About TinyToolkit
TinyToolkit is a network of small websites, one per file format, plus this index. Every tool runs entirely in your browser — no upload, no queue, no account. This page is the directory: search here, then work on the site that owns your format.
Where it came from
Almost every tool here started as a task I did not want to do twice. Convert a directory of screenshots. Get a PDF under a five-megabyte mail limit. Check that a fine-tuning file was actually in the shape a provider wanted before paying to train on it. The available options were the same three every time: install something for a five-minute job, write a throwaway script, or upload a file I would rather not upload to a site covered in ads.
Browsers can do all of this now. Canvas can re-encode images, File and
ArrayBuffer can read any binary format, WebAssembly covers the rest. Once the
processing happens locally, the usual constraints disappear — no size limits, no rate limits,
no privacy trade — and the tool can be a static page that loads in under a second.
Why it is split into 12 sites
An all-in-one converter has to treat every format the same way, so it ends up handling forty of them adequately. Splitting by format lets each site be opinionated about its own:
- PNG and other images — Batch by default, and it goes deeper than the pixels — chunk viewer, PNG validator, perceptual-hash duplicate finder, real pHYs DPI writing.
- JSON — Repair and type generation. Most JSON sites stop at pretty-printing; this one fixes trailing commas, single quotes and unquoted keys, then hands you a typed struct.
- TSV — TSV-native repair. Because there is no quoting to fall back on, ragged-row detection and column re-alignment are the whole game.
- URLs and links — The parse-and-audit half. Redirect maps, tracking-parameter stripping and phishing checks, not just an encoder box.
- CSV — Real .xlsx read and write, and SQL over a CSV — no server, no row cap beyond your own memory.
- JSONL / NDJSON — The dataset tools. Chat-format conversion, token counting, leakage detection between train and eval, k-fold splitting — built for people preparing LLM training data.
- PDF — The document you are merging is a contract or a payslip. It stays in the tab, which is the whole reason this kit exists.
- TOML — Round-trip conversion that survives comments and table layout, plus a diff built for config review.
- JPG photos — Target-file-size compression and real EXIF/GPS removal, which is a privacy fix disguised as an image tool.
- XML — XPath and XSLT testing against your own file, in a browser tab, with no Java toolchain to install.
- GeoJSON — A map preview that runs locally. Location data is the kind you least want to paste into a stranger's server.
- cron schedules — Thirteen cron dialects modelled rule by rule, including the day-of-month OR day-of-week trap that differs between Linux and Kubernetes.
The cost of the split is having to know which site to open, which is what this index exists to solve.
How the tools work without a server
Each site is static HTML, CSS and JavaScript. When you drop a file, it is read locally and processed in memory; the result comes back as a download generated by your own browser. A few tools pull a library from a CDN on demand — pdf-lib and pdf.js for PDFs, DuckDB-WASM for SQL over a dataset — but the library is code coming to you, not your file going anywhere.
The practical test: load a tool, turn off your network, and use it anyway. It works. Several of the harder pieces are written by hand for exactly that reason — the XLSX writer, the animated GIF and APNG encoders and the MinHash near-duplicate index are all dependency-free implementations rather than server calls.
What this is not
- Not a SaaS. Nothing to sign up for, no plan, no seats.
- Not a batch API. These are hand-driven tools. For automation, two of the kits have CLI companions on GitHub.
- Not a place to store anything. Reload the page and your work is gone. That is the design, not an oversight.
- Not audited or warranted. Check anything load-bearing before you rely on it — see the terms.
Who builds it
Made by Parallax — a small independent shop. One person writes the tools; the sites are funded by nothing in particular, which is why they have no ads and no upsell. Bug reports and tool requests genuinely set the roadmap, so if something is missing or wrong, tell me.
— S., [email protected]