ADK-Rust

ADK-Rust Official Documentation

Welcome to the official documentation for ADK-Rust (Agent Development Kit for Rust). This documentation provides comprehensive guides and references for building AI agents using the Rust implementation of the ADK framework.

Getting Started

Core

  • Core Types - Fundamental types: Content, Part, Agent trait, Tool trait, contexts
  • Runner - Agent execution runtime and configuration
  • Plugins - Lifecycle hooks for tool/model interception and middleware

Models

Agents

Coding Agent

A native coding agent: read/edit/run code in a sandboxed workspace, autonomous goal mode, and parallel ultra-review workflows.

  • Overview - Start here: the pieces, install, and a 60-second quick start
  • Dev Tools - adk-devtools: read/write/edit/glob/grep/bash + the sandboxed Workspace
  • Harness - CodingAgent (the coding feature), the plan loop, write_todos
  • CLI - adk-rust code / goal (durable, resumable) / ultracode
  • Workflows - Ultra-review on adk-graph: parallel reviewers + fan-in
  • Examples - The three runnable example crates

Realtime & Multimodal

A dedicated, newcomer-friendly guide to building voice + multimodal agents — architecture, concepts, and runnable examples.

  • Overview - Start here: the mental model, install, and a 60-second quick start
  • Architecture - The four layers, event loop, audio pipeline, and turn lifecycle
  • Providers - OpenAI Realtime & Gemini Live models, voices, and how to choose
  • Tools - Server-side actions: native handlers, bridged adk_core::Tools, parallel calls
  • Multimodal - Video input via send_video_frame and browser camera capture
  • Affective Dialogue - Emotion-aware responses on Gemini native-audio
  • Memory - Context injection, turn storage, and knowledge-graph self-curation
  • Building Web Apps - The server-side bridge, WS protocol, and Web Audio
  • Examples - The four runnable examples and a suggested learning path

Tools

  • Function Tools - Create custom tools with async Rust functions
  • Built-in Tools - Pre-built tools like GoogleSearchTool
  • MCP overview and architecture - Where MCP fits and which ADK-Rust surface to use
  • MCP client - Tools, resources, prompts, completion, subscriptions, elicitation, tasks, and HTTP
  • Dynamic MCP server manager - Runtime local-server registry, persistence, monitoring, and restart
  • MCP server authoring - Publish tools and other capabilities with the aligned official Rust SDK
  • MCP security - Authentication, visibility, authorization, approval, and secret handling
  • MCP testing - Deterministic gates and deployment verification
  • Browser Tools - 46 WebDriver tools for web automation
  • UI Tools - Dynamic UI generation with forms, cards, charts
  • RAG - Retrieval-Augmented Generation for knowledge base search
  • ACP Tools - Use an ACP coding agent as a typed ADK-Rust tool
  • Retry & Reflect - Tool failure recovery with reflection prompts
  • Action Nodes - 14 deterministic node types for workflow graphs
  • Benchmarking - Performance measurement with cargo adk bench

Agent Client Protocol

Use an external coding agent from ADK-Rust, or expose an ADK-Rust agent to an editor through stable ACP v1.

Sessions & State

Memory

Long-term memory that outlives a session — the persistent counterpart to session state. A dedicated, newcomer-friendly guide to both the semantic store and the bi-temporal knowledge graph.

  • Overview - Start here: the two kinds of memory, the mental model, install, and a quick start
  • Concepts - MemoryEntry, the MemoryService trait, search, project scoping, GDPR erasure
  • Backends - The six stores (InMemory, SQLite, Postgres+pgvector, Redis, MongoDB, Neo4j) and how to choose
  • Knowledge Graph - GraphMemoryService: entities, observations, relations, and why bi-temporal matters
  • Tools & Agents - Wiring memory into agents; remember/relate and the memory tools

Callbacks & Events

  • Callbacks - Intercept and customize agent behavior
  • Events - Understanding the event system and conversation history

Artifacts

  • Artifacts - Binary data storage and retrieval

Observability

  • Telemetry - Logging, tracing, and monitoring

Deployment

Evaluation

Managed Agents

Security

  • Access Control - Role-based permissions and audit logging
  • Guardrails - PII redaction, content filtering, schema validation
  • Memory - Long-term semantic memory for agents
  • Payments and Commerce - Agentic commerce journeys, protocol support, and validation paths

Studio

  • ADK Studio - Visual development environment for building agents
  • Action Nodes - Non-LLM programmatic nodes for automation workflows
  • Triggers - Webhook, schedule, and event triggers for workflows

Development

Validation Status

Copy-paste Cargo commands and dependency snippets in the README and official docs are validated by scripts/check-doc-examples.sh. CI also checks cargo-adk scaffolds with scripts/check-cargo-adk-templates.sh, rejects duplicate example target names with scripts/check-example-name-collisions.sh, and compiles workspace examples with cargo check --workspace --examples.