Persistent learning memory and progress tracking for Codex Live voice across sessions and devices

Open 💬 0 comments Opened Jul 30, 2026 by chentinghui

What variant of Codex are you using?

Codex App — Live voice on desktop and mobile.

What feature would you like to see?

I would like Codex Live voice to support persistent, user-controlled learning memory so that it can act as a long-term language teacher across sessions and devices.

My use case

I want to use Codex Live voice to learn a new language through regular spoken lessons.

For this to work well, Codex needs to remember information such as:

  • The language I am learning
  • My current proficiency level
  • Vocabulary, grammar, and topics I have already studied
  • Mistakes I frequently make
  • Words or concepts that need review
  • My lesson goals and preferred teaching style
  • Where the previous lesson ended

Current problem

When I close Codex Live voice and start it again later, the new session does not reliably know my previous learning progress.

I have to explain my level and learning history again, and Codex may start teaching material that I have already learned. This makes it difficult to use Codex as a consistent long-term teacher.

I tried using an Obsidian vault to store my learning notes and progress. This can work on a desktop where Codex has access to the local vault, but it does not solve the mobile use case. On my phone, Codex cannot reliably access or update the same Obsidian repository, so it still does not know what I have learned.

Proposed behavior

Codex could provide a persistent learning profile or project memory with the following capabilities:

  1. At the end of each lesson, Codex saves a short, structured progress record.
  2. When the next voice lesson begins, Codex automatically loads the relevant learning history.
  3. The learning profile is synchronized across desktop and mobile through the user’s Codex account.
  4. Users can view, correct, export, delete, or disable the stored learning memory.
  5. Codex stores structured learning progress instead of relying on an entire conversation transcript.
  6. External tools such as Obsidian can optionally import or export the same progress, but they should not be required for basic continuity.
  7. Codex can generate spaced-review lessons based on previous mistakes and learned material.

An example progress record could include:

Target language: English
Current level: A2
Completed topics: greetings, daily routines, simple past tense
Vocabulary learned: 85 words
Frequent mistakes: verb tense and article usage
Next lesson: review the simple past, then practise ordering food
Preferred teaching style: spoken practice with short Chinese explanations

Expected experience

I should be able to finish a lesson on my computer, close Codex, and continue the next day on my phone.

Codex should be able to say something like:

Last time, we practised the simple past tense. You had difficulty choosing between “did” and the past-tense verb form. Let’s review that briefly before continuing with today’s lesson.

This would make Codex Live voice much more useful as a long-term personal tutor rather than a teacher who starts over in every new session.

Additional information

This is related to the broader cross-session context problem discussed in #24810, but this request focuses specifically on:

  • Structured learning progress
  • Live voice tutoring
  • Continuity across desktop and mobile
  • User-controlled and privacy-conscious memory
  • Optional integration with external knowledge tools such as Obsidian

If Codex already has a recommended way to achieve this workflow, documentation or an official example would also be helpful.

View original on GitHub ↗