adk-telemetry

OpenTelemetry integration for Rust Agent Development Kit (ADK-Rust) agent observability.

Overview

adk-telemetry provides observability infrastructure for the Rust Agent Development Kit (ADK-Rust), built on OpenTelemetry 0.31 and tracing-opentelemetry 0.32:

• Tracing - Distributed tracing with OpenTelemetry 0.31
• Logging - Structured logging with tracing-subscriber
• Metrics - Performance metrics export via OTLP (tonic 0.12 / gRPC)
• Span Context - Propagation across agent boundaries

Installation

[dependencies]
adk-telemetry = "0.8.1"

Or use the meta-crate:

[dependencies]
adk-rust = { version = "0.8.1", features = ["telemetry"] }

Quick Start

use adk_telemetry::init_telemetry;

fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Initialize with service name
    init_telemetry("my-agent")?;

    // Your agent code here...
    Ok(())
}

Configuration

Set the RUST_LOG environment variable:

# Debug logging for ADK
RUST_LOG=adk=debug cargo run

# Trace level for specific modules
RUST_LOG=adk_agent=trace,adk_model=debug cargo run

OpenTelemetry Export

Configure OTLP export for distributed tracing:

use adk_telemetry::init_with_otlp;

fn main() -> Result<(), Box<dyn std::error::Error>> {
    init_with_otlp("my-agent", "http://localhost:4317")?;
    
    // Your agent code here...
    Ok(())
}

Available Functions

Span Helpers

Pre-configured spans for instrumenting ADK operations:

Token Usage Tracking

llm_generate_span pre-declares OpenTelemetry GenAI semantic convention fields. After receiving a response, call record_llm_usage to populate them:

use adk_telemetry::{llm_generate_span, record_llm_usage, LlmUsage};

let span = llm_generate_span("openai", "gpt-5-mini", true);
let _enter = span.enter();

// After receiving the LLM response:
record_llm_usage(&LlmUsage {
    input_tokens: 100,
    output_tokens: 50,
    total_tokens: 150,
    cache_read_tokens: Some(80),
    ..Default::default()
});

Recorded fields: gen_ai.usage.input_tokens, output_tokens, total_tokens, cache_read_tokens, cache_creation_tokens, thinking_tokens, audio_input_tokens, audio_output_tokens.

Re-exports

Convenience re-exports from tracing:

use adk_telemetry::{info, debug, warn, error, trace, instrument, Span};

Features

• Zero-config defaults with sensible logging
• OpenTelemetry 0.31 compatible span export
• OTLP export via tonic 0.12 (gRPC), aligned with adk-server's hyper 1.x / http 1.x stack
• Automatic context propagation
• JSON or pretty-print log formats

OpenTelemetry Dependency Versions

Related Crates

• adk-rust - Meta-crate with all components
• adk-core - Core traits and types

License

Apache-2.0

Part of ADK-Rust

This crate is part of the ADK-Rust framework for building AI agents in Rust.