Web Design

Decode
Local AI, Measured

A calculator and field guide for running AI models on your own machine. Will it fit, and how fast will it run? Arithmetic, not opinion.

Home — Hardware in Wireframe “Will that model run on your machine?”
Client Decode
Industry Technology / Artificial Intelligence
Services UX/UI Design
Web Development
Data Visualization
Deliverables Interactive Calculator
Illustrated Guide
Model Database
Lab Try the Calculator

“Memory is a room. Bandwidth is the door.”

Running a language model on your own computer comes down to two questions: does it fit in memory, and how fast will it answer? Most answers online are opinion. Decode answers with arithmetic. Pick a machine and a model, and the calculator lays out the memory budget — weights, cache, runtime, and what’s left free — then estimates decode speed as an honest range.

We designed the whole site around one metaphor and drew it literally: memory as a room, bandwidth as a door, and tokens as small marks passing through it. The same live diagram appears in the calculator, the guide, and the model pages, so every explanation doubles as a working tool. A restrained dark interface, monospaced figures, and a single warm accent for model weights keep dense technical data calm and readable.

Decode website — the calculator, showing a model that fits with memory and speed estimates
01 The Calculator Pick a machine and a model. The verdict comes first, then the memory budget and the speed range behind it.
Decode website — How Local Inference Works guide page with a live diagram
02 The Guide A plain-language guide with a live diagram beside every chapter, wired to the same arithmetic as the calculator.
Decode website — guide chapter on agent harnesses with an animated tool-call diagram
03 Agents Step-through diagrams show how an agent works, next to a live chart of which models can run one.
Decode website — the Models database table with filters
04 Models Every model in the database, with architecture read from its published config and a filter for what fits your machine.

Outcomes: A technical subject made legible — one metaphor carried through every page, a calculator people can trust, and a guide that teaches by letting readers change the numbers themselves.

← Back to Work