This repo now has two layers:
docs/: public GitHub Pages prototypeapp/: full-stack local MVP for real file parsing and AI review
The GitHub Pages site publishes the static prototype in docs/.
The local app reads drafts, loads private templates from templates/, applies the house rules in references/, and returns a review UI with redlines plus margin comments.
- Open a terminal in
app/ - Install dependencies with
npm install - Copy
.env.exampleto.env - Add
OPENAI_API_KEY - Set
OPENAI_MODELif you want to override the default - Optionally set
OPENAI_BASE_URLif you want to use an OpenAI-compatible non-OpenAI endpoint - Run
npm run dev - Open
http://localhost:3000
Without OPENAI_API_KEY, the app falls back to a local mock review mode so you can still test the workflow.
Keep approved sample statements and incoming drafts out of the public site:
templates/incoming/outputs/
These folders are already ignored by Git.
The app defaults to gpt-5-mini for lower latency and cost, but you can override OPENAI_MODEL in app/.env. OpenAI's current model guidance says to start with gpt-5.4 if you're not sure and use gpt-5-mini when you want lower latency and cost.