Crate adk_runner 

adk-runner

Agent execution runtime for ADK.

Overview

This crate provides the execution runtime:

• Runner - Manages agent execution with full context
• RunnerConfig - Configuration for the runner
• InvocationContext - Execution context implementation
• Callbacks - Hook points during execution

Quick Start

use adk_runner::{Runner, RunnerConfig};
use std::sync::Arc;

// Configure runner with services
// let config = RunnerConfig {
//     app_name: "my_app".to_string(),
//     session_service: sessions,
//     artifact_service: Some(artifacts),
//     memory_service: None,
// };
//
// let runner = Runner::new(config);

Features

• Automatic session management
• Memory injection
• Artifact handling
• Callback hooks at every stage

Structs

CacheMetrics

Metrics computed from session event history.

CachePerformanceAnalyzer

Utility for computing cache effectiveness metrics from session events.

Callbacks

Collection of all callback types

ContextCacheConfig

Configuration for automatic prompt caching lifecycle management.

EventsCompactionConfig

Configuration for automatic context compaction.

InvocationContext

MutableSession

MutableSession wraps a session with shared mutable state.

RequestContext

Identity and authorization context extracted from an HTTP request.

Runner

RunnerConfig

Traits

BaseEventsSummarizer

Trait for summarizing events during context compaction.

CacheCapable

Trait for LLM providers that support prompt caching.

Type Aliases

AfterModelCallback

Callback executed after model response

AfterToolCallback

Callback executed after tool execution

BeforeModelCallback

Callback executed before calling the model

BeforeToolCallback

Callback executed before tool execution