Signoz Studio
Signoz Studio helps you understand AI agent runs.
See which steps failed and why, with evidence from SigNoz.
Toggle a fix to project how much time and money you would save.
Create an alert so the same failure gets caught next time.
Agent runs
2
Total recorded runs in this window
How often runs succeed
50%
Share of runs that finished without errors
Typical time to finish
8.32s
Average duration across recent runs
Typical $ per run
$0.0571
Estimated model cost, averaged
Start here: two example runs
Same kind of agent job. One worked. One didn’t. That’s the whole demo story.
Good agent run
This is what success looks like: retrieval found docs, the model answered, and formatting finished cleanly.
- Time
- 2.45s
- Cost
- $0.000620
- Tokens
- 2871
Broken agent run
Something went wrong: empty search, retries, higher cost. Scroll down to simulate a fix and create an alert without leaving home.
- Time
- 14.20s
- Cost
- $0.1135
- Tokens
- 35200
What’s the difference?
Both runs tried the same job. The broken one likely lost documents, retried steps, spent more tokens/money, or failed at the end. Compare lines up both paths so you can see what changed.
The broken run already has root-cause evidence. Toggle a fix to see what it would have cost, then create a SigNoz alert so it does not happen again.
What if we fixed this?
Toggle a fix to project what this run would have cost. Every number is recomputed from the recorded spans. Nothing is guessed.
Select a fix to project what this run would have cost.
Why it broke
Deterministic findings from telemetry. Every claim has evidence. Create an alert on any one to catch it next time in SigNoz.
Impact: Execution status is failed
- Span: agent.execute
- Error: Final answer quality degraded after empty retrieval
Suggested fix: Investigate: Final answer quality degraded after empty retrieval
Impact: Execution status is failed
- Span: formatter
- Error: Could not format incomplete model output
Suggested fix: Investigate: Could not format incomplete model output
Impact: Downstream planner/LLM likely lacked grounding context
- Documents returned: 0
- Retriever latency: 2.80s
Suggested fix: Investigate vector index freshness, query embedding quality, and collection filters
Impact: Added 2.80s to the critical path
- Retriever duration: 2.80s
Suggested fix: Investigate vector database latency, cold starts, and index size
All recent runs
Each row is one full agent execution recorded in SigNoz. Watch step-by-step replays the path; See what changed diffs against a good run.
| Run | Result | Time to finish | Model talk(tokens) | Est. cost | When | What next |
|---|---|---|---|---|---|---|
Broken agent run exec-regressed-001 Broken run | Failed | 14.20s | 35200 | $0.1135 | 7/23/2026, 10:00:00 AM | |
Good agent run exec-healthy-001 Good run | Success | 2.45s | 2871 | $0.000620 | 7/22/2026, 10:00:00 AM |