Token Workbox is a free, privacy-first toolkit for counting LLM tokens and sketching API cost — without sending your prompts to a server.
Whether you are budgeting a long system prompt, comparing GPT vs Claude cost, or checking whether a Markdown doc fits a context window, the goal is the same: fast, honest numbers on your device.
Why we built this
Most “token counters” either:
- call a vendor API (your text leaves the browser), or
- show a single opaque number with no Exact vs Estimate label.
We take the other path:
- Local-first — counting runs in your browser. Your text never leaves your device.
- Honest accuracy — OpenAI-family models use exact encodings (js-tiktoken). Claude, Gemini, and open-weight models are clearly marked Estimate.
- Cost as a sketch — published list prices help you compare options; invoices may still differ.
What is on the site
| Tool | What it does |
|---|---|
| LLM Token Counter | Paste text or chat turns, count tokens, estimate cost, inspect encodings |
| Split by tokens | Cut long prompts/docs to a max token budget |
| Batch files | Count many local .md / .txt files at once |
Updates and how-tos live here under /blog. Privacy details are on /privacy; common questions on /faq.
Who it is for
Prompt engineers, developers, and writers who ship Markdown-heavy workflows and care about context limits — without pasting confidential prompts into a third-party counter.
Thanks for stopping by. Start with the token counter, or read the next posts for each utility.