Find why your AI agent failed in under 5 minutes.

Trace every step, get root-cause explanations, and cut token cost with actionable fixes.

No credit card. First trace in about 3 minutes.

AI agents are powerful — but hard to debug

Common challenges

Why did my agent fail?

Which prompt caused this issue?

Why is cost increasing?

Which tool broke?

How AgentScope helps

Full run tracing across your agent

Root cause analysis for failures

Prompt, response, and tool visibility

Cost and latency tracking

Everything you need to understand your agents

Run Tracing

Visualize every step of your agent with spans, tools, and LLM calls in a clear timeline.

Root Cause Analysis

Automatically detect why runs fail and pinpoint the exact step that broke.

Optimization Insights

Get suggestions to improve prompts, reduce latency, and lower cost.

Demo Quickstart

Copy your API key, run a local demo app, and inspect traces in under a minute.

Real-Time Status View

Watch your run live with workflow graph updates, active spans, streaming logs, and artifacts as they happen.

Sub-200ms live updates
Workflow Graphrunning
input.parse
planner.llm_call
tool.search
Active Span Timeline00:01.27
Live Logs

[10:24:41] span_started planner.llm_call

[10:24:41] token_stream +42

[10:24:42] artifact_created prompt.rendered

From failure to insight in seconds

Quickly move from a broken run to a clear explanation and actionable fix.

Failed Run #1247

1420ms total
Span
Timeline
0ms
355ms
710ms
1065ms
1420ms
root.run
1420ms
planner.step
360ms
tool.search_docs
420ms
tool.rank_results
180ms
tool.send_email
210ms
llm.final_answer
240ms
Root Cause Analysis

Likely cause:

External API endpoint timeout. This endpoint has a 94% failure rate in the last hour.

Suggested fix:

Increase timeout to 10s or add retry logic with exponential backoff

Send your first trace in 3 steps

Install SDK, add API key, and verify trace delivery.

quickstart.sh
# 1) Install SDK
pip install agentscope-sdk

# 2) Add your API key
export AGENTSCOPE_API_KEY=proj_live_xxx

# 3) Send a minimal trace
import os
import agentscope

os.environ["AGENTSCOPE_API_KEY"] = os.getenv("AGENTSCOPE_API_KEY", "")
agentscope.auto_instrument()
print("trace received: run_01H...")

Control your cost. Improve performance.

Track token usage, latency, and cost across every run and model.

Token Usage

987K
tokens this week

Cost per Run

$0.28
avg per run

Latency Trends

1.4s
avg latency

Built for teams

Organizations and projects

Organize your work by team and project

API key authentication

Secure access with API keys

Onboarding path to first trace

Signup includes organization + project context and guided setup

Secure multi-tenant architecture

Your data is isolated and protected

Start debugging your AI agents today

Understand failures, improve performance, and ship better AI faster.