Every tool on this site that runs directly in the browser. Each has its own page setting out input format, what the output means, the privacy boundary, and the typical ways it fails.
Image tools never upload your files. Decoding, computation and export all happen in your own browser. Previews and download links exist only in the current tab and vanish on reload.
|
VRAM calculator Estimate whether a local LLM fits, from GPU memory, context length and concurrency. Formula calibrated on real hardware. |
Open |
|
dQ/dV incremental capacity analyser Turn battery cycling data into an incremental capacity curve, locate phase-transition peaks, identify the degradation mode. Nothing uploaded. |
Open |
|
HPPC pulse resistance analyser Extract R0, the RC pair and the time constant from pulse data – the whole R(t) curve, not one DCIR figure. |
Open |
|
Image to ANSI art Convert an image into terminal-ready coloured character art using Unicode quadrant blocks and 24-bit truecolour. |
Open |
|
IP reputation scorer See the score this site computes for your address and the weight of each signal – also the anti-abuse documentation. |
Open |
|
Soft mask threshold lab Drag the binarisation cut and the noise, and watch a segmentation mask shatter – and why area metrics cannot see it. |
Open |
|
Animation asset generator Matte, layer and vectorise a character image into SVG animation assets, entirely on your machine. |
Open |
|
Algorithm playground Step-by-step visual execution of classic sorting, searching and graph algorithms. |
Open |
|
Handwritten digit playground Draw a digit on the canvas and watch layer activations and class probabilities update live. |
Open |
|
LLM inference visualiser Step through attention weights and token generation. |
Open |
|
2D animation principles Interactive demos of bone matrices, two-bone IK, skinning weights and keyframe easing. |
Open |
|
Knowledge map Browse how articles on this site depend on and extend one another, by topic. |
Open |
|
Learning index An ordered route from fundamentals through to the specialised series. |
Open |
Tools and articles are maintained in pairs
A tool page handles use, preview and export; the companion article explains the principle, the implementation and the limits. Where no companion exists, the tool page itself must still state the input format, what the output means, the privacy boundary and the failure modes — so that when a result surprises you, you can tell whether it was a parameter, a browser limitation, or an assumption in the algorithm that does not hold.
- Image to Unicode quadrant ANSI: principle and implementation: why a terminal cell carries only two colours, and how the converter picks the lowest-error mask among 16 quadrant patterns.
- Sampling bias in quantisation validation: the measured account behind the threshold lab — int8 dropping to 0.30 IoU out of distribution.
- Character image to SVG animation asset pipeline: IP-Adapter identity locking, OpenPose exploded layout, Grounded-SAM matting, upscaling and offline vtracer assembly.
- Every Gemma 4 variant explained: the companion to the VRAM calculator, covering architectural differences and the quantisation size trap.
- Downloads: public code, datasets and companion resources.
What earns a place here
New tools will keep favouring small tasks suited to running locally in a browser: format conversion, visual aids, experiment parameter previews. Anything touching accounts, mail, surveys or server state gets its own explanation of how the data is handled, rather than dressing an interactive feature up as a technical tutorial.
Whether a tool stays comes down to three questions: does it solve one clearly defined small problem, can it explain why the output looks the way it does, and does it leave a diagnosable trail when it fails. While any of those is unclear, the answer is better documentation rather than more features.