The default Agentix.Tokenizer: a byte-count estimate (≈ English-text token density of ~4 bytes/token) with a deliberate ~20% over-estimate so budgeting errs toward leaving headroom — under-counting risks a hard over-window failure, while over-counting only compacts a little early. A +1 floor keeps non-empty text from counting as zero. No dependency; rough by design — see Agentix.Tokenizer.