Concept
What is Token?
Also known as: tokens, tokenisation
A token is the unit a language model reads and writes — roughly a word-piece of about four characters in English — and the unit that context limits, pricing and speed are all measured in.
Models don't see letters or words; they see tokens. Unbelievable might be three of them, a common word is usually one, and a rough English rule of thumb is four characters or about 0.75 words per token — though code, punctuation and non-English text tokenise far less efficiently. This is the currency of the whole system: context windows are counted in tokens, providers bill per million input and output tokens, and generation speed is quoted in tokens per second. Different models use different tokenisers, so counts don't transfer.
Tokens are why context engineering exists. Every tool schema, retrieved document and prior turn you include is paid for on every call it appears in — in money, in latency, and in the room left for anything else. That arithmetic is the reason Velaris keeps connectors out of the prompt: a thousand tools as schemas would be an enormous standing cost per step, so the capability mesh holds them as searchable data and the agent loads only the few it needs. Model routing is the same calculation applied to price per token.