CLI

The binary is available as both asb and as-bench. Use npx asb <command> (or add scripts to package.json).

asb --version
asb --help

Commands

Command	Purpose
asb run	Build, then run benchmarks with full statistical analysis. The everyday command.
asb build	Compile benchmarks to wasm without running.
asb profile	Per-function work profile (instructions, time, or allocations).
asb watch	Rebuild and rerun on source change.
asb compare	Diff two saved baselines without re-running.
asb init	Scaffold config and an example bench.
asb doctor	Validate config, dependencies, input globs, and runtimes.
asb clean	Remove generated build/chart outputs.

run, build, profile, watch, compare, doctor, and clean accept --config <path> and --mode <name>. See Configuration.

asb run

asb run                              # all files from config input globs
asb run assembly/__benches__/my.ts   # a specific file
asb run --mode quick                 # apply a named overlay
asb run --filter "fib*"              # only benches whose name matches
asb run --json                       # machine-readable JSON to stdout
asb run --verbose                    # print every estimate

Timing flags (override config settings.*):

Flag	Effect
--warmup <ms>	Warmup time cap.
--warmup-tolerance <x>	Stable-met drift threshold (0 = fixed-time).
--warmup-min <ms>	Minimum warmup time.
--measure <ms>	Measurement window.
--samples <n>	Sample count (≥ 10).
--resamples <n>	Bootstrap resamples.
--sampling auto|linear|flat	Sampling strategy.
--confidence <x>	Confidence level, in (0, 1).

Numeric flags are validated against the same bounds as the config file — out-of-range values are rejected rather than silently producing garbage.

Baseline flags:

Flag	Effect
--save-baseline <id>	Save this run's raw samples as <id>.
--baseline <id>	Compare each bench against a saved baseline (node host only).

See Baselines.

Runtime flags:

Flag	Effect
--runtime <r>	node (default), wasmtime, wasmer, wazero, or a command template. Repeat to compare.
--deterministic	Record host imports once, replay them every later iteration (node host only).

See Runtimes.

Other:

Flag	Effect
--filter <pattern>	Run only benches whose name matches (substring or * glob; repeatable).
--json	Machine-readable JSON to stdout; suppresses human output.
--verbose, -V	Print all estimates (mean / median / std dev / MAD / slope).

asb build

asb build                            # compile all input files
asb build --runtime wasmtime         # also emit the pure-WASI WIPC build

Accepts build-selection flags only — no statistical or runtime execution.

asb profile

asb profile            # --instr (default)
asb profile --time
asb profile --alloc    # alias: --heap

Flag	Effect
--instr	Cost-weighted instruction count (default; exact, deterministic).
--time	Wall-clock self time (overhead-corrected; node host only).
--alloc, --heap	Bytes allocated (exact, deterministic).
--top <n>	Rows per bench (default 10).
--all	Include engine/runtime-internal rows.
--iters <n>	(--time / --alloc) iterations per bench.
--min-instrs <w>	(--time) don't wrap functions under w static instructions.

The legacy --heaviest=instr|time|alloc forms still work. See Profiling.

asb watch

asb watch
asb watch --filter "fib*" --mode quick

Accepts all asb run flags. Re-runs on source change, separating runs with a --- line.

asb compare

asb compare main dev
asb compare main dev --mode quick

Flag	Effect
--significance <x>	p-value threshold for "changed" (default 0.05).
--noise <x>	Ignore changes within ±x (default 0.01).
--config <path> / --mode <name>	Config / overlay selection.

See Baselines.

asb init

asb init
asb init --yes --install
asb init ./bench-project

Flag	Effect
--dir <path>	Target directory (also positional).
--install	Run install after scaffolding.
--yes, -y	Accept defaults; no prompts.
--force	Overwrite managed files.

asb doctor

Validates config loading, Node/dependency setup, bench-file discovery, and the selected runtime commands. See Doctor.

asb clean

asb clean                # build/chart outputs; baselines preserved
asb clean --baselines    # also remove saved baselines  (alias: --all)

JSON output

asb run --json writes one JSON document to stdout (all human output suppressed):

{
  "version": 1,
  "benches": [
    {
      "file": "assembly/__benches__/example.ts",
      "runtime": "node",
      "suite": null,
      "name": "fib(20)",
      "result": { "lb": 1.23e-3, "point": 1.30e-3, "hb": 1.37e-3 },
      "delta": { "lb": -0.02, "point": 0.015, "hb": 0.05, "pValue": 0.32, "verdict": "no change", "vs": "baseline 'main'" },
      "outliers": { "lowSevere": 0, "lowMild": 0, "highMild": 2, "highSevere": 0 },
      "warnings": []
    }
  ]
}

Times are in milliseconds.