OpenPrompt for Codex

How Codex prompts are captured

Codex keeps a chat log in each project folder. OpenPrompt reads those on-disk transcripts and files every prompt under the project that owns the folder. No Accessibility permission is needed.

Each project owns one or more folders, and the longest matching folder wins, so a repo nested inside another still files correctly. Prompts from folders no project owns are discarded.

Codex capture can be turned off on its own under Monitor my IDEs.

What you get for every prompt

• One thread for every tool, with code blocks and file paths kept as they were typed. Filter by source or favourites, or search.
• The Prompt Time Machine: open a capture to see the exact lines that prompt changed, then copy it, revert the whole turn, discard one file, or open its diff.
• A project context file at .openprompt/OpenPrompt.context.mdc, refreshed automatically with your instructions, recent prompts and the files they changed.
• Commits straight from the diff, with a generated commit message.

Codex as an agent

Connect Codex in Agent Mode to plan and run tasks. In a routed pipeline a connected CLI plans the work, then each sub-task runs on an agent; expand a step to read what it changed, or stop the run from the composer.

Agent CLIs run under your own accounts and licences. What they send, retain and charge for is governed by that vendor, not by OpenPrompt.

Set it up

1. Download OpenPrompt and sign in.
2. Create a project and add the folders you use with Codex.
3. Check that Codex is on under Monitor my IDEs, and connect the Codex CLI under Agent CLIs if you want to run it from OpenPrompt.

Where the data lives

Captured prompts stay on your machine, in SQLite at ~/.openprompt/prompts.db. Credential-shaped files (env files, .pem and .key files, anything named credentials or containing "secret") are never indexed. You can pause any project from its ⋯ menu in the sidebar, and the privacy policy lists exactly what is sent and when.