At a glance
- Problem
- Writers and prompt authors need instant text analysis, and nobody wants to paste draft copy into someone else's server.
- My contribution
- I designed and built twelve local writing tools, including readability analysis and prompt budgeting with explicit token-count assumptions.
- Outcome
- A live product I operate end to end, with separate input/output budgets and clearly labeled historical cost estimates.
Engineering decisions
01decision
Keep draft text local
Readability, word statistics, keyword density, and token analysis run in the browser. These tools do not need a text-processing backend.
02decision
Distinguish tokens from words
Supported OpenAI entries use the explicitly named o200k_base encoder, including whitespace and literal token spellings. Other providers are labeled unvalidated ratio estimates, with no invented accuracy band.
TradeoffNaming one real encoder and labeling the rest as estimates is less impressive than a universal counter and does not pretend to an accuracy the tool cannot check.
03decision
Expose the cost assumptions
Pasted text, other input context, and an output reservation form a visible budget. Overages remain visible beyond 100%. Costs use an explicitly unverified April 2026 project snapshot; a derived-only report carries the assumptions without exporting the prompt.
What you can verify
- Live writing tools with immediate browser-side feedback.
- Inspect the selected token method and the assumptions behind any cost estimate.
- Compare prose, code, non-English text, and a budget that exceeds the selected context limit.
Source and release links checked .
Scope and limits
Readability formulas are indicators, not an assessment of a writer's ability. Token and cost estimates have model-specific limits and do not replace a provider's billed usage.
A short walkthrough
- 01
Paste a short draft and inspect its readability and word statistics.
- 02
Compare prose, code, and non-English text in the token counter.
- 03
Reserve other input and output tokens, inspect the context limit, and copy the budget report with its assumptions.