Using TOON with LLMs

TOON is a strong fit for LLM workflows because it keeps structured data readable while reducing repeated JSON punctuation and field names.

Practical Prompt Pattern

Use TOON as the structured context block instead of pasting raw JSON.

use Sbsaga\Toon\Facades\Toon;

$context = [
    'account' => $account->only(['id', 'name', 'tier']),
    'orders' => $orders->map->only(['id', 'status', 'total'])->all(),
    'alerts' => $alerts->map->only(['severity', 'message'])->all(),
];

$prompt = <<<PROMPT
Review the following business context and produce a summary:

%s
PROMPT;

$response = $llm->chat(sprintf($prompt, Toon::promptBlock($context)));

Validating Model Output

If you ask a model to return TOON, validate it before trusting it:

$result = Toon::validate($modelOutput, strict: true);

if (!$result['valid']) {
    // Ask the model to regenerate or fall back to JSON handling.
}

Delimiter Tip

The wider TOON ecosystem recommends tab delimiters when maximum compactness matters. This package supports opt-in delimiter changes through config/toon.php.

'delimiter' => 'tab',

Safe Rollout Advice

For existing projects, keep compatibility_mode on legacy unless you are intentionally migrating consumers to the newer nested-data behavior.