Benchmarks

This document tracks reproducible benchmark commands and baseline outputs.

Build Profiles

AgentZero ships two build variants:

Variant	Profile	Features	Binary Size (macOS arm64)	Crate Count
default	release	All (TUI, WASM plugins, gateway, interactive)	~18 MB	~625 deps
minimal	release-min	Core only (memory-sqlite)	~5.2 MB	~262 deps

Build commands:

# Default build
cargo build -p agentzero --release

# Minimal build (size-optimized: fat LTO + opt-level z)
cargo build -p agentzero --profile release-min --no-default-features --features minimal

Prerequisites

• Build release binary once:

cargo build -p agentzero --release

• For default single-message benchmark (agent -m ...), set:

export OPENAI_API_KEY="sk-..."

Command Set

Run criterion core-loop benchmarks (offline, no external provider call):

cargo bench -p agentzero-bench --bench core_loop

Criterion writes reports under target/criterion/ with per-benchmark latency stats.

Run CLI startup benchmark script:

scripts/bench-cli-startup.sh --iterations 20 --command "--help"

Run single-message benchmark script:

scripts/bench-single-message.sh --iterations 10 --message "hello benchmark"

If you want a provider-free path for script validation:

scripts/bench-single-message.sh --iterations 10 --command "status --json"

Baseline Template

Record your local measurements in this table after each benchmark run:

Date (UTC)	Commit	Environment	Variant	Benchmark	Iterations	min_ms	avg_ms	max_ms	Notes
2026-03-03	HEAD	macOS arm64 (M-series)	default	cli_startup	20	19.05	42.90	443.08	—help
2026-03-03	HEAD	macOS arm64 (M-series)	minimal	cli_startup	20	21.18	41.20	307.36	—help