Summary

ADK-Rust is a production-ready, modular framework for building high-performance AI agents in Rust. It offers a flexible, model-agnostic, and deployment-agnostic architecture, empowering developers to create sophisticated AI systems with simplicity and power. The framework targets engineers and power users looking to leverage Rust for robust AI agent development, benefiting from type safety, async execution, and extensive features.

How It Works

ADK-Rust employs a layered, modular architecture, abstracting core functionalities like LLM integrations, tool execution, memory management, and real-time voice processing. It supports diverse agent types, including LLM agents, sequential/parallel workflow agents, and LangGraph-style graph agents for complex stateful orchestration. A key design principle is its unified API across numerous LLM providers and a zero-boilerplate #[tool] macro for seamless tool integration, simplifying agent development.

Quick Start & Requirements

• Primary install/run command:

  cargo install cargo-adk
  cargo adk new my-agent
  cd my-agent && cargo run
  

• Prerequisites: Rust 1.85+ (Rust 2024 edition). API keys for desired LLM providers (e.g., GOOGLE_API_KEY, OPENAI_API_KEY). WebDriver is required for browser automation features.
• Dependencies: The adk-rust crate, with optional features like full for comprehensive functionality or specific features for modules like realtime-openai.
• Links:
  • ADK Playground: https://playground.adk-rust.com
  • Examples: https://github.com/zavora-ai/adk-playground.git
  • API Reference: docs.rs/adk-rust

Highlighted Details

• Extensive Model Support: Integrates with over a dozen LLM providers including Gemini, OpenAI, Anthropic, Groq, Ollama, and more, via a unified API.
• Real-time Voice Agents: Enables bidirectional audio streaming with OpenAI and Gemini, supporting WebRTC and LiveKit bridges for production-grade voice assistants.
• Graph-Based Workflows: Implements LangGraph-style orchestration for complex, stateful agent workflows with checkpointing and human-in-the-loop capabilities.
• Browser Automation: Provides 46 WebDriver-based tools for agents to navigate, interact with, and extract information from web pages.
• Agentic Commerce: Includes support for ACP and AP2 payment orchestration with durable transaction journals.
• Local Inference: Features the adk-mistralrs crate for native local LLM inference, supporting quantization and GPU acceleration.

Maintenance & Community

The project is actively maintained, with a roadmap for 2026 and numerous contributors acknowledged. The ADK Playground offers a free, online environment to run over 70 ADK-Rust AI Agents. Community discussions are encouraged, though specific Discord/Slack links are not detailed in the README.

Licensing & Compatibility

The project is licensed under the Apache 2.0 license, consistent with Google's Agent Development Kit. This license generally permits commercial use and integration into closed-source projects.

Limitations & Caveats

The adk-mistralrs crate is distributed via git dependency, not crates.io, requiring explicit build configurations for local inference, especially for GPU acceleration. Certain advanced features, such as WebRTC or specific local inference backends, necessitate additional external dependencies (e.g., cmake) or specific build features. The v0.5.0 release introduced breaking changes to error handling and runner parameters.