol

🎯Skill

from meetsmore/use-ai

VibeIndex|
What it does

ag-ui-protocol skill from meetsmore/use-ai

ag-ui-protocol

Installation

DockerRun with Docker
docker run -d \
DockerRun with Docker
docker run -d -p 8081:8081 -e ANTHROPIC_API_KEY='your-api-key' -e CORS_ORIGIN='*' ghcr.io/meetsmore/use-ai-server:latest
BunRun with Bun
bun dev
git cloneClone repository
git clone git@github.com:meetsmore/use-ai.git
BunRun with Bun
bun install

+ 1 more commands

πŸ“– Extracted from docs: meetsmore/use-ai
Need more details? View full documentation on GitHub β†’
1Installs
6
-
Last UpdatedFeb 5, 2026
View on GitHubBack to Skills

Skill Details

SKILL.md

AG-UI (Agent-User Interaction) protocol reference for building AI agent frontends. Use when implementing AG-UI events (RUN_STARTED, TEXT_MESSAGE_*, TOOL_CALL_*, STATE_*), building agents that communicate with frontends, implementing streaming responses, state management with snapshots/deltas, tool call lifecycles, or debugging AG-UI event flows.

Overview

# AG-UI Protocol

The Agent-User Interaction (AG-UI) Protocol is an open, lightweight, event-based protocol that standardizes how AI agents connect to user-facing applications.

When to Use This Skill

Use this skill when:

  • Implementing AG-UI protocol events in your code
  • Building agents that communicate with frontends via AG-UI
  • Understanding the event types and their structure
  • Implementing state management, tool calls, or message streaming
  • Debugging AG-UI event flows

Documentation

See the docs/2025-11-27/ directory for complete AG-UI protocol documentation:

  • introduction.md - Protocol overview and integrations
  • concepts/architecture.md - Core architecture and design
  • concepts/events.md - Event types and patterns
  • concepts/messages.md - Message structure and types
  • concepts/state.md - State management and synchronization
  • concepts/tools.md - Tool definitions and lifecycle
  • concepts/agents.md - Agent implementation
  • concepts/middleware.md - Middleware patterns
  • concepts/serialization.md - Event serialization and compaction
  • quickstart/introduction.md - Getting started guide
  • quickstart/server.md - Server implementation
  • quickstart/clients.md - Client implementation

Quick Reference

Event Types

```typescript

enum EventType {

// Lifecycle

RUN_STARTED = "RUN_STARTED",

RUN_FINISHED = "RUN_FINISHED",

RUN_ERROR = "RUN_ERROR",

STEP_STARTED = "STEP_STARTED",

STEP_FINISHED = "STEP_FINISHED",

// Text Messages

TEXT_MESSAGE_START = "TEXT_MESSAGE_START",

TEXT_MESSAGE_CONTENT = "TEXT_MESSAGE_CONTENT",

TEXT_MESSAGE_END = "TEXT_MESSAGE_END",

// Tool Calls

TOOL_CALL_START = "TOOL_CALL_START",

TOOL_CALL_ARGS = "TOOL_CALL_ARGS",

TOOL_CALL_END = "TOOL_CALL_END",

// State

STATE_SNAPSHOT = "STATE_SNAPSHOT",

STATE_DELTA = "STATE_DELTA",

MESSAGES_SNAPSHOT = "MESSAGES_SNAPSHOT",

// Custom

RAW = "RAW",

CUSTOM = "CUSTOM",

}

```

Event Patterns

  1. Start-Content-End: Streams content incrementally (text, tool arguments)
  2. Snapshot-Delta: State synchronization using complete snapshots + JSON Patch updates
  3. Lifecycle: Run monitoring with mandatory start/end events

Message Roles

  • user - User messages (text and multimodal)
  • assistant - AI responses (text and tool calls)
  • system - Instructions or context
  • tool - Tool execution results
  • activity - Progress updates
  • developer - Internal debugging

Tool Definition Structure

```typescript

interface Tool {

name: string; // Unique identifier

description: string; // Purpose explanation

parameters: JSONSchema; // Accepted arguments

}

```

Tool Call Lifecycle

  1. TOOL_CALL_START - Initiates with unique ID
  2. TOOL_CALL_ARGS - Streams JSON arguments
  3. TOOL_CALL_END - Marks completion

State Synchronization

  • STATE_SNAPSHOT - Complete state replacement
  • STATE_DELTA - Incremental JSON Patch (RFC 6902) updates

Source

Documentation downloaded from: https://github.com/ag-ui-protocol/ag-ui/tree/main/docs