Codex Alternatives: Cloud and Terminal Agents Compared

The right Codex alternative depends on which half of Codex you use. If you mostly hand tasks to Codex cloud and review a diff later, the closest alternatives are other cloud agents: Claude Code cloud sessions, Google Jules, the GitHub Copilot cloud agent, Devin and Cursor cloud agents. If you mostly run Codex CLI in a terminal, look at Claude Code, Gemini CLI or OpenCode. Pick by where your code lives, which account your company already pays for, and how much review you want before code exists. Whichever you choose, your AGENTS.md, your MCP servers and your task list carry over almost unchanged.

This page is a map by fit: one short paragraph per agent, each linking to a longer head-to-head where one exists. No prices, because plans change often; check each vendor’s current page. Agents compared against Claude Code rather than Codex are in Claude Code alternatives, and every agent against six questions is in the best AI coding agents.

What you are replacing

Codex is one OpenAI agent in several places: a CLI, an IDE extension, the Codex view of the ChatGPT desktop app, and Codex cloud, which runs tasks in OpenAI-managed containers and takes work from the web, GitHub, GitLab (in beta), Linear and Slack. All of them read AGENTS.md. So “an alternative to Codex” usually means one of two jobs: a cloud agent that works while you are away, or a local agent you pair with in a terminal. How the Codex surfaces divide that work is in Codex app vs CLI vs IDE extension, and the cloud side in depth is in Codex cloud.

Why people look beyond Codex

• Account: Codex cloud needs a ChatGPT sign-in. A team that pays for Claude, Google AI, GitHub Copilot or Cursor may prefer an agent that comes with it.
• Where the code lives: some cloud agents work only with GitHub; others also clone from GitLab, Bitbucket or Azure DevOps.
• A plan before code: some agents write a plan you approve first, instead of returning a finished diff.
• Where work starts: an issue, a Slack thread, a Jira ticket or a terminal. The best agent is often the one that listens where your team already asks.
• Model choice: Codex runs OpenAI models. Open-source terminal agents can run models from many providers, or a local one.

Cloud agents you hand a task to

Google Jules

Jules is the nearest match to Codex cloud in shape, with one difference in order. Google’s Jules documentation↗ (opens in a new tab) says it clones your code into a virtual machine, installs dependencies and modifies files, and that you review and approve a plan before any code changes are made. It works on GitHub repositories, reads AGENTS.md, and has a CLI called Jules Tools and a REST API. Google’s own pages still describe it as an experimental coding agent. Suits teams that want to approve the approach before a diff exists. See Google Jules vs Codex.

GitHub Copilot cloud agent

GitHub renamed the Copilot coding agent to the Copilot cloud agent. GitHub’s page about the cloud agent↗ (opens in a new tab) says it runs in an environment powered by GitHub Actions, where it explores your code, makes changes and runs tests and linters, and that it works only with repositories hosted on GitHub. You start it from the agents panel, an issue, VS Code or an @copilot comment on a pull request, and each session has a hard limit of 59 minutes. MCP servers are supported, with the GitHub and Playwright servers on by default. Suits teams whose issues, reviews and permissions already live in GitHub. See OpenAI Codex vs GitHub Copilot and the GitHub Copilot coding agent.

Devin

Cognition’s Devin runs each cloud session on its own machine and takes work from Slack, Microsoft Teams, Linear and Jira as well as its web app, then returns a pull request. The same brand now covers the Devin CLI, a local agent with handoff to the cloud, Devin Review for pull requests, and Devin Desktop, the IDE formerly called Windsurf. Devin reads AGENTS.md and connects to MCP servers through its marketplace or custom entries. Suits teams with a queue of well-scoped tickets and a habit of assigning them in a tracker. See Devin vs Claude Code.

Cursor cloud agents

What Cursor used to call background agents are now cloud agents. Cursor’s cloud agents documentation↗ (opens in a new tab) says they run in isolated virtual machines with full development environments, clone from GitHub, GitLab, Bitbucket Cloud or Azure DevOps, and can be started from Cursor on the desktop, the web, the iOS app, Slack, Linear, the API or an @cursor comment on GitHub or Bitbucket. They can use the MCP servers configured for your team. Suits teams that already work in Cursor and host code outside GitHub. See Cursor vs Codex and Cursor cloud agents.

Terminal agents on your own machine

Claude Code

Anthropic’s agent covers both jobs. Locally it runs in a terminal, VS Code, JetBrains IDEs and a desktop app, and reads CLAUDE.md, or AGENTS.md when there is no CLAUDE.md. For delegation, Anthropic’s page on cloud sessions↗ (opens in a new tab) describes sessions that run on Anthropic-managed virtual machines and keep going after you close your laptop: claude --cloud "task" starts one from the terminal, and claude --teleport pulls one back. Cloning and pull requests need GitHub, and cloud sessions are available on Pro, Max, Team and Enterprise plans. Suits people who want the same agent for pairing and for handing off. See Claude Code vs Codex and Claude Code on the web.

Gemini CLI

Google’s open-source terminal agent, licensed Apache 2.0, reads GEMINI.md, supports MCP servers and offers sandboxing. Check sign-in before you switch: Google’s Code Assist deprecation notice↗ (opens in a new tab) says that from June 18, 2026, Login with Google no longer works for Gemini CLI on the Code Assist for individuals, Google AI Pro and Google AI Ultra tiers, and points those users to Antigravity. A Gemini API key, Vertex AI, or a Code Assist Standard or Enterprise subscription still work. Suits teams on Google Cloud. See Gemini CLI vs Claude Code vs Codex CLI.

OpenCode

An open-source agent under the MIT license with a terminal interface, a desktop app and an editor extension. It works with many model providers or a local model, reads AGENTS.md from the project root, and connects to MCP servers. Suits people who like how Codex CLI works but want to choose the model, or run one on their own hardware. See what OpenCode is and OpenCode vs Claude Code.

Which one fits

• You want to approve a plan before any code is written: Jules.
• Your issues, reviews and Actions live in GitHub: the Copilot cloud agent.
• Your team assigns tickets in Slack, Linear or Jira and wants pull requests back: Devin.
• Your code is on GitLab, Bitbucket or Azure DevOps and your team uses Cursor: Cursor cloud agents.
• You want one agent for pairing and for handing off, on a Claude plan: Claude Code.
• You are on Google Cloud with a Code Assist Standard or Enterprise license: Gemini CLI.
• You want open source and your own choice of model: OpenCode.

Many teams keep Codex and add a second agent rather than replace it. That works if both read the same instructions and connect to the same tools, which is the next section.

What carries over from Codex

• AGENTS.md: the AGENTS.md format↗ (opens in a new tab) is a plain Markdown file for build and test commands, conventions and no-go areas. Codex reads it, and so do Jules, Devin, OpenCode and Claude Code when there is no CLAUDE.md. Gemini CLI defaults to GEMINI.md, so point it at the shared file. Which tool reads what is in AI context files compared.
• MCP servers: the server URL and its sign-in stay the same; only the config changes. Local agents keep servers in their own config file. Cloud agents keep them in the vendor’s settings for the repository, team or organization, and some allow only servers on a vetted list.
• Environments: the setup script and variables you wrote for a Codex cloud environment have to be rewritten for each cloud agent, because each describes its machine differently.
• What does not travel: config.toml settings, approval and sandbox modes, cloud task history, and anything that lived only in a Codex chat.

# Codex CLI
codex mcp add fenbs --url https://fenbs.ai/api/mcp

# Claude Code
claude mcp add --transport http fenbs https://fenbs.ai/api/mcp

# Gemini CLI
gemini mcp add --scope user --transport http fenbs https://fenbs.ai/api/mcp

Keep the list of delegated work in one place

Cloud agents make it easy to start work and hard to see it all: a Jules plan waiting for approval, two Codex tasks and a Copilot pull request sit in three different lists. fenbs keeps the list on a board instead. Each task has a note saying what and why, a plan, and a test status, in lanes To Do, Next Up, In Progress and Completed. A local agent connects to https://fenbs.ai/api/mcp with OAuth or a token you issue under Settings with a name, scopes and an optional expiry, and clients that only start local servers can use the bridge, npx -y fenbs-mcp. History records each change under the assistant’s name. fenbs has no assignee field, so say in a comment which agent has the task. Set-up is on Codex CLI on fenbs.