Eino V0.9.0 Alpha Release: AgenticMessage

[mrh997]

As capabilities like built-in tools, MCP tools, and caching gain native support in advanced vendor APIs such as OpenAI Responses API and Claude API, models are evolving from "one-shot Q&A engines" to "goal-oriented autonomous agents." We're excited to announce that the Eino framework has launched a brand-new message structure definition AgenticMessage, along with three core components: AgenticModel, AgenticChatTemplate, and AgenticToolsNode, providing powerful support for building more intelligent and autonomous AI applications.

Eino v0.9.0-alpha.2 has been released. Update eino-examples@release/agentic_model now to experience Eino's Agentic capabilities and build next-generation intelligent applications! 🚀

git clone https://github.com/cloudwego/eino-examples.git
git checkout release/agentic_model
cd ./eino-examples/flow/agent/react/agentic
go run main.go

---

✨ Core Updates

1️⃣ AgenticMessage

• AgenticMessage: Carries complex composite content (such as "reasoning process + tool calls + text generation") through a polymorphic ContentBlock structure, providing dedicated extension fields for different model providers like OpenAI, Claude, and Gemini.
• ContentBlock: Supports the following 19 content types, including reasoning, multimodal input/output, tool calls and results:

User Input

• UserInputText - Text input
• UserInputImage - Image input
• UserInputAudio - Audio input
• UserInputVideo - Video input
• UserInputFile - File input

Model Generation

• Reasoning - Reasoning process
• AssistantGenText - Generated text
• AssistantGenImage - Generated image
• AssistantGenAudio - Generated audio
• AssistantGenVideo - Generated video

Tool Calls

• FunctionToolCall/Result - User-defined tools
• ServerToolCall/Result - Server-side built-in tools
• MCPToolCall/Result - MCP protocol tools
• MCPListToolsResult - MCP tool list
• MCPToolApprovalRequest/Response - MCP tool approval

---

2️⃣ AgenticModel

AgenticModel is a model capability abstraction centered on "goal-driven autonomous execution" and serves as an enhanced abstraction of ChatModel.

📦 Interface Definition

type AgenticModel interface {
    Generate(ctx context.Context, input []*schema.AgenticMessage, opts ...Option) (*schema.AgenticMessage, error)
    Stream(ctx context.Context, input []*schema.AgenticMessage, opts ...Option) (*schema.StreamReader[*schema.AgenticMessage], error)
    WithTools(tools []*schema.ToolInfo) (AgenticModel, error)
}

🔥 Model Vendor Support

Vendor	Version
OpenAI	github.com/cloudwego/eino-ext/components/model/agenticopenai@v0.9.0-alpha.1
ARK	github.com/cloudwego/eino-ext/components/model/agenticark@v0.9.0-alpha.1
Gemini	In development
Claude	In development

---

3️⃣ AgenticChatTemplate

A prompt template component specifically designed for AgenticMessage, used for building structured context preparation.

📦 Interface Definition

type AgenticChatTemplate interface {
    Format(ctx context.Context, vs map[string]any, opts ...Option) ([]*schema.AgenticMessage, error)
}

---

4️⃣ AgenticToolsNode

A tool execution component specifically designed for AgenticModel, supporting automatic parsing and execution of tool calls generated by the model.

---

⚠️ Beta Notice

The current version is a Beta release, and the API may be subject to adjustments. We welcome community feedback and contributions!

Feedback Channels

• GitHub Issues: https://github.com/cloudwego/eino/issues