adk-rust: Build AI Agents with Rust

Overview

ADK-Rust is a Rust framework for building AI agents, from single-model assistants to tool-using teams and durable graph workflows. It brings orchestration, model integrations, session handling, and deployment options into a modular crate ecosystem.

It is suited to Rust developers who want typed, async agent infrastructure and the flexibility to select providers and capabilities. Its breadth is useful for production-oriented systems, but may be more than needed for a simple prompt-to-model application.

Key Features

• Integrates with Gemini by default, with optional support for OpenAI, Anthropic, and other providers.
• Provides agents, sequential and parallel workflows, teams, and graph-based orchestration.
• Supports typed tools, MCP integrations, and sandboxed coding-agent capabilities.
• Includes session storage, long-term memory, retrieval-augmented generation, and artifact handling.
• Offers real-time voice and multimodal agent capabilities through optional crates.
• Includes HTTP serving, evaluation, guardrails, authentication, and OpenTelemetry support.
• Uses tiered features and separate crates so projects can select capabilities rather than enabling everything.

Use Cases

• Rust teams building assistants that call typed tools and need an async runtime.
• Developers composing multi-agent workflows, including graph execution with checkpoints and human approval pauses.
• Applications that need to switch model providers while keeping agent logic largely unchanged.
• Services that expose agents over HTTP or integrate retrieval, memory, and observability.
• Voice-agent projects that need realtime audio capabilities alongside other agent components.

Getting Started

Install the project scaffolding command and create an OpenAI-backed API project:

cargo install cargo-adk
cargo adk new quickstart_agent --template api --provider openai

Add the required API key to the generated .env file, then run cargo run. See the README for provider setup, feature tiers, examples, and documentation links.

Alternatives

• rig: Rig is a Rust LLM application library focused on shared provider and vector-store interfaces, while ADK-Rust offers broader agent workflows, memory, and real-time voice.

Considerations

• Most provider integrations and advanced capabilities are opt-in; check the feature requirements for the crates you plan to use.
• A model provider generally requires its own credentials, and managed providers may also require account- and region-specific identifiers.
• The README describes experimental crates, including adk-managed and adk-codeact-monty; review their stability before relying on them in production.
• The supplied repository metadata reports the license as NOASSERTION, while the README states Apache 2.0. Verify the repository license files and terms for your use case.