Metrics
Every ticket the pipeline has processed, and what it cost to do it.
Where tickets stand
Each card filters the inbox.
0
Tickets processed
25 total in the inbox
0
Awaiting approval
Drafted, waiting on a person
0
Escalated
A guardrail took the decision
0
Sent by an operator
8% resolution rate
2 runs failed partway and are excluded from the averages below.
Throughput and efficiency
Change is measured against the previous 11 runs.
375%
0 ms
Avg. latency
538%
$0.0055
Avg. cost / ticket
$0.13 across all runs
71%
0
Avg. tokens / ticket
94%
Avg. triage confidence
How sure the classifier was
Median run (p50)
9.44 s
Slow run (p95)
103.89 s
Slowest recorded
257.44 s
What the agents decided
Guardrail verdicts
21drafts
Every draft lands in exactly one of these, so the shares total 100%. Pick one to see those tickets in the inbox.
Usage by model
- qwen/qwen3.6-27b14 hops56,648 tok$0.10
- openai/gpt-oss-120b24 hops51,072 tok$0.02
- openai/gpt-oss-20b13 hops8,358 tok$0.0016
- llama-3.3-70b-versatile4 hops5,735 tok$0.0040
- llama-3.1-8b-instant12 hops5,470 tok$0.0004
Total across the pipeline127,283 tok · $0.13
Cost is estimated from total tokens at Groq's published rates. The full multi-agent pipeline runs for a fraction of a cent per ticket.
Where the time goes
Tool usage
- kb_search19 calls · 151 ms avg
- account_lookup15 calls · 158 ms avg
- invoice_lookup9 calls · 155 ms avg
- subscription_lookup6 calls · 101 ms avg
- refund_calc3 calls · 99 ms avg
Agent utilisation
- Triage25 hops · 16.22 s
- Router25 hops · 0 ms
- Supervisor21 hops · 123.27 s
- Billing agent9 hops · 402.31 s
- Technical agent7 hops · 129.85 s
- Account agent5 hops · 80.01 s