Use Presets

Presets are named collections of strategies ordered for best results.

Available Presets

Preset	Strategies	Use when
safe	minify, collapse_whitespace	Output must be semantically identical
moderate	minify, drop_nulls, flatten, dedup_values_with_refs, tabular, strip_quotes, collapse_whitespace, compact_tables, strip_html_comments	Structural transforms are acceptable
aggressive	minify, drop_nulls, flatten, tabular, round_numbers, dedup_values_with_refs, strip_quotes, collapse_whitespace, compact_tables, strip_html_comments	Maximum savings, float precision may change

CLI

Bash

# Default (safe) — minify + collapse_whitespace
axm-smelt compact --file data.json

# Moderate
axm-smelt compact --file data.json --preset moderate

# Aggressive
axm-smelt compact --file data.json --preset aggressive

Python API

Python

from axm_smelt import smelt

# Safe
report = smelt(data, preset="safe")

# Moderate
report = smelt(data, preset="moderate")

# Aggressive
report = smelt(data, preset="aggressive")

Choosing a Preset

Use safe when: - The downstream consumer needs byte-identical semantics (e.g., signature verification) - The data contains floats that must not be rounded - You only want whitespace removed

Use moderate when: - Null/empty values are not meaningful and can be dropped - Repeated nested structures can be flattened - Repeated long strings benefit from aliasing - You want significant savings without altering numeric precision

Use aggressive when: - You want the maximum token reduction - Float precision beyond 2 decimal places is not needed - The data is large and savings matter more than exact fidelity

Inspecting what a preset would do

Use check to see estimates for each strategy before committing:

Bash

axm-smelt check --file data.json

Python

from axm_smelt import check

report = check(data)
for strat, pct in report.strategy_estimates.items():
    print(f"  {strat}: {pct:.1f}%")