Analytics
What the agents did this week: requests, latency, tool health and spend. Figures come from the metrics service with their meaning attached.
Requests23,480+12% on the week beforeMeasured, 7 days
Tokens41.2M+9%Input and output, measured
Spend$318.40+4%Measured by billing; estimates excluded
p95 latency2.2 s+0.4 sFirst token, 7-day p95
Requests per day by model
Daily, measuredRequests by model
requests · Source: metrics service · As of: 2026-10-27T08:00:00Z
- Opus 5.5 (solid / circle)
- Sonnet 5.5 (striped / square)
- Haiku 5.5 (dotted / triangle)
Time to first token
p95, secondsp95 latency
s · Source: metrics service · As of: 2026-10-27T08:00:00Z
Tool health
Calls and failures over the period| Last hour | ||||
|---|---|---|---|---|
search_orders | 4,120 | 0.3% | 640 ms | Healthy |
kb.fetch | 2,210 | 4.3% | 880 ms | Failing |
images.resize | 2,140 | 68.2% | 2900 ms | Failing |
translate.glossary | 960 | 0.3% | 410 ms | Healthy |
payments.refunds.preview | 380 | 0.0% | 1210 ms | Healthy |
Sorted by Calls, descending.
Context window, median request
Context window16,900 of 32,000 tokens (53%)
- System prompt
- 2,100 (7%)
- Memories
- 700 (2%)
- Tool results
- 9,800 (31%)
- Messages
- 4,300 (13%)
Measured · Oct 27, 2026, 8:00 AM
Tool results take the most room. Trimming kb.fetch output would free about a quarter of the window.
Spend by agent
Measured and estimated kept apart| Run | Agent | Model | Tokens | Cost |
|---|---|---|---|---|
| Support | Sonnet 5.5 | 18,400,000 | $141.20 | |
| Orders | Opus 5.5 | 6,100,000 | $98.60 | |
| Translator | Sonnet 5.5 | 11,900,000 | $61.40 | |
| Media | Haiku 5.5 | 4,800,000 | $17.20 | |
| Eval service | Mixed | — | ~$28.90, estimated from token price list | |
| Total, last 7 daysReported by your billing | $318.40 | |||