What You Need to Get Started

LLM Grader is deliberately lightweight. Before you start, it is worth knowing how little you actually need – and, just as important, what you do not need.

To Start Authoring

  • A GitHub account. You will fork the llmgrader repository and keep your own course material in a repository of your own. A free personal account is enough.
  • A laptop – any laptop. You do not need a server, a GPU, or a powerful machine. Windows, macOS, and Linux are all fine. All of the heavy lifting is done by the model provider’s API, not by your computer.
  • Python 3.12 (or a nearby compatible version) to install the llmgrader package locally. See Installing the Python Package.
  • An OpenAI API account and key. Grading calls go to the OpenAI API platform, so you need an account with billing enabled and an API key to try your questions and run the grading tests while you build the course. This is a developer account on platform.openai.com – a ChatGPT subscription is a different thing and does not give you API access. Costs are small – on the order of a few tenths of a cent per graded question.
  • VS Code. Course content is written as XML, and you will spend most of your time editing structured text. Any editor will work, but the documentation and workflows here assume VS Code. See Selecting an IDE.
  • An AI coding agent of your choice inside VS Code – GitHub Copilot, Claude Code, Cursor, or similar. LLM Grader ships an MCP server that lets an agent scan your existing course materials, draft question and rubric XML, and validate it for you. This is by far the fastest way to turn homework you already have into a course package.

When You Are Ready to Deploy

None of these are needed while you are authoring and testing on your own machine.

  • A Render account. Render builds and hosts the app directly from your GitHub fork, with a persistent disk for course packages. You create the account by linking it to GitHub. See Deploying on Render.
  • A Google Cloud project with OAuth 2.0 credentials. Sign-in and admin access use Google OAuth. A normal Google account is enough to create the project and the client ID/secret – you do not need a separate account. See Setting up Google OAuth.
  • A Gradescope instructor account, if you want student submissions to flow into Gradescope for the record grade. See Gradescope Integration.
  • Submission signing keys, only if you want the Gradescope autograder to reject hand-edited submission files. These are generated by a script, not bought or registered. See Submission Signing Keys.

What You Do Not Need

  • A server of your own. You can run the whole grader on your laptop while you author, and Render handles hosting when students need it.
  • API keys for your students. Each student supplies their own OpenAI key in the portal. Keys stay in the student’s browser and are never stored on the server, so you are not paying for – or holding – anyone else’s credentials.
  • To replace your existing LMS or Gradescope setup. LLM Grader gives students an immediate try -> grade -> improve loop; the grade of record still flows through Gradescope and your LMS.
  • To rewrite your assignments. The authoring workflow starts from the homework and solutions you already have.

Next Step

Once you have the items in the first two sections, continue to Setting Up LLM Grader.

This site uses Just the Docs, a documentation theme for Jekyll.