Codex team setup: seats, shared config and admin controls

Updated 8 min read

A team gets Codex through ChatGPT Business or Enterprise seats, because Codex comes with the ChatGPT plan. Business costs $25 a user a month billed monthly, or $20 billed annually, with a two-seat minimum. The repository carries what everyone shares: AGENTS.md, .codex/config.toml, rules, custom agents and skills under .agents/skills. Each developer keeps their own ~/.codex files and sign-ins, and an admin enforces the rest with requirements.toml.

Prices and plan details here come from OpenAI’s Codex pricing page and its ChatGPT Business FAQ, read on 20 October 2026. Prices are in US dollars and may vary by country. Several results for “codex team” are about something else: a GitHub issue asking for Claude Code style agent teams (covered below), and unrelated projects that happen to be called Codex Team.

How does a team pay for Codex?

Most teams use the first of three routes.

Route Price What the team gets
ChatGPT Business Standard seat $25 a user a month, or $20 billed annually. Premium seat $125, or $100 annually ChatGPT, ChatGPT Work and Codex on every seat, SAML SSO, MFA, no training on business data by default
ChatGPT Enterprise or Edu Through OpenAI sales Everything in Business, plus SCIM, RBAC and custom roles, audit logs through the Compliance API, analytics, and retention and residency controls
API key API token prices Codex in the CLI, SDK and IDE extension, with no cloud features such as GitHub code review or Slack

The Business rules, from OpenAI’s Business FAQ:

  • A workspace needs at least two paid seats, in any mix of Standard and Premium. Workspaces created since 24 August 2026 can have up to 200.
  • Premium seats have 5x the usage of Standard and no five-hour limit.
  • Only workspace owners buy or change seat types. Added seats are charged pro rata at once; reductions apply from the next billing cycle.
  • A Business subscription does not include API usage. The two are billed separately.

Workspaces that added a Codex seat before 24 June 2026, or had an invitation for one pending that day, can keep using that legacy seat type. It has no monthly fee, draws every Codex task from the workspace’s shared credit balance, and does not include ChatGPT chat. Other workspaces, new or old, cannot add one.

An API key fits automation in shared environments such as CI, according to the pricing page. Business and Enterprise also offer Codex access tokens for trusted automation, which individual plans do not.

How much Codex usage does a Business seat include?

OpenAI publishes ranges, not fixed limits. For a Standard Business seat, the pricing page estimates 15 to 160 local messages per five hours on GPT-6.1 Sol and 350 to 3,000 on GPT-6 Luna, the same as Plus. Weekly limits may also apply, and cloud tasks can use more of the allowance than local messages.

When a seat runs out, a workspace on flexible pricing can buy shared credits so members keep working. Enterprise and Edu workspaces with flexible pricing have no fixed rate limits; without it, they get the same per-seat limits as Plus for most features. Each member can check their remaining usage with /status in the CLI or on the usage dashboard.

The team’s own files affect how fast a seat runs down. Among OpenAI’s tips on the pricing page: keep AGENTS.md small by nesting it in the directories it applies to, and turn off MCP servers you are not using, since each one adds context to every message.

What should a team commit to the repository?

Codex reads several files from the repository, so a team shares its setup through git. OpenAI’s admin rollout guide recommends this for project defaults, rules and skills.

File What it does
AGENTS.md at the root, and in subdirectories Instructions. Codex reads one file per directory from the project root down to where you work, closer files last, up to 32 KiB combined by default
AGENTS.override.md in a subdirectory Replaces that directory’s AGENTS.md, for a part of the codebase with different rules
.codex/config.toml Project settings and project-scoped MCP servers. Files closer to your working directory win
.codex/rules/ Which commands Codex may run outside the sandbox
.codex/agents/*.toml Custom agents with their own model and instructions
.agents/skills/ Skills, at the repository root or in the folder they apply to

Codex loads the project .codex/ layer, with its config, rules and hooks, only after the developer marks the project as trusted, per config basics. Repository configuration supplies defaults and workflows, but it cannot grant workspace, model, API or connected-system access.

Most public repositories stop at the instructions file. On 20 October 2026 we checked the 1,000 most-starred active GitHub repositories from our earlier scans: 381 had an AGENTS.md at the root, 38 had a .codex directory and only 12 had a .codex/config.toml. Six checked in custom agents under .codex/agents. Skills were more common: 116 repositories had skills under .agents/skills, a median of 6 each and 1,557 in total.

Our AGENTS.md guide covers what to put in the file, and which other agents read it.

What stays with each developer?

Shared through the repository or workspace Stays with each person
AGENTS.md and nested overrides ~/.codex/AGENTS.md, read before the project’s files
.codex/config.toml, rules and custom agents ~/.codex/config.toml, ~/.codex/rules/ and ~/.codex/agents/
Skills in .agents/skills Skills in $HOME/.agents/skills
Plugins shared in a Business or Enterprise workspace Their own sign-in to each MCP server and plugin
Cloud environments shared with an Enterprise workspace Secrets in their Personal vault
Policy an admin sets in requirements.toml Local memories, Codex history and usage

When someone adds a command to the allow list in the CLI, Codex writes it to ~/.codex/rules/default.rules, so allow-lists built up that way stay on one machine until someone copies them into the repository’s .codex/rules/. MCP servers that use OAuth need each person to run codex mcp login <server-name> or sign in from the app.

Memories are local files under the Codex home directory, off by default, and tied to one machine. OpenAI’s own advice is to keep required team guidance in AGENTS.md or checked-in documentation and to treat memories as a recall layer.

What can an admin enforce?

A developer can change the defaults in config.toml but not an admin’s requirements. From OpenAI’s managed configuration page, requirements come from four places:

  1. A system file at /etc/codex/requirements.toml on macOS and Linux.
  2. Agent Security in the ChatGPT admin console, delivered to the members it is assigned to when they sign in to the workspace. Cloud-managed policies are available on Business and Enterprise.
  3. Fields from the older managed_config.toml that the client reads as requirements.
  4. macOS managed preferences pushed through MDM.

A typical requirement limits which permission profiles members may choose, as in this example from the rollout guide:

default_permissions = ":workspace"

[allowed_permission_profiles]
":read-only" = true
":workspace" = true

Enterprise adds role-based access control and custom roles on top. Owners can create custom roles and grant “Use Codex in the cloud” and “Manage workspace environments” separately, per roles and workspace permissions. Codex Cloud is off by default on Enterprise until an admin turns it on. Custom roles, SCIM, the analytics dashboard and API, and the Compliance API audit log are Enterprise only, according to the pricing page’s plan table.

Can Codex run a team of agents?

Yes, as subagents inside one session. Current Codex releases turn subagent workflows on by default: you ask Codex to split work across parallel agents, or an AGENTS.md or skill tells it to, and the main thread collects their results, per OpenAI’s subagents page. In the CLI, /subagents switches between the agent threads. Each subagent does its own model and tool work, so a subagent run uses more tokens than a single agent.

A team can check custom agents into .codex/agents/ as TOML files, each with a name, its own instructions and model settings, for example a reviewer or a docs researcher.

The GitHub issue asking for “Codex Teams”, similar to Claude Code’s agent teams where one session leads several others, was closed as a duplicate on 15 March 2026. Our Claude Code team guide covers agent teams on the Claude side.

Where should a team’s Codex sessions keep what they learn?

AGENTS.md holds the rules a team writes in advance. What one developer’s Codex works out during a task, such as why a migration was reverted or which service owns a queue, stays in that developer’s history or local memories. The next person’s Codex, and the team’s other agents, start without it.

Dexio is a hosted, open source (AGPL-3.0) wiki that AI agents read and write over MCP. Codex, Claude Code, Cursor and your other agents search, read, write and link markdown pages in the same wiki, and people see what their agents know as a page graph. Every change is kept, with the agent and the person behind it.

Each developer adds it to ~/.codex/config.toml with their own API key in place of dxk_..., as in Dexio’s agent instructions, which Codex can also follow to set itself up:

[mcp_servers.dexio]
url = "https://app.dexio.wiki/mcp"
http_headers = { "Authorization" = "Bearer dxk_..." }

Our Codex MCP guide covers checking the connection with /mcp and the AGENTS.md lines that tell Codex when to use the wiki. Codex also supports bearer_token_env_var for a project-scoped server in .codex/config.toml, which would keep keys out of git; we have not tried Dexio that way.

Dexio is free for one person, then $10 (Team) or $20 (Business) a member a month, and agents never count as members. If your team works only in Codex and its shared knowledge fits in AGENTS.md, you do not need a wiki. One is worth adding when the same notes have to reach Codex and your other agents, such as Claude Code or Cursor, and you want to read and correct what they wrote.

FAQ

Is Codex included in ChatGPT Business?

Yes. Standard and Premium Business seats both include Codex, along with ChatGPT and ChatGPT Work. Our ChatGPT Business guide covers the rest of the plan.

How much is Codex for a team of five?

Five Standard Business seats are $125 a month billed monthly, or $100 a month billed annually. Usage past each seat’s limits comes from workspace credits if the workspace has flexible pricing.

Can a team share one Codex usage limit?

Not on Standard or Premium seats: each seat has its own allowance. Credits bought for the workspace are shared and cover eligible usage once a member’s included limits run out.

Should .codex/config.toml be committed?

Yes, for settings the whole team should have, as long as it holds no secrets. Personal settings belong in ~/.codex/config.toml, and Codex only loads the project file once a developer trusts the project.

Sources

Ask a question

Ask anything about Dexio.

About
We reply by email.