Skip to main content

AI tools

AI tools — token counts, prompt utilities, no logging

The hub for SnapToolz' AI-adjacent tools — the ones that help you work with LLMs and on-device models without handing your data to a third party. Some are text utilities for prompts (token counting, cleaning); others run a real model in your browser (background removal, upscaling, transcription). This page points you to the right one for the job.

  • No uploads — files stay on your device
  • WebAssembly fast
  • Works offline
  • 100% free, no sign-up

5 tools in AI tools

Browse all categories
AI Token CounterNew

Estimate token counts for GPT-4, GPT-4o, Claude, Gemini and Llama from any text.

AI Background RemoverNew

Remove image backgrounds with an on-device AI model. Downloads once (~25MB), then works offline. Your photos never leave your browser.

AI Image UpscalerNew

Double image resolution with on-device AI super-resolution. Sharper edges, recovered detail. Model downloads once (~50MB), then offline.

AI Voice TranscriptionNew

Whisper speech-to-text running on-device. 15 languages, SRT/VTT subtitles. Audio never leaves your browser.

Prompt Formatter & CleanerNew

Format, clean and JSON-escape your prompts — strip invisible chars, fix smart quotes, embed safely.

Start here: which AI tool for which job

The tools split into prompt utilities and on-device models. Pick by the task and follow the link:

You want toUseNote
Estimate LLM token cost / context fitToken CounterHeuristic, within ~10% for English
Clean / format / JSON-escape a promptPrompt ToolsStrips invisible chars that bloat tokens
Remove an image backgroundBackground RemoverONNX model, runs locally
Upscale a low-res imageImage UpscalerSuper-resolution, not magic
Transcribe speech to textTranscribeWhisper-class; accuracy varies with audio

Who this hub is for, and where to begin

Prompt engineers checking a prompt's cost before they send it, developers wiring a system prompt into a config, and anyone who wants an image or recording processed by AI without it becoming someone else's training data. The distinction worth knowing: the model tools download their weights once and then run on your device, while the prompt utilities are pure local text operations — either way, nothing you paste leaves the tab.

Where to begin: Token Counter to size a prompt against a context window, Prompt Tools to clean and JSON-escape it before you embed it, and the on-device models — Background Remover, Image Upscaler, Transcribe — when you need real inference on private media.

Frequently asked questions

  • Are the token counts exact?

    They're heuristic estimates — within ±10% of the real tokenizer for English prose, wider for code-heavy or non-Latin text. We don't bundle a real tokenizer because the smallest one would add 150 KB to every page load. For exact billing-grade counts, run the prompt through the provider's API.

  • Why would I clean a prompt?

    Pasted prompts often carry zero-width characters, BOM markers, smart quotes, and non-breaking spaces from Word, ChatGPT's UI, or PDFs. These bloat the token count, occasionally trigger sanitisation filters, and sometimes interfere with downstream parsing. The cleaner removes them deterministically.

  • What does JSON Escape do?

    It converts any text into a string body safe to embed inside a JSON string literal — backslashes, quotes, newlines all escaped. Drop the output between the quotes of an existing JSON config field. Useful when wiring system prompts into Anthropic / OpenAI / Vertex configs.

  • How is this different from pasting into ChatGPT or Claude's web tools?

    The major LLM web UIs send your prompt to their server — that's how they work. Even when a vendor pledges 'we don't train on your data', the prompt still leaves your machine and lives in their logs for some retention period. SnapToolz' prompt utilities run entirely in your browser, so the text never moves. Use SnapToolz for cleaning/inspection; use the LLM vendor when you actually need a response.

  • Can I use this for proprietary system prompts or client data?

    Yes — that's the use case it's best at. Token counting and prompt cleaning are pure text operations that don't require model inference. Doing them locally means your trade-secret prompts and customer data never become someone else's training data, log entry, or breach exposure surface.

  • Are my files uploaded anywhere?

    No. Every SnapToolz tool runs entirely in your browser using JavaScript and WebAssembly. Files are read locally, processed in memory, and the result is offered as a download. There is no backend.

  • Is everything free?

    Yes — every tool in this category and on the entire site is free with no sign-up, no daily limit, and no watermark.

Learn more about SnapToolz