Configure a synthetic agent workload, click Run Simulation, and see an investor-grade
cost & time breakdown generated in seconds — no API key required in mock mode.
Simulation Parameters
10 – 2 000 (300 is the headline demo)
Same seed → identical results every run
Controls the ratio of easy/medium/hard/expert tasks in the workload
0.10
Override the % of tasks that truly need the frontier model
1.0×
Scales token count on hard + expert tasks
12%
Fraction of tasks that are exact cache repeats
4%
Fraction of tasks that are runaway loops (SPRT stops them)
65%
Fraction of verbose tasks eligible for context compression
Use real Claude API
(ANTHROPIC_API_KEY not set — live mode unavailable)
When OFF: deterministic mock — free, no API key needed. When ON: real Claude calls.
Running simulation…
Report ready. The full investor-grade analysis is shown below —
cost & time savings decomposed by routing, compression, caching and loop-stopping.