EinoExt

EinoExt provides additional capabilities beyond core component implementations:

• Callback Handlers: Official implementations of the Eino CallbackHandler interface. These are used for observability and tracing, such as the Langfuse tracing callback.
• DevOps Tools: An IDE plugin designed for Eino that provides visualized debugging and UI-based graph editing. For setup and usage, refer to the Eino Dev tooling documentation.

Use the agenticark package when you need a model.AgenticModel implementation backed by the Volcengine Ark Responses API. This implementation works with *schema.AgenticMessage types.

Import the package using:

import "github.com/cloudwego/eino-ext/components/model/agenticark"

The Tencent Cloud Hunyuan embedding implementation includes several built-in capabilities:

• Automatic Batch Processing: The embedder automatically splits large text arrays into batches of up to 200 texts to comply with Tencent Cloud API limitations.
• Token Usage Tracking: Tracks Prompt tokens and Total tokens via Eino's callback system.
• Callbacks Support: Fully supports Eino's callback system for error tracking, start/end event monitoring, and token usage statistics.

The XLSX Parser is a document parsing component for Eino that implements the Parser interface. It converts Excel (.xlsx) files into document formats, supporting flexible configurations for headers, worksheet selection, and metadata injection.

Key Capabilities

• Header Support: Process files with or without a header row.
• Worksheet Selection: Choose specific worksheets to parse.
• Custom ID Prefixes: Define a prefix for the generated document IDs.
• Metadata Preservation: Automatically converts row data into structured metadata and allows injecting additional custom metadata.

The XLSX Parser is an Eino component that implements the Parser interface to convert Excel (.xlsx) files into document formats. It allows for flexible configuration, including selecting specific worksheets, handling files with or without headers, and customizing document ID prefixes.

Key capabilities:

• Automatic conversion of table data to document format.
• Preservation of complete row data in the _row metadata field.
• Support for injecting additional metadata via the _ext field.
• Ability to target a specific sheet within a multi-sheet workbook.

The libs/acl/opentelemetry library provides a way to configure and initialize OpenTelemetry exporters and providers specifically for use with Eino. It simplifies the setup of observability instrumentation within Eino-based applications.

For complete implementation details and usage patterns, refer to the examples/ directory within this repository.

The Sequential Thinking Tool is a component designed for dynamic and reflective problem-solving. Inspired by the Model Context Protocol (MCP) sequential thinking implementation, it helps users and LLMs solve problems through a structured series of guided questions and self-reflection steps.

Key Features

• Guided step-by-step thinking processes.
• Dynamic questioning and self-reflection capabilities.
• Enhanced problem-solving through structured analysis.

Ideal Use Cases

• Decomposing complex problems into manageable steps.
• Planning and design where iterative modifications are expected.
• Analyzing directions that may require correction.
• Handling tasks where the full scope is not immediately clear.
• Maintaining context across multiple sequential steps.
• Filtering out irrelevant information during reasoning.

Human-in-the-loop (HITL) allows agents to pause execution, request human input, and resume from the exact point they stopped. This is achieved through three core components:

1. Interrupt: The agent pauses and sends information to the user (e.g., asking for approval or clarification).
2. Checkpoint: The framework saves the current execution state to a CheckPointStore.
3. Resume: The user provides input, and the framework restores the saved state to continue execution.

Common patterns include Approval (pausing before an action), Review & Edit (modifying tool arguments), Feedback Loops (refining content), and Follow-up (asking for clarification).

Because the Gemini API does not provide tool calling IDs in its responses, this implementation automatically generates a unique UUID (v4) for every tool call. This ensures compatibility with the Eino framework and allows for correct tool execution tracking in complex Agent workflows.

Key properties:

• Each tool call receives a new, globally unique UUID.
• This prevents ID conflicts across multiple model interactions in a single session.
• It follows the standard UUID v4 format (e.g., 550e8400-e29b-41d4-a716-446655440000).