Measurement setup

We timestamped every stage: market data ingestion, prompt evaluation, decision propagation, order submission, and exchange acknowledgment. Tests ran on the same hardware.

Results

HyperAgent averages 274ms total cycle time. PMM scripts hovered around 780ms because they polled REST and recalculated state naively. That delta let our agent capture 63 additional micro-trends during the +80% campaign.

Engineering choices

Executor keeps persistent WebSocket pools, batches updates, and only calls expensive models when Boss requires it. It also runs close to Hyperliquid’s matching engine for deterministic latency.

Reproduce it

This article links Grafana dashboards plus the instrumentation snippet. Run it against your bots and see exactly where HyperAgent saves precious milliseconds.