What is AgentTrace?

AgentTrace (v0.3.41, released May 3, 2026) is a terminal-based observability tool for AI coding agents. It provides comprehensive monitoring across Claude Code, Codex CLI, Gemini CLI, Aider, and Cursor exports — tracking token usage, costs, latency, tool failures, and anomalies to help teams identify waste and optimize performance [citation:6].

Key Metrics Tracked

  • 💰 Token Costs: Input/output/cache tokens, estimated costs, top token-heavy agents
  • ⚠️ Tool Failures: Repeated tool calls, retry loops, long gaps, hanging sessions
  • ⏱️ Latency: P50/P95/P99 latency, per-tool latency ranking
  • 🧠 Thinking Quality: Blocks, character counts, shallow reasoning detection
  • 💯 Health Scores: Overall session health, anomaly detection, CI-ready reports [citation:6]

What It Catches

AI coding agents "burn tokens in loops, retry failures silently, and leave you with a surprise bill." AgentTrace finds the waste — so you stop paying for nothing and start shipping faster [citation:6].

Output Format

Beautiful terminal UI (TUI) with session comparison, anomaly detection, and CI integration:

  • Health scores (0-100) per session
  • Anomaly types (hanging sessions, shallow thinking)
  • Session comparison across agents
  • JSON reports for CI/CD integration

Pricing

Free and open source — 43 releases available on GitHub.

Pros

  • Captures hidden token waste and failures
  • Beautiful terminal interface (TUI)
  • Multi-agent support (Claude Code, Codex, Gemini, Aider, Cursor)
  • CI/CD integration with JSON reports
  • Health scores for at-a-glance assessment

Cons

  • Terminal-only (no web dashboard)
  • Requires installation and setup
  • Data stored locally (no cloud sync)
  • Advanced features may require configuration

Who Should Use It?

Perfect for: Teams using AI coding agents at scale, engineering managers tracking agent costs, and developers debugging agent behavior.

Verdict

AgentTrace is essential infrastructure for any organization using AI coding agents seriously. The token waste it catches pays for itself many times over.

Rating: 4.5/5 - The observability platform AI agents desperately needed.