Offline & deterministic
Token counts come from gpt-tokenizer — pure JS, no network, no API key.
Same input, same number, every time.
tokcost
An instant, offline token + cost counter for LLM text. Point it at a file,
pipe text in, or pass a handful of files — tokcost counts the tokens locally
and, on request, estimates what that input would cost.
Token counts drive both your context budget and your bill, but they’re easy to
lose track of while iterating on a prompt. tokcost is a daily-use tool for
anyone building with LLMs: a fast, deterministic counter that runs entirely on
your machine — no network round-trip, no API key — and always tells you exactly
which encoding it used.
$ tokcost --cost -m gpt-4o prompt.md[1m42[0m tokens [2m(prompt.md)[0m[2mmodel: gpt-4o encoding: o200k_base[0m[2m≈[0m [1m$0.0001[0m (input)[2m — approximate, editable pricing[0mFiles or stdin
Count a single file, pipe text in over stdin, or pass multiple files for per-file counts plus a combined total.
Cost estimates
The optional --cost flag estimates input dollars from a small built-in
pricing table for common models.
Library too
Import countTokens and estimateCost into your own code — the CLI is a
thin wrapper over the same functions.
How it works
Section titled “How it works”tokcost resolves the model you pass (default gpt-4o) to a tiktoken encoding,
runs the text through gpt-tokenizer
locally, and prints the count. Add --cost and it multiplies that count by an
approximate per-million-token input price from a small built-in table. Add
--json and the whole result comes back as a structured object you can pipe
into other tooling.
Counts are exact for OpenAI models. For claude-* and gemini-* there is
no public local tokenizer, so tokcost approximates with the closest OpenAI
encoding (o200k_base) — and it always prints the encoding it used, so the
number is never a black box. See Models & pricing for the
full mapping and caveats.
Next steps
Section titled “Next steps” Quick start Install, count a file, pipe stdin, add --cost, and read JSON — in a few steps.
CLI reference Every flag, multi-file totals, exit codes, and examples.
Models & pricing The model → encoding mapping and the approximate pricing table.
Library API countTokens, estimateCost, resolveEncoding, and the exported types.