Opsphere
← All articles

Codex Download: Install the App or CLI, Then Add Production Context

Codex Download: Install the App or CLI, Then Add Production Context

Repository and production context bridged by a teal plane

Codex is available as a desktop app, a command-line tool (CLI), an IDE extension and a cloud service, and you download the first two from OpenAI. This guide covers the official download paths, the system requirements, the install commands and the setup choices that matter for DevOps, SRE and platform teams. It also covers the step most guides skip: giving Codex read-only access to production context, so it can help investigate failures instead of guessing from source code alone.

Opsphere
Give Codex Production Context
Opsphere connects Codex to read-only operational context across cloud and engineering tools, helping teams investigate incidents beyond the repository.

TL;DR

  • Problem: Downloading Codex takes minutes, but an agent that sees only your repository cannot explain a production incident.
  • Desktop app: Get it from the official ChatGPT download page. It combines Chat, Work and Codex, and is documented for macOS, Windows and Linux.
  • CLI: Install with the standalone script, npm or Homebrew. Pin versions in CI and shared environments.
  • Next step: After installing, connect Opsphere through MCP so Codex gets read-only operational context. See the Codex integration page.

Table of Contents

Which Codex download do you need?

Search results for this keyword mix several products, and some of them are outdated or unofficial. Pick the surface first, then download from OpenAI only.

Surface Best for Where it comes from
Desktop app Parallel tasks, reviewing diffs, a visual workspace ChatGPT desktop app download
CLI Terminal workflows, scripting, remote shells, CI Install script, npm or Homebrew
IDE extension Staying inside VS Code, Cursor or Windsurf Your editor’s extension marketplace

The OpenAI documentation for the desktop app says the ChatGPT app for macOS, Windows or Linux is the install step, then you sign in and choose Codex when you send a message. Older guides describe a macOS-only Codex app. That changed: OpenAI’s announcement notes that the Codex app became available on Windows on March 4, 2026.

A caution on third-party mirrors. Some ranking pages host installers or unofficial ports, such as community builds for Linux. For a work machine with cloud credentials on it, install from OpenAI’s own pages or from a package registry you already trust.

Download the Codex desktop app

The desktop app is the fastest way to start if you want parallel agent tasks and a review interface.

  1. Open the official ChatGPT download page and choose your operating system.
  2. Install and sign in with your ChatGPT account.
  3. Choose a project folder or open a repository.
  4. Select Codex when you start a task.

OpenAI’s help center states that the desktop app now combines Chat, Work and Codex, and lists the compatible Mac hardware and software. Check that page for your exact macOS version before installing, since requirements change between releases.

Install the Codex CLI

The CLI runs locally and is the natural choice when your debugging happens in a terminal, over SSH or next to kubectl. The official CLI guide lists these install methods:

macOS and Linux (standalone installer)

curl -fsSL https://chatgpt.com/codex/install.sh | sh

Windows (PowerShell)

powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"

npm

npm install -g @openai/codex

Homebrew

brew install --cask codex

The npm package page shows the current published version and notes that the CLI runs locally on your computer. Releases move quickly: the GitHub releases page lists new alpha builds almost daily, so “latest” changes under you.

After installing, run codex from a project directory. With no subcommand it launches the interactive terminal UI, as the command reference describes.

System requirements

The install documentation on GitHub lists supported systems as macOS 12+, Ubuntu 20.04+ or Debian 10+, and Windows 11 through WSL2. It also recommends at least 4 GB of RAM, with 8 GB preferred. If you run the CLI on a shared bastion or a small cloud workstation, size it accordingly.

What does Codex cost?

There is no separate download fee. Codex usage is included in ChatGPT plans, and OpenAI’s Codex page lists Plus at $20 and Pro at $100, with Business and Enterprise options that add admin controls. Plans and limits change, so confirm on that page before you budget.

Standardize the install for a team

One engineer running the install script is fine. Twenty engineers and a CI pipeline need consistency.

  • Pin versions. The install documentation recommends pinning specific releases from GitHub rather than relying on the generic script when you need reproducible environments. Record the version in your onboarding doc or image.
  • Share a config. Codex reads ~/.codex/config.toml. Platform teams can use it to set defaults for permissions, proxy settings and agent behavior, as the CLI reference covers.
  • Start restrictive. The same reference describes sandbox and approval flags. For low-friction local work it points to --sandbox workspace-write with approval settings. Keep approvals on while engineers learn what the agent does.
  • Decide on proxies and egress early. If production networks restrict outbound traffic, find that out before the first rollout, not during an incident.

The gap after the download: Codex cannot see production

Codex reads your repository well. During an incident, though, the answer is rarely in the repository alone. It is in the combination of a failing deployment, a Kubernetes event, a latency graph, an error spike and a recent merge.

Without that context, a developer pastes logs and screenshots into the prompt by hand. That is slow and incomplete, and it leaves the agent reasoning from whatever you remembered to copy.

Opsphere is the operational intelligence layer that closes this gap. It does not replace Codex. Codex remains where you write, review and reason about code. Opsphere brings live production context into it: cloud infrastructure, Kubernetes, observability, deployments and code changes, available through the Model Context Protocol (MCP).

Production systems supplying context to Codex through MCP

How Opsphere connects to Codex

MCP is an open protocol that lets an AI agent call external tools in a structured way. Opsphere exposes its operational tools through an MCP operational gateway, and Codex connects to it as a client. The full setup is on the Codex integration page.

Two properties matter for security teams:

  • Read-only by default. Opsphere gives Codex controlled access to production context without allowing direct infrastructure changes. The agent can look, query and correlate. It cannot scale a deployment, edit a security group or restart a pod.
  • Your tools stay the source of truth. Opsphere connects to the tools you already run and does not require replacing them or duplicating telemetry.

The design questions behind this are the same ones covered in material on agent tool calling patterns: what the agent may call, what each call returns and where the boundaries sit. Read-only tools with typed, scoped responses are the safest answer for production.

For details on how access is handled, see Opsphere’s security page.

Practical workflows from the coding environment

These are the workflows teams run once Codex has operational context. Each one stays inside the terminal or editor.

1. Investigate a production failure

You get an alert that checkout latency is up. Instead of switching between consoles, you ask Codex to investigate the service.

With Opsphere connected, the agent can pull the service’s dependencies, current pod state and recent events. It can compare those signals against the code it already has open. You get an evidence trail, not a guess: which signals it checked and what each showed.

2. Check recent changes first

Most incidents follow a change. A useful first prompt is: what changed in this service and its dependencies in the last few hours?

Codex can combine the commit history in your repository with deployment and infrastructure changes surfaced by Opsphere. The answer points at a candidate cause, with timestamps you can verify.

3. Correlate application errors with infrastructure signals

An exception in your logs may come from the code, or from a node under memory pressure, a throttled database or an expired credential. Codex sees the stack trace. Opsphere supplies the infrastructure side: Kubernetes state, AWS resource health and observability data.

Together they let the agent separate “this code path is broken” from “the platform under it is unhealthy”.

4. Debug with live context, then fix in code

Once the cause is clear, the fix often lives in the repository: a missing timeout, a bad retry loop, a resource limit in a manifest. Codex drafts the change, and you review it through your normal pull request process. Opsphere stays out of the write path, so the change goes through the same CI/CD and approvals as any other.

5. Test several hypotheses in parallel

Incident work often stalls on one theory. Opsphere supports structured, evidence-backed investigation with parallel hypothesis testing. From Codex, you can ask for the three most likely causes and the evidence for and against each, rather than chasing the first plausible one.

Three incident hypotheses tested against evidence

Example session: a failing rollout

Here is an illustrative flow, not a customer story.

  1. A deployment finishes and error rates on one service climb.
  2. You open Codex in the service repository and ask it to compare the failing pods with the last healthy revision.
  3. Through Opsphere, it reads pod status, restart counts and events, and the deployment timeline.
  4. It finds that the new revision lowered a memory limit while a commit raised batch size, and points to both the manifest line and the commit.
  5. You ask Codex to propose a fix in code and manifests. It opens a diff. You review and merge as usual.

No one granted the agent write access to the cluster at any point.

Decision criteria: what to check before rolling out

Before you give any AI agent production context, work through this list:

  • Access scope. Is every tool read-only? Can you see which tools exist and restrict them per team?
  • Auditability. Can you review what the agent queried during an incident?
  • Tool fit. Does the layer connect to your existing observability, CI/CD and cloud stack, or does it ask you to add another agent?
  • Model choice. Can you bring your own model or agent? Opsphere supports this on its Bring Your Own AI page.
  • Noise. Does the context the agent receives come correlated and curated, or is it raw alerts?
  • Change path. Do fixes still flow through pull requests and approvals?

Measure results with numbers your team already tracks: time to first useful hypothesis, time to resolution and the number of consoles opened per incident. Capture a baseline before the rollout.

Opsphere plans for teams using Codex

Opsphere is a subscription with tiered monthly plans. The published prices are Community at 0 € per month, Developer at 19 € per month, Team at 500 € per month and Enterprise at 450 € per month. Custom Solutions has no published price. Details and current terms are on the pricing page.

If you already use Codex and want production context in the same workflow, START FREE TRIAL or BOOK A 20-MIN DEMO from the Opsphere site. To see how the pieces fit for AI-assisted teams, read the use case for AI-native engineering teams.

FAQ

Is Codex free to download?

The download itself is free. Using Codex requires a ChatGPT plan, and OpenAI states that Codex usage is included in Plus, Pro, Business, Enterprise and Edu plans within each plan’s limits. Check OpenAI’s pricing page for current limits.

Is there a Codex download for Windows and Linux?

Yes, in different forms. The desktop app has been available on Windows since March 2026, and OpenAI’s docs say the ChatGPT desktop app is offered for macOS, Windows and Linux. The CLI runs on macOS, Linux and Windows, with Windows 11 supported through WSL2 per the install documentation.

Should I install the app or the CLI?

Choose the app for parallel tasks and visual review. Choose the CLI for terminal-centered work, SSH sessions and scripting. Many teams install both, since they share the same account and usage.

Can Codex change my production infrastructure?

That depends on what you connect and how you set permissions. With Opsphere, access is read-only by default, so Codex gets production context without the ability to change infrastructure directly. Code changes still go through your repository and review process.

Sources

Opsphere
Discuss Your Codex Setup
Contact Opsphere to discuss connecting Codex with operational context across your cloud, observability, CI/CD, security and engineering tools.

This article is provided for general informational purposes only and does not constitute professional, legal, security, or compliance advice. Please evaluate recommendations against your organization’s specific environment and requirements.