Deploying the App on a Local Machine

Before deploying the app on a public portal, you should test it on your own machine. Running locally needs far less setup than a Render deployment – in particular, you do not need Google OAuth credentials to run locally.

Running the Student Portal

Navigate to the llmgrader repo, activate your virtual environment, and run:

python run.py --soln_pkg <soln_pkg>

where <soln_pkg> is the path to the solution package that was created following the packaging instructions. Then open http://127.0.0.1:5000/ in your browser.

The --soln_pkg flag loads the course directly from disk, so you can skip the admin upload step entirely while you are authoring. To try this immediately with the bundled example course:

python run.py --soln_pkg example_repo/soln_package

The student-facing portal – browsing units, answering questions, and grading – works with no environment variables set at all. Students supply their own OpenAI key in the browser, so nothing needs to be configured on the server.

Reaching the Admin View Locally

The Admin view (loading course packages, the database viewer) is protected by login. For local work you have two options:

Option 1: Development bypass (recommended for local use)

Set one environment variable before starting the app:

# macOS / Linux
export LLMGRADER_AUTH_MODE=dev-open

# Windows PowerShell
$env:LLMGRADER_AUTH_MODE = "dev-open"

This grants admin access without any Google configuration. It is intended for local development only – never set it on a public deployment.

Option 2: Full Google OAuth

If you want to exercise the real sign-in flow locally, follow the Google OAuth setup and register http://127.0.0.1:5000/auth/callback as an authorized redirect URI. This is required for Render, but optional for local testing.

---

Next: Deploying on Render