Report Abuse

Basic Information

Agentic ADK is an agent application development framework from Alibaba International AI Business, based on Google-ADK and ali-langengine. It is intended to help developers build, construct, evaluate, orchestrate, and deploy powerful, flexible, and controllable AI Agents and multi-agent collaborations. The project provides engine modules, infrastructure and demo examples to support a range of agent applications from simple tasks to complex coordinated workflows. It emphasizes streaming interaction, visualization debugging, workflow integration, and context extension for conversations including session, memory and artifact handling. The repository documents connectors and configuration examples for many LLM providers, vector stores and infra components and specifies a JDK 8+ runtime requirement.

Links

App Details

Features
Compatibility with Google ADK interfaces and enhanced streaming interaction and visualization debugging tools. Integration with Alibaba"s Ovis multimodal model to fuse visual and textual understanding with improved chain-of-thought reasoning and multilingual OCR/structured-data extraction. Flexible multi-agent framework supporting synchronous, asynchronous, streaming and parallel execution modes and A2A protocol. High-performance workflow engine built on SmartEngine and RxJava3 with a node-based process system for defining agent behaviors and bidirectional communication. Hundreds of API tools plus an MCP integration gateway. Built-in best-practice patterns such as DeepResearch/RAG, ComputerUse, BrowserUse and Sandbox. Context extensions including Session, Memory and Artifact with short and long-term memory plugins. Prompt automation tuning and security risk control examples.
Use Cases
The repository reduces development effort by providing reusable engine components, a reactive workflow engine, and debugging and visualization tools that simplify building and orchestrating agents. Multimodal integration with Ovis enables combined visual and textual reasoning useful for tasks like OCR, table and chart extraction and cross-language understanding. The node-based workflow and reactive model let developers express complex agent behaviors and communication patterns including synchronous, asynchronous and streaming flows. Prebuilt configuration samples and connectors for many LLMs, search, and vector services accelerate external integration. Built-in memory modules, session management and artifact support simplify conversational state handling. Examples for prompt tuning and security controls help teams adopt best practices when deploying agent applications.

Please fill the required fields*