Trace every tool call
Follow each step, prompt, and output with parent-child context so you can inspect agent decisions end to end.
§ Open-Source Observability
See what your agents are doing before production breaks.
AgentScope captures traces, cost hotspots, and failure patterns across your agent graph so teams can debug behavior fast and keep budgets predictable.
42%
faster incident root-cause analysis in benchmark runs
Core promise
Open-source observability for multi-agent systems.
§ EARLY ACCESS
Get on the list.
Join the AgentScope launch list to receive the OSS starter bundle and roadmap updates.
Cost and latency heatmaps
Find expensive nodes and bottlenecks by agent, task type, and environment without exporting logs manually.
Failure pattern detection
Spot retry loops, hallucination clusters, and timeout cascades before they become production incidents.
Open data model
Integrate traces into your stack using open schemas and push events into existing monitoring pipelines.
§ HOW IT WORKS
Three steps.
Step 01
Instrument your runtime
Drop in the SDK and start streaming spans from local and distributed agent executions.
Step 02
Correlate behavior in real time
Inspect linked traces, token usage, and external API calls in one timeline-oriented workspace.
Step 03
Act on regressions quickly
Trigger alerts from cost or quality thresholds and pin the responsible run in your incident workflow.