Set up a local math tutor with Codex

They built ari, an open-source tutor app that uses Codex on the user’s machine to coach pen-and-paper math.

Done with Codex· Intermediate· Oct 2026

What they did
The author created ari and shared the project publicly. Instead of calling an external API, it runs Codex locally, so it relies on the user’s own subscription access. They also added a TUTOR.md brief, adapted from the SOUL.md approach in openclaw, to tell the agent how to behave as a math teacher.
How it went
The author said the setup was fun to use and that making a similar version is easy. No measurements, hard outcomes, or specific drawbacks were mentioned.
Tips
A local setup can lean on the user’s subscription access, so it avoids separate API billing. For a similar tutor, keep the interaction simple: one question, then the learner’s attempt, then hints or feedback.

Try it yourself with Codex

Create a local math tutor in [folder] for [topic] at [level]. Write a TUTOR.md brief that instructs Codex to act as a pen-and-paper coach: ask one problem at a time, wait for me to write or describe my steps, and give hints before showing the full solution. Before any action that would remove files, replace files, send data, or finalize something, pause and ask me to approve it first. When the setup is finished, open a session with an initial problem and wait for my reply.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.