AgentScope Studio — Agent Tracing UI

Practical Notes

• GitHub: 542 stars · 57 forks; pushed 2026-03-06 (verified via GitHub API).
• README recommends Node.js >=20 and provides an npm global install with a single as_studio entrypoint.
• README highlights OpenTelemetry tracing and evaluation analysis as first-class UI features.

Main

Make the UI actionable, not decorative:

1. Define a “run” contract: inputs, tools allowed, timeouts, and environment.
2. Capture traces for every run and keep them searchable by version (prompt/model/agent code).
3. Add simple evals (pass/fail rubrics) and correlate them with trace patterns (retries, tool failures, long spans).
4. Use the UI to produce a reproducible bug report: trace link + inputs + expected vs actual.

This is how agent debugging becomes engineering work instead of guesswork.

FAQ

Q: Do I need to use AgentScope? A: README positions it as AgentScope-focused, but the tracing concepts are OpenTelemetry-based.

Q: Is Docker required? A: No—README lists Docker as optional; the recommended install is via npm.

Q: What should I track first? A: Start with latency and tool failures, then add token usage and cost annotations.