/ chat simulator

Every word costs tokens, even for "hi"

Chat with a simulated assistant and watch the token meter for every prompt and reply. Misspellings get fragmented into more sub-pieces, which is why "thnks" can cost more than "thanks". No real model is called, since all metrics are computed locally.

Simulated chat· model:
Try sending hi, thanks, or a misspelled question like wht is teh capital of france?

Per-interaction token ledger

One row per prompt and its reply. Every metric in the side panel is summed from this table.

#PromptReplyIn tokOut tokTotal tokCostLatencyTypos
No interactions yet. Send a message above to start filling the ledger.

Token breakdown per interaction

For each turn, see exactly where the tokens went: prompt pieces by kind, response pieces by kind, and chat-protocol special tokens (role markers like <|im_start|> that every API adds per message).

wordsubwordmisspellingpunctuationwhitespacespecial

No interactions yet. Send a message to see its token breakdown.

Token stream scrubber will appear once you send a message.