Guide OpenAI · Chapter 1

    OpenAI at a glance: ChatGPT, Codex and the platform.

    Employees work in ChatGPT Business or Enterprise, Codex handles tasks in the repository, and the platform carries agents of your own on the Agents API.

    As of: October 2026 · Editor: Marcel Haas

    • ChatGPT Enterprise
    • ChatGPT Business
    • Codex
    • Responses API
    • Agents SDK
    • Agents API

    Stack map

    The map of the platform.

    For every block: what it does, who uses it, how it is bought, where it runs and which chapter goes deeper.

    • Workspace
    • Platform
    • Administration

    For everyone

    Workspace

    ChatGPT Business and Enterprise

    Writing, research and analysis in projects; Deep Research, Company Knowledge and workspace agents for recurring routines.

    Who uses it
    Teams that work with OpenAI's tools
    How it is bought
    Seat per person in the workspace
    Where it runs
    At OpenAI; sign-in through Entra ID, administration in the admin console
    Workspace

    Apps and Company Knowledge

    Apps read documents from SharePoint and Outlook, issues from GitHub or data from your own systems; Company Knowledge answers across several sources.

    Who uses it
    Everyone in the workspace, after approval
    How it is bought
    In Business and Enterprise; Connector Registry in Enterprise
    Where it runs
    At OpenAI, with the person's access to the source
    Workspace

    Dots

    Persistent agents with their own cloud computer that keep working towards goals and report by chat, voice, Slack or Teams.

    Who uses it
    Individuals with governed access
    How it is bought
    Business Premium, Enterprise as a beta
    Where it runs
    At OpenAI, with the person's access

    Development

    Workspace

    Codex

    Works through tickets in the repository, writes code and tests and opens a pull request; locally in the desktop app, terminal and IDE or in Codex Cloud.

    Who uses it
    Development teams
    How it is bought
    Included in ChatGPT Business and Enterprise
    Where it runs
    Locally or in Codex Cloud, switched off at first in Enterprise workspaces
    Administration

    Codex rules

    requirements.toml and the agent security settings define which sandbox profiles, approval policies and MCP servers are allowed; the repository cannot widen them.

    Who uses it
    Your IT
    How it is bought
    Part of the workspace
    Where it runs
    Distributed through MDM and the workspace

    Applications of your own

    Platform

    Responses API and Conversations API

    One call with input, tools and structured output; the Conversations API or your application keeps the history.

    Who uses it
    Development teams
    How it is bought
    Consumption per project with keys and usage limits
    Where it runs
    At OpenAI, per project with a data region
    Platform

    Agents SDK

    The agent loop inside your own application, with guardrails, traces and handoffs between agents; you decide operation, memory and approvals.

    Who uses it
    Development teams
    How it is bought
    Consumption of the model calls per project
    Where it runs
    Wherever your application runs, for instance in Azure Container Apps
    Platform

    Agents API

    The runtime OpenAI itself runs Codex on: sessions, sandboxes, context compaction, several agents, computer use in the hosted browser.

    Who uses it
    Development teams
    How it is bought
    Consumption per project and runtime of the agent
    Where it runs
    At OpenAI; not listed for the Europe region
    Platform

    ChatKit and Decisions API

    ChatKit provides the embedded chat interface, the Decisions API picks one answer from a fixed set, for classification and routing.

    Who uses it
    Development teams
    How it is bought
    Consumption per project; Decisions API in preview
    Where it runs
    At OpenAI

    Administration

    Administration

    Admin console

    SAML SSO with Entra ID, SCIM, roles and groups, approved apps and agents, the Compliance API for audit, eDiscovery and loss prevention.

    Who uses it
    Your IT
    How it is bought
    SCIM and the Compliance API in the Enterprise plan
    Where it runs
    At OpenAI
    Administration

    API projects

    One project per application with keys that expire, hard spending limits and the data region fixed at creation.

    Who uses it
    Your IT and development
    How it is bought
    Consumption per project
    Where it runs
    At OpenAI, in the chosen region
    Administration

    Directive and test set

    What is allowed, which apps and agents are approved, and the test set every version is measured against.

    Who uses it
    Management, business team, data protection
    How it is bought
    Your own work
    Where it runs
    With you
    01

    The work starts in the workspace.

    A communications department creates a project for the annual report in ChatGPT, attaches the quarterly reports and has Deep Research look up the industry figures; an app reads last year's edition from SharePoint. That is the first layer: ChatGPT Business or Enterprise, a workspace with projects, apps for Microsoft 365, GitHub and your own systems, Company Knowledge across several sources and workspace agents for recurring routines. OpenAI does not use workspace content for training by default. Dots, persistent agents with their own cloud computer, sit on the same layer and are reachable by chat, voice, Slack or Teams.

    02

    Codex works in the repository and submits changes for review.

    A development team gives Codex a ticket: a new interface to the billing system, with tests. Codex reads the code, writes the change and the tests, shows what it executed and opens a pull request that a person reviews. It works locally in the desktop app, in the terminal and in the IDE, or in Codex Cloud on a connected repository; rules and skills for your repository live in the .codex or .agents folders. In Enterprise workspaces Codex Cloud starts switched off: administrators enable use through roles, define environments, secrets and internet access, and distribute rules for local runs in a requirements.toml file.

    03

    The platform carries applications and agents of your own.

    An application that turns an association's meeting minutes into resolutions with owner and deadline needs a model call with tools and no chat window. That is what the Responses API is for: one call sends input and tools, such as file search, web search, the code interpreter or an MCP server, and gets structured output back; the Conversations API or your application keeps the history. With the Agents SDK the agent loop runs inside your own application, with guardrails for inputs and outputs and traces for every step. The Agents API provides the runtime OpenAI itself runs Codex on: sessions that persist, sandboxes, context compaction and collaboration between several agents. ChatKit provides the embedded chat interface, the Decisions API picks one answer from a fixed set.

    04

    Administration belongs to your IT.

    Employees sign in through SAML single sign-on with Entra ID; in ChatGPT Enterprise, SCIM creates accounts and removes them when someone leaves, roles and groups control who uses Codex, apps or agents, and the Compliance API delivers conversations and activities to your tools for audit, eDiscovery and loss prevention. On the platform every application gets its own project with keys that expire, hard spending limits and a data region you set when creating it. This layer stays in place when a model or a tool changes.

    05

    Where OpenAI and Microsoft meet.

    The same GPT models run in Microsoft Foundry as Azure OpenAI and are available under Copilot; ChatGPT reaches SharePoint and Outlook through apps and connectors; sign-in runs through Entra ID on both sides, and Purview's data security posture view also captures a connected ChatGPT Enterprise workspace. The Model Context Protocol connects tools with agents on both platforms, since December 2025 under the roof of the Agentic AI Foundation. For your overview this means: you set identities, data rules and logs once, and both worlds hang off them.

    What you decide.

    Workspace, Codex or platform: where to start?

    Options

    1. 1With the workspace for a team that wants to use the tools directly
    2. 2With Codex in a development team's repository
    3. 3With an application on the platform

    Our recommendation: With the part a case exists for. A business team that already works with private accounts gets a workspace with sign-in through Entra ID and approved apps; a development team with a pipeline and review gets Codex for a named group; an application comes when a workflow without a chat needs a model with tools. Projects, keys and usage limits are set up by your IT first in every case.

    What we advise against: A workspace for everyone before roles, approved apps and the directive exist. The private accounts only disappear once the offer holds up in daily work.

    Business or Enterprise?

    Options

    1. 1ChatGPT Business for teams that start quickly
    2. 2ChatGPT Enterprise with SCIM, the Compliance API and data residency

    Our recommendation: Business when a team starts with sign-in through Entra ID and approved connectors and nobody needs audit exports. Enterprise as soon as accounts should come and go through SCIM, the Compliance API must deliver evidence, or data residency and retention rules of your own are required; Codex is included in both plans.

    What we advise against: Enterprise for a pilot team of a few people without audit requirements, or Business for an organisation whose data protection officer demands residency and evidence.

    Who keeps the overview current?

    Options

    1. 1Your IT as the owner of workspace and projects
    2. 2One person per application in the business or in development
    3. 3A partner who maintains both

    Our recommendation: Your IT owns the workspace, the roles and the API projects; every application has a named person who maintains instructions, tools and the test set; and every entry carries a review date. We maintain the overview with you for as long as you want, and hand it over so your team carries it on.

    What we advise against: An overview without a date. What OpenAI changes in one quarter turns an undated plan into a guess.

    Status and sources.

    As of: October 2026

    • As of October 2026

      Agents API, Agents SDK or Responses API.

      OpenAI matches the Agents API to long-running tasks where OpenAI runs the agent, the Agents SDK to agents with their own tools and workflows inside your application, and the Responses API to direct model calls or an agent built from scratch. ChatKit provides an embedded chat interface.

      Source: OpenAI: Agents
    • Since 26 August 2026

      Assistants API shut down, Agent Builder next.

      OpenAI removed the Assistants API on 26 August 2026; the successors are the Responses API and the Conversations API. Agent Builder and the Evals platform shut down on 30 November 2026, and existing evals become read-only from 31 October 2026. ChatKit remains.

      Source: OpenAI: Deprecations
    • As of October 2026

      Codex Cloud starts switched off in Enterprise workspaces.

      Administrators enable use through roles, define environments, secrets and internet access, and distribute rules for local runs in a requirements.toml file. Codex Cloud respects the repository permissions of the source system.

      Source: OpenAI: Codex admin rollout guide
    • Since 29 September 2026

      OpenAI brings dots.

      Dots keep working towards goals in ChatGPT on their own cloud computer and are reachable by chat, voice, Slack or Teams. In Switzerland through Business Premium and as an Enterprise beta, not for Pro accounts at launch.

      Source: OpenAI Help Center: Getting started with your dot

    Frequently asked questions.

    Do employees sign in to ChatGPT with their Microsoft account?

    Yes, through SAML single sign-on with Entra ID; in ChatGPT Enterprise, SCIM creates the accounts and removes them when someone leaves. Roles and groups control who may use Codex, apps or agents, and the Compliance API delivers conversations and activities to your tools for audit and eDiscovery.

    How do ChatGPT Business and Enterprise differ?

    Both give teams a workspace with projects, apps and workspace agents, and Codex is included in both. Enterprise adds user provisioning through SCIM, role-based access, the Compliance API for audit and eDiscovery, data residency and retention rules of your own. Which plan fits is decided by the audit requirements and by how many people come and go.

    What distinguishes a workspace agent from an agent on the Agents API?

    A workspace agent is created in ChatGPT from instructions, apps and files and works for the people in the workspace; your team maintains it in the interface. An agent on the Agents API runs in your own application or workflow, with sessions, sandboxes and tools your developers define, and with approvals before every action with consequences.

    Do we need the API if we have ChatGPT Enterprise?

    Only when AI belongs in a workflow of your own: an application without a chat, an agent in your process, batch processing. For people who work with documents, research and drafts, the workspace with apps and workspace agents is enough; workspace and API are administered and billed separately.

    Where do employees find apps and agents?

    In the ChatGPT workspace: apps and Company Knowledge in Business and Enterprise, your own apps from Developer Mode, workspace agents someone has shared, and dots where they are enabled. Who sees what is controlled by roles and groups in the admin console, in Enterprise additionally by the Connector Registry.

    What may Codex do on a managed computer?

    What requirements.toml and the workspace's agent security settings allow: which sandbox profiles (read-only, workspace-write, danger-full-access), which approval policies and which MCP servers. Rules from the repository cannot widen these limits, and Codex Cloud respects the repository permissions of the source system.

    Next chapter · Chapter 2

    Which OpenAI model does which work.

    Which OpenAI model does which work, and how do we choose?