Average success rate
93.8%
Agent minutes today
14h 24m
Across tasks completed today
Spend today
$41.40
Monthly budget
$1,570
Per-agent caps sum to this ceiling
Agent performance & cost
Throughput, reliability and spend against each agent's cap
| Agent | Tasks today | Success | Avg minutes | Spend today | Monthly cap | Cap used |
|---|---|---|---|---|---|---|
Product Analyst reasoning-v3 · v1.4.0 | 3 | 97.2% | 14 | $1.90 | $60 | 3% |
Architecture Agent reasoning-v3 · v1.2.1 | 1 | 95.8% | 22 | $2.40 | $80 | 3% |
Frontend Agent reasoning-v3 · v2.1.0 | 6 | 92.4% | 31 | $7.80 | $240 | 3% |
Backend Agent reasoning-v3 · v2.0.3 | 5 | 90.1% | 38 | $9.20 | $280 | 3% |
Database Agent reasoning-v3 · v1.8.0 | 2 | 94.6% | 26 | $4.10 | $120 | 3% |
QA Agent fast-v2 · v3.0.1 | 11 | 98.3% | 9 | $3.60 | $150 | 2% |
Security Agent reasoning-v3 · v1.6.2 | 4 | 96% | 18 | $5.00 | $160 | 3% |
DevOps Agent fast-v2 · v2.3.0 | 3 | 88.5% | 17 | $2.20 | $140 | 2% |
Reliability Agent fast-v2 · v1.9.4 | 8 | 93.7% | 7 | $2.90 | $130 | 2% |
Support Agent fast-v2 · v2.4.1 | 14 | 91.2% | 5 | $1.40 | $90 | 2% |
Licensing Agent fast-v2 · v1.1.0 | 0 | 89.4% | 6 | $0.00 | $70 | 0% |
Documentation Agent fast-v2 · v1.5.3 | 2 | 97.8% | 12 | $0.90 | $50 | 2% |
A cap is a hard stop: the agent pauses rather than continuing past its budget. Cost is tracked per run in agent_tool_calls.
Evaluation datasets
A version ships only after it beats the baseline
| Dataset | Cases | Cadence | Score |
|---|---|---|---|
| Code change quality | 240 | Regression suite on every PR | 94% |
| Incident triage accuracy | 120 | Weekly, replaying past alerts | 91% |
| Support reply quality | 180 | Weekly, human-graded sample | 88% |
| Policy compliance | 60 | Every agent version change | 100% |
Cost & reliability controls
Kept deliberately boring
- Per-task and per-day caps; a breach pauses the agent and notifies its owner.
- Workload routing sends tasks to the cheapest agent that meets the quality bar.
- Failures are reviewed weekly; recurring findings become evaluation cases.
- Provider changes are configuration, not a rewrite — spend stays comparable.