agtrace-sdk 0.5.7

Public SDK for building observability tools on top of agtrace
Documentation

agtrace-sdk

Installation

Add this to your Cargo.toml:

[dependencies]
agtrace-sdk = "0.5"

agtrace-sdk: SDK for AI agent observability.

Note: README.md is auto-generated from this rustdoc using cargo-rdme. To update: cargo rdme --workspace-project agtrace-sdk

Overview

agtrace-sdk provides a high-level, stable API for building tools on top of agtrace. It powers agtrace's MCP server (letting agents query their execution history) and CLI tools, and can be embedded in your own applications. The SDK normalizes logs from multiple providers (Claude Code, Codex, Gemini) into a unified data model, enabling cross-provider analysis.

Quickstart

use agtrace_sdk::{Client, types::SessionFilter};

// Connect to the local workspace
let client = Client::connect_default().await?;

// List sessions and browse the most recent one
let sessions = client.sessions().list(SessionFilter::all())?;
if let Some(summary) = sessions.first() {
    let handle = client.sessions().get(&summary.id)?;
    let session = handle.assemble()?;

    println!("Session: {} turns, {} tokens",
        session.turns.len(),
        session.stats.total_tokens);

    // Browse tool calls
    for turn in &session.turns {
        for step in &turn.steps {
            for tool in &step.tools {
                println!("  {} ({})",
                    tool.call.content.name(),
                    if tool.is_error { "failed" } else { "ok" });
            }
        }
    }
}

For complete examples, see the examples/ directory.

Architecture

This SDK acts as a facade over:

  • agtrace-types: Core domain models (AgentEvent, etc.)
  • agtrace-providers: Multi-provider log normalization
  • agtrace-engine: Session assembly and analysis
  • agtrace-index: Metadata storage and querying
  • agtrace-runtime: Internal orchestration layer

Usage Patterns

Session Browsing

Access structured session data (Turn → Step → Tool hierarchy):

use agtrace_sdk::{Client, types::SessionFilter};

let client = Client::connect_default().await?;
let sessions = client.sessions().list(SessionFilter::all())?;

for summary in sessions.iter().take(5) {
    let handle = client.sessions().get(&summary.id)?;
    let session = handle.assemble()?;
    println!("{}: {} turns, {} tokens",
        summary.id,
        session.turns.len(),
        session.stats.total_tokens);
}

Real-time Monitoring

Watch for events as they happen:

use agtrace_sdk::Client;
use futures::stream::StreamExt;

let client = Client::connect_default().await?;
let mut stream = client.watch().all_providers().start()?;
while let Some(event) = stream.next().await {
    println!("Event: {:?}", event);
}

Diagnostics

Run diagnostic checks on sessions:

use agtrace_sdk::{Client, Diagnostic, types::SessionFilter};

let client = Client::connect_default().await?;
let sessions = client.sessions().list(SessionFilter::all())?;
if let Some(summary) = sessions.first() {
    let handle = client.sessions().get(&summary.id)?;
    let report = handle.analyze()?
        .check(Diagnostic::Failures)
        .check(Diagnostic::Loops)
        .report()?;

    println!("Health: {}/100", report.score);
    for insight in &report.insights {
        println!("  Turn {}: {}", insight.turn_index + 1, insight.message);
    }
}

Standalone API (for testing/simulations)

use agtrace_sdk::{SessionHandle, types::AgentEvent};

// When you have raw events without Client (e.g., testing, simulations)
let events: Vec<AgentEvent> = vec![/* ... */];
let handle = SessionHandle::from_events(events);

let session = handle.assemble()?;
println!("Session has {} turns", session.turns.len());

Contributing

Contributions are welcome! Please see the main agtrace repository for contribution guidelines.

License

Licensed under either of:

at your option.