The anti-homework homework tool: an ergonomic LLM wrapper

What it does

Skid-Homework is a Next.js web app that acts as a specialized chat interface for solving homework. You upload images or PDFs of problems, pipe them to a Gemini or OpenAI-compatible API, and get answers back. The entire thing runs in the browser, with no phone number, no telemetry, and no desktop install required.

The interesting bit

The project treats homework as a user-experience problem rather than an AI research problem. It leans heavily into keyboard shortcuts for uploading, submitting, and navigating problems, and even includes experimental WebUSB ADB support so you can screenshot questions from an Android phone directly into the browser. The README also candidly admits the tool is a “workaround” for excessive busywork, which is a refreshingly honest stance for this category of software.

Key highlights

• Keyboard-first workflow with customizable shortcuts and answer styles via a global prompt editor.
• Supports multi-image and PDF parsing, with optional local persistence for parsed data.
• Can render diagrams via Mermaid, SVG, and an unstable JSXGraph integration.
• Left-handed friendly and e-ink display compatible (light mode only).
• No telemetry, no ads, and no account required beyond bringing your own API key.

Caveats

• Requires your own Gemini or OpenAI-compatible API key; the app is just a frontend.
• LLM-based solving is slower than traditional question-bank apps, and the README admits there is “no good optimization method” for latency.
• Several features are explicitly marked experimental, including local persistence, ADB screenshot support, and JSXGraph rendering.

Verdict

Students drowning in repetitive problem sets who already have an API key and value speed over pedagogical depth might find this a pragmatic time-saver. Teachers and anyone hoping for guaranteed correct, tutor-style explanations should look elsewhere—this is a browser wrapper for an LLM, not a curated learning platform.