eino-adk（埃诺阿德克）

Eino ADK Development Guide

When to Use This Skill

Use this skill when:

• User needs to build AI Agent applications
• User asks how to use Eino ADK
• User needs to implement multi-agent systems, workflows, human-in-the-loop
• User wants to understand core concepts like Agent, Runner, Tool

Quick Start

Step 1: Import Required Packages

import (
    "context"
    "errors"
    "fmt"
    "io"

    "github.com/cloudwego/eino/adk"
    "github.com/cloudwego/eino/adk/prebuilt/planexecute"
    "github.com/cloudwego/eino/adk/prebuilt/supervisor"
    "github.com/cloudwego/eino/components/model"
    "github.com/cloudwego/eino/components/tool"
    "github.com/cloudwego/eino/schema"
)

Step 2: Create ChatModelAgent

agent, err := adk.NewChatModelAgent(ctx, &adk.ChatModelAgentConfig{
    Name:        "my_agent",
    Description: "An agent that helps with tasks",
    Instruction: "You are a helpful assistant.",
    Model:       chatModel, // model.ToolCallingChatModel
    ToolsConfig: adk.ToolsConfig{
        Tools: []tool.BaseTool{myTool},
    },
})

Step 3: Create Runner and Execute

runner := adk.NewRunner(ctx, adk.RunnerConfig{
    Agent:           agent,
    EnableStreaming: true,
})

events := runner.Run(ctx, []schema.Message{
    schema.UserMessage("Hello!"),
})

for event, ok := events.Next(); ok; event, ok = events.Next() {
    if event.Output != nil && event.Output.MessageOutput != nil {
        if stream := event.Output.MessageOutput.MessageStream; stream != nil {
            for {
                chunk, err := stream.Recv()
                if errors.Is(err, io.EOF) {
                    break
                }
                if err != nil {
                    // handle error
                    break
                }
                fmt.Print(chunk.Content)
            }
        }
    }
}

Note: MessageStream is automatically closed by ADK, no manual cleanup required.

Core Workflow

Basic Execution Flow

1. Create Agent - Use NewChatModelAgent or prebuilt agents
2. Create Runner - Use NewRunner to wrap the Agent
3. Execute - Call runner.Run() or runner.Query()
4. Process Events - Iterate AsyncIterator[*AgentEvent] to handle output

Key Interfaces

Common Patterns

1. Basic Agent

See examples.md

agent, _ := adk.NewChatModelAgent(ctx, &adk.ChatModelAgentConfig{...})
runner := adk.NewRunner(ctx, adk.RunnerConfig{
    Agent:           agent,
    EnableStreaming: true,
})
events := runner.Query(ctx, "Hello!")

for event, ok := events.Next(); ok; event, ok = events.Next() {
    if event.Output != nil && event.Output.MessageOutput != nil {
        if stream := event.Output.MessageOutput.MessageStream; stream != nil {
            for {
                chunk, err := stream.Recv()
                if errors.Is(err, io.EOF) {
                    break
                }
                if err != nil {
                    // handle error
                    break
                }
                fmt.Print(chunk.Content)
            }
        }
    }
}

2. Workflow Agent

See examples.md

// Sequential - executes sub-agents in order
agent := adk.NewSequentialAgent(ctx, &adk.SequentialAgentConfig{
    SubAgents: []adk.Agent{agent1, agent2},
})

// Parallel - executes sub-agents concurrently
agent := adk.NewParallelAgent(ctx, &adk.ParallelAgentConfig{
    SubAgents: []adk.Agent{agent1, agent2},
})

// Loop - iterates until condition met
agent := adk.NewLoopAgent(ctx, &adk.LoopAgentConfig{
    SubAgents:     []adk.Agent{agent1},
    MaxIterations: 5,
})

3. Multi-Agent Systems

See examples.md

// Supervisor pattern
supervisor := supervisor.New(ctx, &supervisor.Config{
    Supervisor: supervisorAgent,
    SubAgents:  []adk.Agent{agent1, agent2},
})

// Plan-Execute-Replan pattern
entryAgent := planexecute.New(ctx, &planexecute.Config{
    Planner:   planAgent,
    Executor:  execAgent,
    Replanner: replanAgent,
})

4. Agent as Tool

See examples.md

agentTool := adk.NewAgentTool(ctx, subAgent,
    adk.WithFullChatHistoryAsInput(),
)

5. Interruption and Resume

See examples.md

// Trigger interruption (in tool's InvokableRun method)
err := adk.StatefulInterrupt(ctx, &ReviewInfo{...}, args)
return "", err

// Resume execution
iter, _ := runner.Resume(ctx, checkpointID)

6. Session Values

// Store
adk.AddSessionValue(ctx, "key", value)

// Retrieve
value, _ := adk.GetSessionValue(ctx, "key")

Configuration

Key configuration options in configuration.md:

• ChatModelAgentConfig
• RunnerConfig
• ToolsConfig
• AgentMiddleware

API Reference

Key APIs in api-reference.md:

• NewChatModelAgent
• NewRunner
• Run/Query/Resume
• NewAgentTool

Error Handling

Common error handling:

1. Tool call failure: Check tool definition and parameters
2. Model call failure: Check Model configuration and API Key
3. Resume failure: Verify CheckpointStore and ResumeInfo
4. Concurrency issues: Ensure Agent instances are not shared

References

• Core Concepts - Detailed core concepts
• API Reference - API documentation
• Examples - Example scenarios
• Configuration - Configuration options