How to run multiple Codex sessions in parallel in 2026: 8 ways compared

Disclosure: Enjoy is our product, and it runs Codex. We checked every way below against OpenAI’s documentation and each maker’s own pages on November 2, 2026, and we say where another way is the better pick, including OpenAI’s own tools, which need nothing but your ChatGPT plan.

The ways to run Codex in parallel, compared

You can run several Codex sessions at once, and OpenAI builds for it: its worktrees guide says worktrees “let Codex run multiple independent chats in the same project without interfering with each other.” The questions are who keeps track of the sessions, where they run, and how you keep them out of each other’s files.

Way	Best for	Price, checked November 2, 2026	Runs where	Who coordinates
The ChatGPT desktop app, with worktrees	Most people who only use Codex	Included in every ChatGPT plan, from Free; Plus $20 a month	Mac, Windows; Linux in preview	You
Terminals with codex --worktree	No extra cost, if you’re at home in a terminal	Included from ChatGPT Plus ($20 a month), or an API key	Mac, Windows, Linux	You
Subagents	Splitting one task inside a chat	Included with Codex, at a higher token cost	Inside one chat	Codex
Codex Cloud	Work that keeps going while your computer sleeps	Included from ChatGPT Plus ($20 a month)	OpenAI’s cloud	You
Enjoy	Codex next to other labs’ agents, across all your projects	A monthly subscription, free for 7 days (pricing)	Mac, Windows, Linux	You, with a superagent over everything
Orca	Many agents, free and open source	Free	Mac, Windows, Linux	You
T3 Code	A free, open-source app you can drive from your phone	Free	Mac, Windows, Linux	You
Conductor	A polished app, if you’re on a Mac	Free; Pro $50 a month	Mac	You

Every way here runs on your ChatGPT plan or an OpenAI API key, and every session draws on it. OpenAI’s pricing puts Codex in the desktop app on every plan, but Free and Go, at $8 a month, get “GPT-6 Luna at Standard speed in the desktop app, subject to rollout.” Plus, at $20, adds “Codex on the web, in the CLI, in the IDE extension, and on iOS,” and Pro is $100, $200 or $500.

Paseo, Paradev, formerly Superset, and CodexMonitor run several Codex agents in worktrees too. Our Conductor alternatives guide compares the first two, and CodexMonitor, built only for Codex, hasn’t had a release since March 24, 2026.

Before you start: worktrees and usage

Three things apply to every way below.

• Give each session its own copy of the files. Two sessions in one folder can overwrite each other’s edits. A Git worktree is a second checkout that shares the repository’s history: in OpenAI’s words, “Each worktree has its own copy of every file in your repo but they all share the same metadata (.git folder) about commits, branches, etc.” Files Git ignores, such as .env, aren’t in a new one. In OpenAI’s desktop app, a .worktreeinclude file at the repository root lists the ones to copy.
• A branch can live in only one place. “Git only allows a branch to be checked out in one place at a time,” so a branch a worktree is using can’t be checked out in your main folder too.
• Every session spends the same plan. OpenAI’s pricing says “Local messages and cloud chats share your plan’s usage allowance. Weekly limits may also apply,” and that “Pro plans currently have no five-hour limit.” ChatGPT Work “uses the same pricing, credits, and usage limits as Codex,” and OpenAI says that “Starting in December, Plus, Pro, and Business plans’ usage allowance will be shared across ChatGPT and Codex” (what changes). When several sessions run you out, see what to do when Codex hits its usage limit.

The ChatGPT desktop app, with worktrees

Best for: most people who only use Codex and want parallel chats without a terminal.

OpenAI’s desktop app is where Codex runs on your computer: “Run projects in parallel, work with files, use your computer, and keep long-running work moving from one desktop workspace.” To keep a chat out of your main folder, select Worktree under the composer, choose the branch to start from, and send. Codex creates the worktree in $CODEX_HOME/worktrees by default, “in a detached HEAD state,” so it can “create several worktrees without polluting your branches.” When the work is ready, Create branch here turns it into a branch you can push, or Hand off moves the chat and its code into your main folder.

A local environment runs a setup script, such as npm install, “automatically when Codex creates a new worktree at the start of a new chat.” The Windows app “supports core workflows such as worktrees, scheduled tasks, Git functionality,” and the Linux app is in preview.

Price: included in every ChatGPT plan, from Free. Free and Go get GPT-6 Luna only; Plus, at $20 a month, adds GPT-6.1 Sol.

Watch for: “Worktrees require a Git repository.” “By default, Codex keeps your most recent 15 Codex-managed worktrees” and deletes older ones, after saving a snapshot you can restore, because “Worktrees can take up a lot of disk space.” It runs only OpenAI’s agents. For work that continues while your computer is asleep, OpenAI points to Codex Cloud.

Terminals with codex –worktree

Best for: people at home in a terminal who want nothing extra.

Open a terminal tab per task and start each session in its own worktree:

codex --worktree

OpenAI’s command reference says the flag will “Start in a new managed Git worktree,” for new sessions, codex fork, codex exec and codex exec fork. Inside a running session, type /worktree and “Choose whether to start a new chat or continue the current conversation in a new worktree,” or choose Browse worktrees to find the ones you made. For scripts, codex exec --worktree "implement the next step" runs a task without you, and codex exec fork <SESSION_ID> --worktree "try an alternative approach" tries a second idea from the same session. codex agents opens “the command center to browse chats and switch between agents on a shared server.”

Price: included from ChatGPT Plus, at $20 a month. With an API key, Codex in the CLI is billed at API rates.

Watch for: you track every session, review every diff and merge every branch yourself. --worktree “requires local execution,” and codex resume “accepts model and sandbox overrides, but not --worktree.” /worktree “is unavailable while a task is running,” and the command center “requires a shared local or remote server.” OpenAI’s CLI guide adds: “Create Git checkpoints before and after a task so you can revert changes.”

Subagents

Best for: letting Codex split one task into pieces inside one chat, such as a review from several angles.

“ChatGPT Work and Codex can run subagent workflows by spawning specialized agents in parallel and then collecting their results in one response,” says OpenAI’s subagents guide, and “Current Codex releases enable subagent workflows by default.” Ask in plain words, “spawn two agents,” “delegate this work in parallel,” or “use one agent per point,” and say whether Codex should wait for all of them. Codex ships default, worker and explorer agents, and you can write your own as TOML files in ~/.codex/agents/ or a project’s .codex/agents/. In the CLI, /subagents switches to a helper’s thread.

Price: included with Codex, but “Because each subagent does its own model and tool work, subagent workflows consume more tokens than comparable single-agent runs.”

Watch for: you steer the main chat, and the helpers report to it. OpenAI suggests subagents first “for read-heavy tasks such as exploration, tests, triage, and summarization,” and to “Be more careful with parallel write-heavy workflows, because agents editing code at once can create conflicts.” “Subagents inherit your current sandbox policy.” For when one agent with skills beats several, see why Anthropic says to stop using most subagents.

Codex Cloud

Best for: tasks that should keep going after you close your laptop, or several tries at the same task.

“Codex Cloud runs coding tasks in the cloud, so work can continue while your computer is asleep,” say OpenAI’s cloud environment docs, and “Each new task gets its own isolated workspace from the published environment,” so tasks can’t touch each other’s files. Start them on the web, from Work in > Cloud in the desktop app, in the phone app, or from a terminal, where codex cloud exec takes --attempts from 1 to 4, the “Number of assistant attempts (best-of-N) Codex cloud should run”:

codex cloud exec --env <environment-id> --attempts 3 "Fix the flaky checkout test"

By default, each task runs in a virtual machine with 2 vCPUs and 8 GiB of memory on Plus, and 4 vCPUs and 16 GiB on Pro and Business.

Price: included from Plus, at $20 a month. OpenAI’s pricing lists Codex Cloud as unavailable with an API key.

Watch for: “Cloud tasks may use more of your allowance than local messages.” An environment starts from GitHub repositories you choose, so push your work first, and the docs list “GitLab and self-hosted GitHub Enterprise Server” and “Computer and browser use” as not yet supported.

Enjoy

Best for: running Codex next to Claude Code and other labs’ agents, across all your projects, without stopping when one account runs out.

Enjoy runs Codex, Claude Code, Grok Build, Antigravity, Muse Code, and local models through OpenCode, on Mac, Windows and Linux, on the AI plans you already have. Each conversation runs its own agent, and “There’s no limit on how many agents work at once, beyond what your computer and your AI plans can handle.” An agent works in a worktree when you ask it to try something “in a new worktree,” and the conversation list groups conversations by the copy they work in. What it adds:

• Keep going through limits. When Codex runs out of usage, Enjoy switches to another of your ChatGPT accounts that still has usage, or waits for the limit to reset, then continues. When OpenAI gives your account a free reset, the waiting card offers Use a reset.
• Let an agent fan out the work. Ask “Start separate conversations to write the About page and the FAQ,” and an agent starts up to five conversations per request, with a link to each.
• See what needs you. The conversation list shows Working and Needs you for every conversation, the activity panel lists Codex’s own subagents while they work, and JOY watches all your projects at once.
• Switch agents mid-conversation. When Codex is the wrong fit, Claude Code or another agent picks up the same conversation.

Price: one monthly subscription, and the first seven days are free, with a card to start. It includes your phone, any browser and your other computers, and each teammate you invite needs a subscription of their own. See pricing.

Watch for: Enjoy isn’t open source, and it’s a subscription on top of your ChatGPT plan once the free week ends. It runs the Codex CLI, which OpenAI includes from Plus, so ChatGPT’s Free and Go plans won’t do. It doesn’t stop two conversations from editing the same files, so overlapping work needs a worktree, which you ask the agent for rather than switching on, and a new worktree leaves out uncommitted changes and files Git ignores. Scheduled workflows run only while your computer is awake with Enjoy open. If you only use Codex, OpenAI’s desktop app does the core of this at no extra cost.

Orca

Best for: a free, open-source app for many agents side by side, with several Codex accounts.

Orca, from Stably AI, says: “Run Claude Code, Codex, and any other coding agent in parallel, each in its own worktree. Terminals, diffs, a browser, and a CLI, in one app built for agents.” It’s “Free and open source for macOS, Windows, and Linux,” under the MIT license. Its agent list gives Codex “Deep integration: usage, hot-swap,” and its homepage says you can “hot-swap Codex accounts without re-logging in.” A mobile companion for iOS and Android follows your agents, and a Remote Orca Server keeps them running “after your laptop sleeps” on a machine you control.

Price: free.

Watch for: “Orca pre-fills each supported CLI’s permission-bypass flag for new launches,” which for Codex is --dangerously-bypass-approvals-and-sandbox; switch to Manual under Settings → Agents → Agent Permissions if you want Codex to ask first. Its ways to run page says “Orca does not sell managed VPS hosting,” so an always-on setup is a machine you run.

T3 Code

Best for: a free, open-source app for Codex and other agents, at your desk or on a server you drive from your phone.

T3 Code calls itself “The open-source control plane for coding agents,” to “Orchestrate Claude Code, Codex, Antigravity, OpenCode, Cursor, and Grok from one surface.” “Every agent thread writes to its own branch,” and one button opens a pull request. For Codex, you “install Codex CLI and run codex login.” It runs on Windows, Mac and Linux, with phone apps, and its FAQ says “Run npx t3 up on any box and drive it from anywhere.” It’s under the MIT license, with more than 41,000 stars on GitHub, and version 0.1.6 came out on October 31.

Price: “Nothing. There’s no way to pay us to use T3 Code.”

Watch for: its README says “We are very very early in this project. Expect bugs.” Its FAQ warns that “Worktrees add up fast. One user cleared 500 GB,” and points to automatic cleanup under Settings > Storage. It sends “basic product usage events,” with “No prompts, no file contents,” until you set T3CODE_TELEMETRY_ENABLED=false.

Conductor

Best for: Mac users who want a polished app for parallel Codex and Claude Code workspaces.

Conductor’s harnesses page says it “can run Claude Code, Codex, Cursor, and OpenCode sessions in isolated workspaces,” and “Conductor includes a managed Codex executable. You can use your own Codex install instead.” Its free plan runs “multiple coding agents in parallel” in “Local workspaces on your Mac,” with your own subscriptions and keys.

Price: free. Pro, at $50 a month, adds “Conductor Cloud and a bunch of cloud workspace hours,” live collaboration, its API and the mobile app; Teams is $60 a user a month.

Watch for: “Conductor is not available for Windows or Linux yet,” and “You need GitHub and at least one agent provider to use Conductor.” “Multiple chats in one workspace share the same branch and current code state,” so parallel work needs separate workspaces. Its pricing FAQ says local sessions end “When your machine shuts off or you close the app.”

When OpenAI’s own tools are enough

If Codex is your only agent, you don’t need another app. The desktop app’s Worktree option, or two or three terminals with codex --worktree, isolate every session at no extra cost. Subagents split one task inside a chat, and Codex Cloud keeps tasks going while your computer sleeps, all on the plan you already pay for.

Reach for an app when you use more than one lab’s agent, when you want every project’s sessions in one place, or when you want work to carry on after one account’s limit. And whichever way you pick, run only as many sessions as you can review: every result still needs checking.

How we compared

We read OpenAI’s pages on worktrees, local environments, subagents, the Codex CLI and its commands, Codex Cloud, cloud environments, the desktop app for Windows and Linux, and pricing; each app’s homepage, pricing and docs; and the GitHub pages of the open-source apps, on November 2, 2026, and every one is linked below. Star counts and release dates come from GitHub that day. We quote each maker’s own words for its limits, don’t count features a maker hasn’t documented, and didn’t test the independent apps ourselves. Codex changes weekly, so we recheck this guide against its sources at least once a month, and the date at the foot of the page says when we last did.