Performance

Measured, not marketed.

Transparent benchmarks for every middleware module. We publish the overhead so you don't have to guess.

0.12ms

Median module overhead

0.8ms

P99 overhead

2.1MB

Avg memory per module

4.2ms

Full chain (58 modules)

Methodology

Environment: Single-core VM, 2GB RAM, Ubuntu 24.04, Go 1.23. stockyard v1.0.0 compiled with -trimpath -ldflags="-s -w".

Workload: 10,000 synthetic requests per module. Payload: 1,200 input tokens, 500 output tokens (GPT-4o format). Cold start excluded — first 100 requests discarded.

Measurement: Per-module latency measured by wrapping each middleware handler with time.Now() / time.Since(). Memory via runtime.MemStats delta before/after module init. Reported values: median, P95, P99.

Transparency: These are module-only overheads — they exclude network I/O to upstream providers. A real request's total latency is dominated by the LLM provider (typically 500ms–3s), not the proxy chain. The full 58-module chain adds ~4ms total.

Sort by: Filter:

Module	Category	Median	P95	P99	Memory	Overhead	Grade