Choose Codex directly when the session needs its sandbox and approvals.
Reviewing a local diff, trying disposable edits, and running commands with the network off belong in a live Codex session, with you approving each step.
ProPR and Codex CLI: Codex as a worker in your PR system
Codex CLI is OpenAI's Apache 2.0 agent for exploring, editing, running, and reviewing code from a sandboxed terminal. ProPR runs Codex unattended: you approve a plan, Codex implements in isolation, and the result arrives as a pull request your team reviews in GitHub.
vs
Codex CLI
Sandboxed session, or headless worker
Codex owns the sandboxed terminal session, its approvals, and its code review. ProPR turns a scoped task into a reviewed pull request on your ChatGPT plan, with every run on record.
Codex CLI path
Prompt → sandboxed terminal session → code, tests, or review
You steer the live session and decide how its result moves into the repository workflow.
ProPR path
Review plan → isolated Codex run → GitHub PR → follow-up commit
Approve the plan, come back to a pull request. Comment on it to get the next commit.
Codex CLI is a full local coding agent with an interactive terminal UI, a plan mode, configurable approvals, operating-system sandboxing with network access off by default, code review, and a non-interactive execution mode. It runs OpenAI models on a ChatGPT plan or API key, and its provider configuration also reaches Amazon Bedrock, Google Cloud, Ollama, and LM Studio. Direct use gives you a tight loop for investigation, implementation, testing, review, and command-level control. The GitHub side of Codex lives in Codex Cloud: @codex review on a pull request posts a review, and other @codex comments start a cloud chat that can push a fix to the branch.
Codex inside ProPR implements on your ChatGPT plan or OpenAI API key. You approve the plan, Codex works in an isolated workspace on your server, and ProPR opens the pull request with the full run attached.
Codex's own comment flow, @codex on a pull request, runs in Codex Cloud. Inside ProPR the same request is a plain comment, answered on your server by Codex or by any of the other four agents, with each run in the task history. Codex also runs goals inside ProPR with live steering.
A sandboxed Codex session versus Codex as a ProPR worker.
| Question | Codex CLI | ProPR with Codex |
|---|---|---|
| Interactive terminal session | Yes | No, unattended runs |
| Open source | Apache 2.0 | Apache 2.0 |
| Plan before implementation | Plan mode in the session | Reviewed plan on every task |
| Isolation for a run | OS sandbox, network off by default | Isolated workspace per task |
| Branch, commit, push, and PR after implementation | Agent-driven or scripted | Every task |
| Shared task history and logs | Local session state | Yes |
| A PR comment starts the next commit | @codex comments run in Codex Cloud | Any plain comment |
| Claude, Gemini, Mistral, or OpenCode on the same path | No, Codex only | Five supported agents |
Choose ProPR when Codex tasks should come back as pull requests.
Run several Codex tasks at once, each ending in a PR. A second model reviews it, plain comments drive the fixes, and a Claude or Gemini run can pick up the same branch.
Codex Cloud
OpenAI-hosted tasks started from ChatGPT, with @codex review and fixes on the pull request.
Run Codex inside ProPR
Codex implements, reviews, or follows up on the ChatGPT plan or API key you already hold.
Verified November 2026: github.com/openai/codex, Codex CLI docs, sandboxing, config reference, GitHub integration, Codex pricing.
Use Codex CLI with ProPR.
Stays in Codex CLI
- Sandboxed terminal sessions with approvals, network off, and local code review.
- Disposable edits and plan mode for changes you steer yourself.
- Your ChatGPT plan or OpenAI API key; ProPR signs in with the same account.
Moves to ProPR
- Scoped tasks that should come back as a pull request, with the plan approved first.
- Follow-up commits from plain GitHub comments on your server, with no Codex Cloud chat.
- A second-model review, task history, logs, and usage per run.
Wire them together
Sign in Codex inside ProPR
Install ProPR (quickstart) or pick a ProPR Hosted plan, add the repository, and sign in Codex with your ChatGPT plan or API key.
Keep the terminal for thinking, send tasks to ProPR
Investigate in a live session as before. When the change has a shape, write it as a GitHub issue and add the AI label, or run propr issue implement; ProPR plans it, runs Codex in isolation, and opens the pull request.
Steer from the pull request
A plain comment becomes the next commit; /review and /fix handle AI findings, and a Claude or Gemini run can pick up the same branch. The CLI playbook drives the loop from the terminal.