IELTS Prep (BandReady)

At a glance

IELTS Prep, also called BandReady, is a free desktop app for practising all four IELTS skills on your own computer. It marks writing and speaking against band descriptors using an AI model you connect, either a local Ollama server or a cloud endpoint, and includes a live voice examiner you talk to. The practice material is original, not official IELTS content, and there is no account.

Good fit

Self-studying IELTS candidates who are comfortable in a terminal and want timed mock tests with repeated feedback and no subscription. Reading and Listening are auto-marked using exam answer rules. Writing gets inline error notes and an on-demand model answer. Mistakes from writing and speaking feed a spaced-repetition vocabulary bank. With all models local, essays and recordings stay on your disk.

Consider something else when

Skip it if you want to install it and start practising in minutes, or if you need a score you can treat as a prediction. There is no downloadable installer, so you build the app yourself. The project says its band scores have never been calibrated against expert-marked samples and nobody has independently checked its answer keys. A tutor or official practice material fits that need better.

Start here

Install Node.js 20 or newer, pnpm 9, Python 3.11 and uv. Clone the repository. It lands in a folder named ielts-prep, although the README's next line says cd bandready. Run pnpm install, then uv sync --extra dev --extra voice inside the sidecar folder, then node scripts/dev.mjs --no-mock so the app uses your real providers instead of mocks. Under Settings, Providers, pick a model and press Verify. The placement test then builds a study plan.

Cost and requirements

The software is MIT-licensed, the practice content is CC0, and nothing is metered. Fully local use needs no API key. A cloud provider such as OpenRouter bills you directly, and the FAQ cites roughly $0.0002 per essay on a mid-size model. Live speaking adds a 2 to 3 GB Python download. First launch fetches about 340 MB of voice weights, and local speech-to-text needs more.

Limits to know

The project lists what is missing: code signing, any Windows or Linux build a human has run, automated tests of the live voice call, score calibration and an answer-key check. Pronunciation feedback only works with local Whisper, because cloud transcripts carry no per-word confidence, and it flags words that were hard to make out rather than scoring accent. A configured cloud model receives the essay or transcript being marked. Packaged builds check GitHub Releases for updates every six hours.

Still unclear

• What computer a local model needs to give useful examiner and marking feedback. The README suggests qwen3:14b as the easy local start but does not state the memory or speed required.