OpenAI's consumer and enterprise AI assistant, including web/desktop/mobile clients, ChatGPT Apps built with the Apps SDK, and workspace admin/governance controls.
Software capabilities Integrations & Extensibility App templates Gives admins guided setup flows for workspace-specific apps for providers like GitHub Enterprise, Snowflake, and Databricks. App Templates Lets a workspace admin turn an OpenAI-provided template into a workspace-specific app by supplying tenant hostnames, OAuth credentials, scopes, and MCP server URLs. Apple Intelligence integration Lets admins enable or disable workspace access to Apple Intelligence, allowing users to select their workspace when signing in. Apple Messages plugin Reads and searches iMessage, SMS, and RCS conversations and prepares or sends messages through Apple Messages on Apple silicon Macs. Apps and plugin directory Provides a directory where users discover and install apps and packaged plugins, with admin installation policies. ChatGPT Plugins via MCP Lets third-party developers extend ChatGPT and Codex with custom tools and UI by running an MCP server, published as a plugin (renamed from the Apps SDK). Company Knowledge Search/Fetch Compatibility Lets an app's read-only tools be surfaced as a citable Company Knowledge source when it implements the standard MCP search/fetch tool schemas. Connector templates (Snowflake, Databricks) Connects ChatGPT to Snowflake via a Snowflake-managed MCP server and to Databricks via a customer OAuth app connection, available to workspace admins and Plus/Pro users with egress IP allowlist guidance. Connector write actions Enables admin-gated write actions in Google and Microsoft apps such as drafting email, creating docs, and managing calendar events. Connector Write Actions Lets connectors perform write actions in third-party apps, not just read data. Connectors Connects third-party tools so ChatGPT can search and reference their content in chat and deep research under admin controls. Custom MCP Apps (Developer Mode) Lets admins and authorized developers build, privately test, and publish custom MCP-based apps with read, write, and interactive-UI actions for their workspace. Custom MCP connectors Lets organizations build, publish, and govern custom MCP connectors with read and write actions in their own tools. Education plugins Provides guided plugins for educators and students covering lesson design, assessments, and study practice in Edu workspaces. Excel Sidebar Provides a ChatGPT sidebar inside Excel for working with spreadsheet data. GPT Actions Connects a GPT to an external REST API through an OpenAPI schema with no-auth, API-key, or OAuth authentication. GPT Actions Data Retrieval Framework Guides building Custom GPT actions that retrieve data from REST APIs, relational databases, or vector databases. Healthcare plugins Provides HIPAA-eligible plugins for public healthcare data search and read-only Epic EHR patient information review. Multi-Account Connections with Profile Tool Lets users connect multiple accounts to the same plugin and lets developers declare an authenticated profile tool, marked _meta["openai/profile"]: true, that returns a stable opaque profile ID so OpenAI can distinguish and label connected accounts. Office and Sheets add-ins Brings a ChatGPT sidebar and add-in to Excel, Google Sheets, and PowerPoint for spreadsheet and presentation work inside those tools. Outlook Shared Mailbox Access Lets ChatGPT access a shared Outlook mailbox through the Outlook connector. Outlook shared mailbox and calendar Supports delegated Outlook workflows for shared mailboxes and shared calendars, including reading, moving, and sending mail and managing events. Plugin Directory Provides the directory where users browse and enable plugins and their underlying apps for ChatGPT and Codex, replacing the app directory as of July 9, 2026. Plugin iframe embedding Lets plugins embed editors and admin interfaces from their MCP server's own registrable domain in widget iframes under the iframe policy. Plugin marketplace import Lets admins distribute plugin marketplaces from GitHub repositories with automatic daily sync. Plugin sharing Lets users share locally built plugins with their workspace for teammates to install from the Codex plugin directory. Plugins Packages skills, apps, and app templates into a single installable unit that admins enable per role to deliver a workflow across ChatGPT and Codex. Role-specific plugins Packages role-specific skills, app integrations, and workflow guidance for functions like sales, analytics, design, and finance. Sign in with ChatGPT Lets a user create or link an account with a supported plugin or partner site directly from the plugin connection flow, sharing only name, email, and profile picture. Site Tools (WebMCP) Lets ChatGPT discover and use tools a website exposes via WebMCP directly in the desktop app's built-in browser, without a separate plugin connection. Skills Provides reusable instructions, prompts, and workflow patterns that a user or workspace can create, upload, share, and publish for ChatGPT and Codex to follow. Skills Turns proven workflows into reusable, shareable instructions that ChatGPT applies automatically when relevant. Slack connector actions Lets ChatGPT take Slack actions such as joining channels, creating reminders, and uploading files under admin action controls. Vertical Conversion Plugin Specs Defines beta, partner-gated conversion flows that let ChatGPT invoke partner plugins directly for high-intent local-services quotes, restaurant reservations, and product checkout. Workspace Agents API (Trigger Runs) Lets an external system programmatically trigger a published ChatGPT workspace agent and poll its run status via a workspace-scoped access token. Workspace Agents in Slack Deploys a published ChatGPT workspace agent into Slack channels under its own handle, where it answers on mention or on all relevant messages and can post on a schedule.
Platform Trust & Compliance Active Session Controls Lets a user view and revoke active login sessions across devices. Active sessions Lets users review sessions associated with their account and sign out of unrecognized ones. Ad Controls Lets Free and Go users turn off ad personalization, dismiss ads, see why an ad was shown, clear ads data, and switch the Free plan to an ads-free experience. Advanced Account Security Provides advanced account security options such as stronger authentication and session protections. Age Prediction Estimates whether a user is likely under 18 to apply age-appropriate protections. Agent website blocking Lets Enterprise and Edu workspace owners request blocking of specific websites or domains from ChatGPT agent browsing and actions. App Plugin Submission & Review Requires apps to pass a review against quality, safety, privacy, and IP guidelines before being packaged and published as a plugin. Chat and File Retention Controls Applies a workspace-configured retention period to chats and Library files and permanently deletes user-deleted content within 30 days. ChatGPT for FedRAMP Provides a FedRAMP Moderate accredited version of ChatGPT Enterprise with a restricted feature set for US government workloads. ChatGPT for Teens Serves eligible under-18 accounts a version of ChatGPT with age-appropriate protections, teen-specific onboarding and starter prompts, and optional parental controls. Compliance API Exposes Enterprise and Edu workspace conversations, files, and recording artifacts to audit, SIEM, eDiscovery, and DLP systems, and supports programmatic deletion. Compliance Logs Platform Exports immutable, time-windowed JSONL compliance, audit, authentication, and Codex usage logs through the Compliance API. Computer History Optionally records interaction events from selected apps and websites on macOS so Enterprise members can bring that context into ChatGPT, with admin gating and member opt-in. Content Security Policy Controls for Widgets Lets a developer declare an allowlist of domains a widget may fetch from, load assets from, or embed as iframes inside its sandbox. Corporate Network Workspace Blocking Restricts which ChatGPT workspaces users on a corporate network can sign in to by matching a ChatGPT-Allowed-Workspace-Id HTTP header injected by the network. Data Residency Stores in-scope ChatGPT customer content at rest in a selected country or region. Data residency for sync apps Supports in-region data residency for synced connector content across supported regions including Japan. Domain Verification Confirms ownership of a domain through a DNS record before SSO and public GPT Builder Profiles can be enabled for that organization. Enterprise Key Management Encrypts ChatGPT Enterprise customer content with keys held in the customer’s own AWS, Google Cloud, or Azure key management service. Enterprise Key Management Encrypts customer content with the organization's own external key management system on AWS, GCP, or Azure. HIPAA-Eligible Workspaces Makes ChatGPT for Healthcare, Regulated Workspaces, FedRAMP, and Clinicians available under a Business Associate Agreement covering a defined set of ChatGPT functionality. Inference Residency Keeps model inference on in-scope ChatGPT customer content executing inside a selected region for workspaces that already have data residency there. IP allowlisting Restricts ChatGPT and Compliance API access to admin-specified IP addresses. IP Allowlisting Blocks requests to an Enterprise or Edu workspace and to the Compliance API from any IP address outside an admin-defined allowlist. Location Sharing Lets a user share device location with ChatGPT, with parental controls to disable it. Lockdown Mode Provides a heightened-security lockdown mode that restricts higher-risk account features. Multi-Factor Authentication Requires a second verification step at sign-in using an authenticator app, push notification, SMS or WhatsApp code, or passkey across ChatGPT and the API Platform. OAuth 2.1 Authentication for MCP Servers Lets an app authenticate end users through a standards-based OAuth 2.1 flow with discovery metadata, client registration, PKCE, and mTLS verification. Parental Controls Lets a parent link a teen's account and manage its settings and safety features. Passkeys Lets a user register device-bound or hardware-security-key passkeys to sign in to an OpenAI account without a password or to satisfy a second factor. Reduce Sensitive Content Applies an account-level setting that makes ChatGPT respond more cautiously to graphic, sexual, or otherwise age-inappropriate topics, on by default for teen accounts. SCIM Provisioning Provisions and deprovisions ChatGPT workspace members and groups automatically from an identity provider directory, with an SFTP CSV fallback. Single Sign-On Authenticates workspace members through a SAML or OIDC identity provider, sharing one connection between a ChatGPT workspace and its paired API organization. Trusted Access for Cyber (Daybreak) Grants vetted enterprise security teams tiered access to cybersecurity workflows that default safeguards otherwise restrict. Trusted Contact Lets a user designate a trusted contact for account safety or recovery scenarios. Workspace data export Lets Edu workspace members request and download an export of workspace conversation data when admins enable it.
Access & Governance Additive role evaluation Evaluates a member's applicable roles additively with Default, On, and Off permission settings per role. Admin API keys Lets admins create workspace-scoped Admin keys in the Global Admin Console for administration, spend controls, and analytics APIs. Admin-Managed App Connections Lets an admin connect a third-party source such as Google Workspace or Microsoft Teams for the whole workspace using a service account, selecting which content syncs and which members get access. App permission prompts Lets admins set when ChatGPT asks members before using connected apps, with workspace defaults and per-app overrides. App Permissions Controls when ChatGPT must ask a user before an app reads data or takes an action, and lets admins set which actions an app may take. Centralized identity management Provides tenant-wide SCIM user and group sync in Admin Console with assignment to ChatGPT workspaces and Ads accounts. Chat model defaults Lets workspace owners configure the starting Chat model and reasoning level for new chats. Codex policy audit logs Records changes made through the Codex Policies & Configurations interface in workspace audit logs reviewable in Admin Console and retrievable via the Compliance API. Desktop app update controls Lets admins turn off in-app updates for the ChatGPT desktop app so releases are reviewed and deployed through device management. Global Admin Console Provides a tenant-level console that spans multiple ChatGPT workspaces and API organizations for managing domains, a single SSO connection, global admins, analytics, and usage limits. Global Admin Console Gives admins a tenant-level console for members, groups, roles, settings, audit logs, analytics, and billing across workspaces. GPT Workspace Controls Lets workspace owners govern who can build, share, and use GPTs, restrict GPT actions to allowlisted domains, and control access to third-party GPTs. Group Managers Lets workspace owners designate members as group managers with delegated group administration such as analytics, usage limits, and feature permissions. Groups Lets owners and admins create named groups of workspace members and use them to assign roles and grant access to GPTs, projects, and apps. Groups Admin API Provides Admin API endpoints for creating, updating, and deleting workspace groups programmatically. Managed browser and computer-use controls Gives admins policy settings for browser and native-app use by desktop clients, including site defaults, uploads, downloads, and app allow/block rules. Model access controls Lets workspace owners manage access to supported models at the workspace and role level. Model access testing Lets admins inspect a member's model access and its underlying settings from the Admin Console without granting access. Role-based access controls Lets workspace owners create custom roles with granular permissions, assign them to groups, and set default member roles. Role-Based Access Controls Lets a workspace owner define custom roles with granular feature permissions and assign them to groups, evaluated additively so any granting role retains access and denial requires every applicable role to be off. SCIM group discoverability control Lets workspace owners control whether SCIM-managed groups appear in sharing flows like projects and GPTs. Sites Workspace Controls Lets Business, Enterprise, and Edu admins decide through RBAC who can create Sites and who can publish them publicly. Workspace Admin & Governance Controls Gives ChatGPT Enterprise and Business admins controls to manage workspace membership, roles, and data governance. Workspace Admin APIs Provides workspace-scoped Admin APIs for automating invitation and member administration, including roles, seats, and removals. Workspace Admin Keys Issues workspace-scoped Admin keys in Admin Console with per-category read and write permissions and an expiry, authenticating the ChatGPT and Codex administration APIs for one workspace.
Client Apps & Surfaces 1-800-ChatGPT Lets a person talk to ChatGPT by calling 1-800-CHATGPT or messaging the verified contact on WhatsApp without a ChatGPT account. Android App Provides a native Android app for ChatGPT. Apple CarPlay Provides a hands-free ChatGPT voice experience through Apple CarPlay. Apple Intelligence Integration Connects a ChatGPT account to Apple Intelligence on iOS, iPadOS, and macOS so Siri, Writing Tools, and Visual Intelligence can route requests to ChatGPT. Appshots Attaches the frontmost app window — screenshot plus available text — to a chat via keyboard shortcut on macOS and Windows. Built-In Browser (Desktop App) Opens a browser inside the ChatGPT desktop app that the user and ChatGPT share, supporting tabs, sign-in, autofill, downloads, and page annotations under per-host access approval. Chat Bar Opens a floating macOS prompt window with a keyboard shortcut so a user can ask ChatGPT, attach files, or capture a screenshot without switching to the app. ChatGPT Atlas Provides a standalone macOS web browser with an Ask ChatGPT sidebar, browser memories, and an in-browser agent mode that acts on pages for the user. ChatGPT Desktop App Provides a native macOS and Windows desktop app that hosts Chat, Work, and Codex in one client. ChatGPT for Intune Provides a separate iOS app managed through Microsoft Intune and Entra with app protection and Conditional Access policies. ChatGPT for KakaoTalk Provides a ChatGPT conversation surface inside KakaoTalk in Korea, with optional linking to an existing ChatGPT account. ChatGPT in Viber Powers translation, summaries, message polish, image remix, a ChatGPT tab, and @ChatGPT mentions inside Rakuten Viber, with optional ChatGPT account linking for higher limits. Desktop app Combines Chat, Work, and Codex in native macOS and Windows apps with unified recents, projects, and cross-device sync. iOS App Provides a native iOS app for ChatGPT. iOS Home Screen Widget Provides an iOS home screen widget for quick access to ChatGPT. Linux desktop app Installs and updates the ChatGPT desktop app on Ubuntu, Debian, Fedora, and Arch Linux. Mobile Skills Menu Surfaces available ChatGPT tools and skills in a menu on mobile. Quick chats Lets users send typed quick chats and check task progress from floating desktop Pets controls without opening the main window. Safari Default Search Engine Lets a user set ChatGPT search as the default search engine in Safari. Screenshot Capture (macOS) Captures a selected open application window from the macOS app composer and attaches it to the prompt. Vision Pro App Provides a ChatGPT app for Apple Vision Pro on the visionOS App Store. Windows App Deployment Lets IT install and update the ChatGPT Windows desktop app via self-service, Microsoft Store/Intune management, or an offline Store-signed MSIX package that auto-updates once a device can reach persistent.oaistatic.com. Work with Apps Lets the macOS app read and edit content in supported local applications such as IDEs, terminals, and Notes, via a companion extension for VS Code and its forks.
Conversation Experience Chat History Saves each conversation to a browsable, searchable history that a user can revisit, rename, archive, or delete. Code Interpreter / Data Analysis Runs Python in a sandboxed environment to analyze uploaded data, run calculations, and produce files and charts. Continue Generating Resumes a response cut off by a length limit from where it stopped. Conversation Branching Lets a user branch a conversation from an earlier message to explore an alternate direction without losing the original thread. Conversation Drafts Preserves an unsent message draft in a conversation so it is not lost when navigating away. Cross-Device Session Persistence Keeps a user's conversations and sign-in available and synced across devices and sessions. Edit Messages with Images Lets a user edit a previously sent message that includes image attachments and regenerate the response. Fast Answers Returns quick answers to simple questions with lower latency by routing to a faster model path. Instant Access (No Login) Lets people start using ChatGPT without creating an account or signing in. Interface Localization Localizes the ChatGPT interface into multiple languages. Keyboard Shortcuts Provides keyboard shortcuts for common actions in the ChatGPT web and desktop interface. Model picker and reasoning levels Lets users choose among Instant, Medium, High, Extra High, and Pro reasoning levels from the composer. Model Switcher Lets a user choose which underlying model handles a conversation from a model selector. Prompt Examples Shows example prompts on the empty conversation screen to help a user get started. Rich Visual Responses Renders answers with richer inline visuals such as formatted cards, tables, and charts. Steerable Thinking Lets a user control how much reasoning effort the model spends before answering. Stop Generating Lets a user halt an in-progress model response mid-generation. Study mode Guides learning through Socratic questions, sectioned explanations, and understanding checks grounded in uploaded materials. Suggested Follow-ups Suggests follow-up questions or next prompts after a response. Table of Contents Generates a navigable table of contents for long responses or conversations. Temporary Chat Runs a conversation that is excluded from chat history, memory, and model training and is deleted from OpenAI systems within 30 days. Temporary Chat Runs chats that skip history and new memories while optionally personalizing responses with existing memories, custom instructions, and plugins. Web Browsing Lets ChatGPT search and read live web pages to answer questions with current information and citations.
Billing & Monetization ChatGPT for Academic Researchers Gives eligible verified faculty/postdoc research teams 12 months of complimentary access to a dedicated ChatGPT workspace with business data protections and Pro-level usage limits. ChatGPT for Financial Services plan Provides a dedicated ChatGPT plan for financial institutions combining premium financial datasets, GPT-6 Astra reasoning, and Enterprise-grade controls. ChatGPT for Teachers Provides verified US K-12 educators a district-scoped ChatGPT workspace, free through June 2027, gated by a school-email verification flow. Codex Access on a ChatGPT Plan Lets a user sign in to the Codex CLI, IDE extension, desktop app, and web client with a ChatGPT account and draw on that plan’s included usage. Codex seat type Offers a flexible-pricing seat type that grants Codex-only access with no fixed per-user cost. External Checkout Monetization Lets an app monetize by redirecting the user to a merchant-hosted checkout flow outside ChatGPT. Flexible pricing Provides credit-based access to advanced models and features through a shared workspace credit pool. In-App Payment Sheet Checkout Opens a ChatGPT-hosted payment sheet so select marketplace partners can collect payment and complete an order without leaving the conversation. Multi-Currency Billing Charges a web-purchased ChatGPT subscription in the local currency and payment methods supported for the account’s country. Plugin credit promotions Grants eligible ChatGPT Business workspaces promotional agentic credits when members install promoted partner plugins and complete qualifying workflows. Seat Management Buys Standard and Premium seat capacity for a workspace and assigns, reassigns or swaps individual seats between members independently of their roles. Subscription Sign-In Access Lets a user sign in within a plugin to an already-paid subscription and access features already included in it, while the plugin must not display subscription plans, initiate new subscriptions, or promote upgrades. Token-based credit pricing Bills Excel/Sheets, Workspace Agent, and PowerPoint usage by input and output tokens rather than fixed credit costs. Usage Credits Lets eligible Plus and Pro users buy prepaid credits, with optional auto top-up, that extend Codex past their plan's included limits. Usage limits and overages Lets admins set monthly credit limits by workspace, group, and user, review increase requests, and manage overage settings. Workspace Credits Lets a workspace buy a shared credit pool with optional automatic reload so members can keep using credit-based features past their included limits. Workspace Usage Limits and Overages Lets Enterprise and Edu admins set monthly per-user, per-group, and workspace-default spend limits with alerts, plus a workspace overage limit once the committed credit pool is exhausted.
Task Experiences Deep Research Plans and runs multi-step research across the web, uploaded files, and connected apps in Chat, Work, and Codex, producing cited, editable reports under plan or Work/Codex allowances. GPT Builder Creates and edits a GPT either conversationally or through a configuration editor, with preview testing and version history. GPT Store Provides a public directory where users browse GPTs by category and eligible builders publish their own under a verified Builder Profile. GPTs Provides purpose-built versions of ChatGPT that combine instructions, uploaded knowledge, selected tools, and connected apps or actions. Health Provides health-related guidance and information features within ChatGPT. Logged-in Free/Go/Plus/Pro users in the US 18+ can connect supported health records and Apple Health data to securely view health dashboards and ask questions grounded in personal health context. Interactive Learning Provides interactive learning tools such as quizzes that help a user practice a topic. Job Search Finds relevant live job listings personalized to a user's experience and goals. OpenAI Certified Surfaces OpenAI-certified partner apps to ChatGPT Enterprise and Edu workspaces, on an invite-only basis. Personal Finances Provides personal finance guidance and tools within ChatGPT. Prism Provides a collaborative LaTeX authoring workspace with an editor-native assistant that generates snippets, diagnoses compile errors, proofreads, and imports existing projects. Pulse Delivers proactive daily research updates based on a user's chats, memory, and feedback. Resume Builder Lets a user create or upload a resume, tailor it to a role, and download a formatted version. Shopping Research Researches products across the web and compiles comparisons to help a user shop. Sites Turns a user's ideas into an interactive website or lightweight app without leaving ChatGPT. Study Mode Guides a user through a topic with interactive Socratic questions instead of giving direct answers. Visualizations Turns a question or dataset into an interactive chart, map, diagram, calculator, or simulation inside a ChatGPT conversation.
Search & Knowledge Retrieval Apps with Sync Indexes content from a connected third-party app in advance so ChatGPT can retrieve it on demand without querying the source at request time. Chat History Search Searches titles and message content across a user’s past conversations from the sidebar on web and mobile. ChatGPT for Healthcare Provides clinical teams an enterprise ChatGPT workspace whose search draws on peer-reviewed literature and clinical guidelines and returns per-source citations, gated by workspace or RBAC. ChatGPT Search Searches the web from within a conversation and answers with links to the sources used, rewriting the prompt into targeted queries sent to search providers. Company knowledge Answers questions using cited context from enabled connectors while respecting each user's existing permissions. Company Knowledge Answers questions using an organization’s own content retrieved from enabled plugins and apps that expose search and fetch, with citations back to the source records. Deep Research Runs a multi-step research task over the public web, uploaded files, and connected apps against a user-reviewable plan, and returns a structured report with citations. Financial data sources Indexes included premium financial datasets such as SEC filings, earnings-call transcripts, and private-company profiles for cited research in ChatGPT for Financial Services. Healthcare Public Data Bundles nine read-only apps for searching public healthcare sources (biomedical research, clinical trials, medication information, Medicare data, provider records) inside ChatGPT for Clinicians, without accessing patient charts. Offline Web Search Answers web-search requests from OpenAI indexed and cached web content instead of calling live external search providers, for workspaces with stricter governance requirements. Sidebar search Searches a user's chats, projects, images, documents, and Library files across all plans on web and mobile, including archived chats and connected Google Drive content, with type filters and Ctrl/Cmd+K. Visual Retrieval with PDFs Reads images, charts, and diagrams embedded in an uploaded PDF alongside its text when answering questions about the file.
Collaboration & Sharing GPT Sharing Shares a GPT at a chosen visibility level with specific users, groups, a workspace, or by link, at chat, view-settings, or edit permission. Group Chats Lets multiple people and ChatGPT share one conversation to collaborate together. Library sharing Shares Library files and folders with specific people (Viewer or Editor) or across the workspace, with access management in the sharing dialog and a Shared with me view. Project sharing Lets members share projects with teammates and collaborate over shared files, instructions, and chats. Projects Groups related chats, files, and custom instructions into a project workspace. Projects Organizes chats, files, and app sources into persistent project workspaces with project-scoped memory. Shared links Generates shared links to ChatGPT conversations, individual responses, and scheduled tasks, with workspace-restricted access, link management, and deletion controls. Shared Links Creates a shareable read-only link to a conversation that others can view and continue in their own account. Shared Projects Lets Business, Enterprise, and Edu members share a project’s chats, files, and instructions with teammates and work in it together. Sites co-editing Lets workspace members co-edit Sites, with owners controlling publishing, settings, and editor access. Sites external sharing Lets Site owners share live Sites with named external viewers who sign in without joining the workspace.
Agentic Execution AI phone support Answers general product questions over the phone through an AI agent, on per-country numbers, with no account required. ChatGPT agent Completes complex online tasks on the user's behalf by combining research across web, files, and connected sources with actions like filling forms. ChatGPT Work Lets a user delegate longer, multi-step tasks to ChatGPT, gathering context from files, plugins, and connected sources, using approved tools, and producing a reviewable deliverable such as a document, spreadsheet, or presentation, running locally or, when chosen, in the cloud so it continues after the app closes. ChatGPT Work Runs longer agentic tasks that research, work across connected apps and files, and produce documents, spreadsheets, presentations, reports, and Sites. Cloud Browser Completes a task on supported public websites in a remote browser, pausing for confirmation before consequential actions. Data plugin Analyzes connected business data in ChatGPT Work and Codex: answers business questions, investigates changes, and builds interactive dashboards and reports using team metric definitions and business context. Memory-Aware Tool Calls Lets the model optionally include a user's saved ChatGPT memory as context when selecting or parameterizing a tool call, under user control. Scheduled and webhook-triggered tasks Runs prompts on a schedule or in response to events such as Gmail, Slack, and GitHub activity, and lets members share tasks within a workspace. Scheduled Tasks Runs one-off or recurring ChatGPT tasks on a schedule from a dedicated Scheduled page and notifies the user by push or email when a task completes. Workspace agents Lets teams build, share, and schedule agents that own repeatable workflows across connected apps and Slack.
Content & Creative Generation App Display Modes Renders an app through defined ChatGPT UI surfaces, including inline cards and carousels, fullscreen views, or a persistent picture-in-picture window. Canvas Opens a shared editable document or code workspace where a user and ChatGPT revise the same text or code with inline edits and code execution. Canvas Provides a shared writing and coding surface with inline editing, Python execution, previews, and sharing. ChatGPT Images Generates images from a text prompt and edits generated or uploaded images, including selection-scoped edits, added text, and transparent backgrounds. Code Blocks Writes and edits code inline in a chat response, previews diagrams and mini apps in place, reviews changes side by side, and exports generated flowcharts and visualizations as images. Document, Spreadsheet, and Presentation Creation Creates or edits documents, spreadsheets, and presentations in ChatGPT Work from instructions, attached source material, a reference file, or a reusable template. Image generation Generates and edits images from text prompts, with template-based creation, sketch-to-image on mobile, aspect-ratio control, an area editor, and shareable prompt links. Image Library Collects images a user has generated or uploaded in ChatGPT into a browsable personal library. Sticker Packs Turns a prompt, photo, or template into a set of stickers a user can download or add to iMessage or WhatsApp. Writing Blocks Drafts and edits long-form documents inline in a chat response, with a full-screen editor, in-document table of contents, download, undo/redo, and saving to the Library.
Voice & Multimodal ChatGPT Voice (Desktop Task Delegation) Lets a user talk through and coordinate multi-step Chat, Work, and Codex tasks in the ChatGPT desktop app, starting, checking, or steering work in other threads and sharing on-screen context via macOS appshots. Image Input Lets a user attach images to a message so ChatGPT can analyze them and answer questions about them. Pronunciation Guidance Provides pronunciation guidance for words and phrases in voice mode. Record Mode Records and transcribes a spoken meeting or session in the macOS desktop app and summarizes it. Voice Provides real-time voice conversation powered by GPT-Live in Chat and the desktop app, with plan-based model choice (GPT-5.6/GPT-6 Astra), usage limits, and per-minute credit or USD pricing. Voice Conversations Lets a user speak with ChatGPT and hear spoken replies inside a chat, with a Live option that listens and speaks simultaneously and can use web search, memory, and visual widgets. Voice Dictation Transcribes spoken input into a text prompt via dictation. Voice Live Translation Provides real-time spoken translation between languages in voice mode. Voice with Video and Screen Share Lets voice mode see live video from the camera or a shared screen during a conversation.
Data Movement Data Export Lets a signed-in Free, Plus, or Pro user request a downloadable archive of their conversations and account data from settings or the Privacy Portal. File Library Collects files a user has uploaded to ChatGPT into a browsable library reusable across chats. File Upload/Download in Widgets Lets a widget upload a user file, pick an existing file from the ChatGPT file library, or fetch a temporary download URL for a file reference. File Uploads Attaches files to a ChatGPT message for analysis across supported document, text, and code formats, subject to per-message and per-plan limits, with large pastes converted into attachments. Library Stores and reuses files uploaded to or created in ChatGPT under workspace retention policies, with Compliance API endpoints. Workspace Data Export Lets an Edu workspace owner grant members, by default or by custom role, the ability to request and download an export of their own workspace data.
Developer Tooling & Ergonomics ChatGPT Sites Builds, previews, and deploys lightweight full-stack web apps and internal portals from ChatGPT with hosted URLs and workspace-internal access. Custom App UI Widgets (MCP Apps Bridge) Renders a developer-supplied HTML/JS component inside a ChatGPT-hosted iframe that exchanges JSON-RPC messages with the host. Interactive code blocks Lets users write, edit, and preview code, diagrams, and mini apps directly in chat with split-screen review. Tool Definitions & Safety Annotations Requires each MCP tool to declare a name, description, input/output schema, and readOnly/destructive/openWorld hints the model uses to decide when and how to call it. Widget State Management Defines a three-tier state model spanning authoritative server data, ephemeral UI state in the widget, and durable cross-session state in the developer's backend.
Usage Analytics Impact surveys Collects and exports self-reported outcome surveys, including admin-created on-demand surveys for Enterprise and Edu. Impact Surveys Collects self-reported outcomes from workspace members through in-product surveys and aggregates the responses into the Impact section of workspace analytics. Personal Analytics plugin Lets members ask about their own Work and Codex activity and privacy-protected aggregate workspace trends. Workspace analytics Gives admins workspace-level analytics on adoption, engagement, benchmarks, task insights, and group-level breakdowns. Workspace Analytics Reports workspace adoption and engagement to admins through trend views, industry benchmarks, per-user tables, SCIM-group breakdowns, and CSV export.
Memory & Personalization Custom Instructions Lets a user set standing preferences about themselves and how ChatGPT should respond, applied across conversations. Improved memory Uses relevant context from past chats to keep memory current, with memory summaries, sources, and user correction controls. Memory Remembers details and preferences from past conversations and applies them in future chats, under user control. Personalities Lets a user pick from preset assistant personalities that change ChatGPT's tone and style.
OpenAI's agentic coding product, available as a CLI, cloud service, IDE extension, GitHub integration, App Server/SDK, and security-scanning tool.
Software capabilities Developer Tooling & Ergonomics Appshots Captures the frontmost macOS app window as chat context for Codex. Browser Annotations / Visual Feedback Lets a user click to comment on elements of a page Codex is working with. Chronicle (Screen-Context Memory) An opt-in macOS research preview that augments Codex's memory with recent on-screen context so a user restates less context when prompting. CLI Customization Lets a user customize themes, shell completions, and the prompt editor in the Codex CLI. Code Review (Review Pane) Provides an inline-comment code review pane with staging and reverting of changes. Codex CLI Runs Codex as a terminal interactive agent, with a non-interactive exec mode for scripting. Codex GitHub Action (CI/CD) Runs Codex inside a GitHub Actions workflow via the openai/codex-action to apply patches, post reviews, or gate changes on Codex-driven checks. Codex Micro Hardware Controller Provides a physical control surface with Agent Keys, Command Keys, an analog stick, and a dial for switching between and acting on parallel Codex chats in the ChatGPT desktop app, including Bluetooth pairing, Voice Chat mic mode, custom action/skill assignments, and four dial modes. Codex Security CLI & SDK Public @openai/codex-security CLI and TypeScript SDK for running security scans from the terminal, CI, or custom tooling, versioned separately from the hosted plugin. Codex Security Workbench Provides a desktop-app UI for starting and tracking Codex Security scans, browsing findings across repositories and scans, and inspecting per-repository scan history, alongside the existing Codex Security CLI and plugin. Codex-Spark Model Provides a near-instant, less-capable model tier for ChatGPT Pro users. Desktop App Provides a native desktop application for Codex. Desktop Custom File Handlers Lets a user register additional editor targets in the ChatGPT desktop app's Open in menus via desktop.custom_file_handlers entries in config.toml. Dev Containers Support Runs Codex inside VS Code Dev Containers or Docker for an additional layer of isolation. Environments (Local / Worktree / Cloud) Lets a Codex chat run in an isolated local worktree or a configured cloud environment with setup and maintenance scripts, dependencies, and environment variables, separate from the primary local project directory. Experimental Context Management Activates token-budget context, history notes, and a new_context tool for eligible ChatGPT Plus/Pro/Pro Lite sessions on the Codex backend, opted into via features.context_management.experimental_mode. File Preview and Annotation Review Previews and lets a user annotate documents Codex generates. Hooks Runs user-configured matcher groups on lifecycle events such as PreToolUse, PermissionRequest, PostToolUse, compaction, session and subagent boundaries, UserPromptSubmit, Stop and Interrupt. Hooks (Agentic Lifecycle Scripts) Lets a user inject custom scripts at defined points in the Codex agentic loop, such as logging, validation, or auto-generated memories. IDE Extension Provides a Codex extension for IDEs such as VS Code. Integrated Terminal Provides a terminal panel scoped to the current project or worktree inside a ChatGPT desktop app chat, with its output readable by the model. Mobile Remote Access Lets a user start and monitor Codex cloud tasks remotely from the ChatGPT mobile app. Multi-Folder Local Projects Lets a local project in the ChatGPT desktop app attach multiple folders, with one primary folder used for new chats, Git operations, and AGENTS.md/skills/config discovery while secondary folders remain available for file search, reading, and editing. Notifications Lets a user configure when and how they are alerted to turn completions, approval requests, and questions across ChatGPT/Codex surfaces. Pets (Animated Companion) Shows an optional animated companion that floats above other windows or appears in-app to reflect a chat's running, needs-input, ready, or blocked status. Record and Replay Records a Codex session so it can be replayed or reviewed. Remote Connections Lets a user access and steer a Codex or ChatGPT chat running on a connected Mac, Windows device, or SSH host from the ChatGPT mobile app or another desktop install. Slash Commands (CLI TUI Controls) Provides keyboard-driven in-session commands inside the Codex CLI terminal UI. Speed / Fast Mode Trades higher credit cost for faster Codex responses. Vim Mode (TUI Composer) Lets a user edit the CLI composer with Vim-style keybindings, including undo/redo, search motions, and preserving drafts, pasted content, and attachments. Web Search Tool Modes Lets Codex use cached or live web search results depending on the configured mode. Weekly What's New Digest Publishes a weekly digest of ChatGPT and Codex feature changes. Windows enterprise deployment Enterprise deployment guidance for the ChatGPT Windows app covers Intune Store and offline MSIX (LOB and Win32) deployment, SCCM packages, generic MSIX deployment via other management tools, user self-install, and troubleshooting for error 0x80073D28. Windows Support (Native Sandbox & WSL) Runs the ChatGPT desktop app, CLI, and IDE extension natively on Windows via a PowerShell-native sandbox or, alternatively, inside WSL2.
Agentic Execution AGENTS.md Custom Instructions Loads layered global and project instruction files that are always injected into Codex's context. Build Skills (Agent Skills) Packages reusable, task-specific instructions that Codex can invoke as a unit. Built-in Browser (Computer Use) Provides an in-app browser that Codex can navigate and interact with as part of a task, with managed-desktop controls that restrict origins, uploads, downloads and developer access. Chrome Extension (Computer Use in Chrome) Lets Codex control the user's real Chrome browser profile and sessions. Code Mode Exposes built-in and MCP tools to the agent as a nested, code-callable surface executed in a sandboxed runtime instead of as flat per-turn tool calls. Codex Memories Lets local Codex clients build and recall a separate on-device memory store of useful context from earlier work, independent of ChatGPT web memory. Computer Use Lets Codex see, click, and type in desktop applications on macOS and Windows, governed by managed and local allow/deny rules for native apps by macOS bundle id, Windows AUMID or signed executable identity. Context management Manages the agent's context across long sessions, with an experimental mode that preserves accumulated details using notes and searchable history instead of repeated compression into a single summary. Custom Agents Defines user-configured named subagents with their own model and instructions. Fast mode Increases supported model speed by 1.5x at a higher credit consumption rate, toggled via /fast commands or the service_tier setting. Goal Mode Lets Codex pursue a high-level goal by planning and executing multiple steps autonomously. Local/Cloud Handoff Lets a Codex task move between local and cloud execution while preserving context. Scheduled Tasks Runs recurring background Codex tasks, optionally in an isolated worktree. Skills Maintains a catalog of available skills within a configurable token budget, defaulting to 2% of the model's context window and capped at 10,000 tokens. Subagents / Subagent Workflows Runs parallel specialized agent threads to divide a larger task via a reworked [agents] config schema: agents.enabled disables multi-agent tools, agents.max_concurrent_threads_per_session (renamed from max_threads) caps concurrent threads, and agents.default_subagent_model / default_subagent_reasoning_effort set defaults; max_depth, display-nicknames, and the CSV-batch workflow were removed.
Agent Safety Controls Agent Internet Access (Codex Cloud) Applies per-environment internet access controls and allowlists to Codex Cloud tasks. Approval Policies Determines when Codex must pause and ask a user before acting, configurable per app and per connected account within an app. Automatic Approval Review Runs a reviewer agent to assess risk before executing approval-gated actions. Browser Use origin policies Applies per-origin allow/deny policies to agent-driven browser work, covering access, uploads, downloads, full CDP access, automatic review, persistent approvals and approval lifetime, with deny-wins resolution across managed and local policy sources. Computer Use app access policies Controls which native desktop apps agent-driven Computer Use may operate, via a default app-access policy plus per-app rules keyed by macOS bundle identifiers, Windows packaged-app AUMIDs, and signed Windows executables. Credential Storage Lets a user configure local credential storage between an OS keyring and a file. Execpolicy Prefix Rules Uses Starlark rules to control which commands run outside the sandbox. Locked Use Lets an approved Computer Use task continue on a locked Mac through a macOS authorization plugin that temporarily unlocks the machine for an active turn, relocking when local keyboard or pointer input is detected. Managed network policy Routes sandboxed local commands through an administrator-managed network proxy governed by a domain allow/deny map with wildcard matching and deny-wins resolution, with browser tools separately checking managed denies and exclusive allowlists. Network Access Controls Applies domain allow/deny rules to network access from sandboxed commands via a network proxy. Safety Monitoring and Paused Tasks Runs asynchronous monitoring of model behaviour during a Codex or ChatGPT Work task and pauses the task when it detects potentially unsafe behaviour, surfacing findings and a resume option on clients that support them. Sandbox Modes Restricts what Codex's agent can do on disk and network via OS-level enforcement. Shell Environment Policy Controls which environment variables Codex passes to spawned subprocess commands, via a canonical include/exclude filter map or legacy allow/deny arrays.
Integrations & Extensibility Amazon Bedrock Model Provider Support Routes Codex model requests through AWS Bedrock as an alternate model provider. Build an App (Apps SDK / MCP-backed apps) Gives Codex a framework for building MCP-backed ChatGPT apps. Build Plugins (Plugin Marketplace) Packages skills, MCP servers, and hooks into installable plugins distributed through a marketplace. Codex Code Review in GitHub Runs an automated Codex review on GitHub pull requests via an @codex review comment, guided by Code Review Rules defined in the repository's AGENTS.md files. Codex via Slack & Linear Lets a user delegate a coding task to a Codex cloud chat by mentioning @Codex in a Slack channel or thread, or a Linear issue or comment. MCP server configuration Connects and tunes MCP servers, covering startup commands, HTTP headers and header-helper commands, scopes, per-tool approval overrides and per-tool output token limits, plus a shared startup grace period for optional servers. Plugin continuous review Published plugins receive continuous review: OpenAI periodically fetches the MCP server's tools, removes deleted tools as soon as a scan detects them, and makes new or changed tool definitions live once automated checks pass without a new plugin version. Plugin marketplace configuration Defines Git or local plugin marketplaces and per-plugin enablement in system, cloud-managed, user, or trusted-project config.toml, separate from enforced requirements. Plugin multi-account connections Plugins let users connect multiple accounts (for example personal and work) to one plugin, with each tool call routed via the selected account's credentials and a designated authenticated profile tool marked with _meta["openai/profile"]: true returning a stable opaque profile ID.
Access & Governance Cloud-managed config defaults Delivers workspace cloud-managed config.toml defaults in the configuration precedence order, including plugin marketplace and plugin enablement defaults. Codex Authentication (ChatGPT Sign-In / API Key) Offers two sign-in modes for Codex, ChatGPT account sign-in or an API key, with different billing and admin controls. Codex service accounts Provides workspace-scoped machine identities that run automated Codex workflows separately from personal access tokens. Compliance API & Audit Events Exposes auditable Codex activity records for security, legal, and governance investigation workflows, with organizational retention and legal-hold support. Managed Browser Use & Computer Use requirements Lets administrators restrict agent-driven Browser Use and native-app Computer Use from requirements.toml, including a master switch, browser-pane and settings-import controls, history access, automatic review, persistent approvals and locked-use enablement. Managed login & authentication requirements Lets administrators restrict ChatGPT login to listed workspace IDs, allow only ChatGPT or API login methods, enforce the ChatGPT base URL, and pin the CLI credential store via requirements.toml or macOS MDM. User Lifecycle Management Provisions, updates, and deprovisions employee access to a Codex/ChatGPT Work workspace from a connected identity provider group as people join, change teams, or leave.
Evaluation & Observability Browser Use OpenTelemetry events Emits OpenTelemetry events for Browser Use capability checks — site access, uploads, and downloads — recording the decision and its source, exported to the customer's OpenTelemetry endpoint when configured. Codex Analytics API Exposes aggregated Codex usage and activity metrics for a ChatGPT workspace for automated reporting, distinct from raw audit-log records. OpenTelemetry Monitoring Provides opt-in telemetry export for audit and compliance monitoring of Codex activity. Usage Insights (Admin Console) Admin Console Insights dashboard groups ChatGPT Work and Codex usage into use cases and tasks, showing messages, credit share, active users, model/reasoning/speed breakdowns, plugin and skill invocations, and code review measures like PRs reviewed and issues found. Usage Insights for ChatGPT Work in Codex Provides usage and credit-consumption insights across ChatGPT Work and Codex so admins can assess the value of their team's workflows.
Billing & Monetization Codex Access Tokens Issues enterprise non-interactive tokens for automating Codex outside of a signed-in session. Rate Limit Reset Banking Banks unused rate-limit allowance from one reset window for later use. Usage Profiles Lets a user switch between usage profiles that trade off model capability and rate-limit consumption.
Developer Platform APIs Codex App Server Runs a local server over stdio, WebSocket, or Unix socket that programmatic clients can drive, including ephemeral in-memory thread/fork, isPinned thread filtering/metadata, and app/installed + app/read methods. Codex MCP Server Mode Lets Codex itself run as an MCP server that other tools can call. Codex SDK (TypeScript library) Provides programmatic control of Codex threads from a TypeScript library.
Data Movement Import from Claude Code Migrates configuration and chat history from Claude Code into Codex.
OpenAI's advertising platform for running, targeting, and measuring ad campaigns shown within ChatGPT.
Software capabilities Ads Measurement & Targeting Ad Performance Insights Returns impressions, clicks, and spend metrics at ad-account, campaign, ad-group, or ad scope, with time bucketing, segmentation, filtering, and sorting. Campaign Geo Targeting Restricts where a campaign's ads can serve by country, region, or designated market area using a location lookup endpoint. Conversion Event Setup Provisions web pixels, server-side API keys, and event definitions used to measure ad conversions. Conversions API (Server-Side Tracking) Sends purchase and other conversion events from a server to OpenAI for more reliable ad attribution than browser tracking alone. Custom Audiences Creates a custom audience for ad targeting from hashed email or phone identifiers, entered inline or uploaded as a file, for use in campaign targeting. Event Quality Scores each pixel or data source daily on how usable its conversion events are for measurement and raises warnings for missing identifiers, delayed server events, unmatched ad clicks and concentrated attribution. Image Tag Fallback Tracking Sends a single conversion event via a hidden 1x1 image request when JavaScript is unavailable. Measurement Partner Integrations Connects a third-party web or mobile measurement partner to a ChatGPT Ads account so conversion events flow in for attribution and reporting. Measurement Pixel Installs a browser SDK on a merchant site that measures website conversions after a user clicks an ad in ChatGPT.
Content & Creative Generation Ad Campaigns Creates and manages top-level ad campaigns with budget, schedule, a bidding type (impressions, clicks, or conversion-optimized oCPC bidding tied to a conversion event), and location targeting via the Advertiser API. Ad Creative Creates and manages individual ads with chat-card or product-template creative, subject to an approval review workflow. Ad group context hints Lets advertisers add ad-group-level context hints describing their product, audience, and use cases to help ChatGPT Ads systems judge ad relevance. Ad Groups Groups ads under a campaign with bidding configuration and free-form audience or context hints that guide when ads appear. Ads Manager Provides a self-serve web console for creating campaigns through a guided workflow or bulk upload, monitoring delivery in tables and charts, and managing account members, API keys, billing, and change logs. Bulk API Creates or updates up to 1,000 campaigns, ad groups, and ads in one asynchronous job, polled for per-operation results. Sponsored Agents Offers an advertising format, currently in limited alpha, that lets people chat with an AI representative of a business from an ad in ChatGPT.
Reporting & Insights Conversion attribution reporting Reports conversions through a dedicated endpoint that separates click-through and view-through attribution and computes attributed purchase value and ROAS. Flexible attribution windows Lets advertisers choose a 7-, 14-, or 30-day click-through reporting window and a 0- or 1-day view-through window in Ads Manager, with eligible view-through conversions included in the Conversions total when enabled. Insights platform segmentation Splits Insights delivery metrics by ChatGPT platform (iOS, Android, desktop and mobile web) via the platform segment. Insights reporting Queries aggregated delivery metrics by account, campaign, ad group, or ad with time granularity, segments, filters, and zero-impression includes. Product-segmented insights Breaks insights down by feed product, reporting per-item impressions, clicks, spend, CTR, and carousel product-card engagement for product-feed campaigns. Serving issues diagnostics Exposes explicit serving issues, review outcomes, and low-bid warnings on ad, ad-group, and campaign retrieval so advertisers can diagnose why active resources are not delivering.
Targeting & Audiences Context hints Attaches up to 2,000 free-text hints per ad group describing products and use cases to inform ad matching. Custom audience management Creates, merges and replaces custom audiences from uploaded identifier files, and lists, resumes or cancels the asynchronous membership ADD/REMOVE operations that apply them. Custom audience targeting Includes or excludes ready custom audiences on campaigns and applies fixed-bid multipliers for matching users. Geographic targeting Targets campaigns by country codes and location IDs with inclusion and exclusion lists up to 2,500 entries, using geo lookup and a downloadable catalog. Platform targeting Restricts campaign delivery to chosen ChatGPT surfaces — iOS and Android native apps, desktop web, iOS web, Android web, or all web platforms via the catch-all web value.
Ads & Creative Ad creative management Uploads images and creates, updates, and reviews chat-card ads with title, body, image, and landing-page URL. Ad previews Renders ad previews and surfaces serving issues and review status without confirming delivery eligibility. Landing page tracking parameters Attaches UTM query-string templates at campaign, ad-group, or ad level with defined precedence for duplicate parameters. Product-ad templates Renders ads from a product_ad_template creative whose macro tokens are filled with the selected feed product's brand, title, description, price, image, and destination URL at serving time.
Bidding & Budgets Account spend limits Manages dated spend-limit windows and an evergreen account-wide daily billable spend cap on postpaid invoice billing accounts, with revision-guarded edits and enforcement metadata. Campaign budgets Sets daily or lifetime spending limits on campaigns in account-currency micros, shared across the campaign's ad groups. Fixed bidding Sets manual per-impression or per-click bids on ad groups in currency-scaled micros, with bid_too_low delivery guidance. Maximize Results bidding Selects the campaign-compatible AutoBid strategy by default when no bidding strategy or max bid is set on an ad group.
Billing & Monetization Bid strategies Lets advertisers set bid controls including the Maximize results strategy, which automatically adjusts bids toward the campaign goal and is the default for eligible new ad groups. Budget pacing Paces Ads Manager campaign delivery against daily and campaign total budgets over the day or the campaign schedule. Conversion-optimized impression billing (oCPM) Runs conversion-optimized campaigns billed per thousand impressions (oCPM, beta) toward a selected conversion event, alongside the existing click-billed oCPC option. Invoice tax IDs Adds or updates supported VAT and tax IDs on the Ads Manager billing profile, applied to future invoices.
Conversion Measurement Conversion event monitoring Inspects a sample of recent Pixel and server events to verify that conversion integrations are delivering data. Conversion event settings Defines standard or custom conversion events with attribution windows and attaches them to campaigns for reporting and optimization. Conversions API Sends server-side conversion events with deduplication against Pixel events using shared Pixel IDs and event IDs. Measurement Pixel Sends browser conversion events from advertiser websites through the Measurement Pixel with consent controls and event IDs.
Product Feeds Feed delta updates Updates individual product fields such as price and availability in an existing feed through the Delta API without re-uploading the full catalog. Feed upload monitoring Reports feed upload history with processing status, accepted and rejected row counts, and field-level diagnostics for resolving ingestion problems. Product feed management Creates and manages product feeds for an ad account, including country configuration, SFTP access credentials, feed listing with product counts, and archiving. Product sets & filters Selects which feed products an ad group can use through attribute filters such as brand, category, and price, with a preview query to check matching products before saving.
Account Administration Account branding Updates the advertiser's brand name, URL, and icon and tracks the brand-review status that gates delivery. Account status control Pauses and activates the ad account and inspects its configured status, brand review, and account-integrity review state. Additional ad account creation Lets eligible self-serve advertisers create additional ad accounts from an existing credit-card-billed account in Ads Manager, capped at 10 accounts per login.
Campaigns & Ad Groups Ad group management Creates and manages ad groups that apply a shared bid configuration and context hints to a group of ads within a campaign. Campaign management Creates, retrieves, lists, updates, and pauses, activates, or archives campaigns with objectives, schedules, budgets, and targeting via the Ads API.
Data Movement Creative Asset Upload Uploads images by URL or binary upload for use as ad creative or account brand assets. Product Feeds (Catalog Ads) Connects a merchant's product catalog to a campaign so ads are auto-generated from eligible catalog items at serve time.
Bulk Operations Bulk mutation jobs Submits asynchronous jobs of up to 1,000 campaign, ad-group, and ad operations with validation, partial failure, and idempotent retries.
Client Apps & Surfaces Platform targeting Targets campaign delivery across five platforms (Android app, Android web, Desktop web, iOS app, iOS web) with per-platform segmentation in Insights.
Integrations & Extensibility Self-serve Ads Manager setup in ChatGPT Lets eligible business advertisers set up an Ads Manager advertiser account from inside ChatGPT, selecting or creating an organization and confirming advertiser settings before finishing billing in Ads Manager.
Platform Trust & Compliance Ad Account Brand Review Gates ad serving on an approved brand review of the account's name, URL, and favicon.
The programmatic API and developer platform (Responses, Assistants, Realtime, Agents SDK, Agent Builder, ChatKit, fine-tuning, embeddings, evals) for building AI-powered applications.
Software capabilities Developer Platform APIs .NET SDK Provides an official OpenAI SDK for .NET. Admin API SDK Support Adds Admin API support to the official SDKs for managing organization resources. API Webhooks Delivers asynchronous event notifications to a developer endpoint via webhooks. Background Mode Runs a Responses API generation asynchronously so long-running tasks avoid client-side connection timeouts, with the response polled for status until it completes. Batch API Processes a batch of API requests asynchronously at reduced cost. Cancel In-Progress Response Lets a client cancel an in-progress model response via the API. Chat Completions Logprobs Returns per-token log probabilities for Chat Completions outputs. Chat Completions Tools Parameter Lets Chat Completions requests declare callable function tools the model can invoke. Configuration Update Items Adds a configuration_update conversation item that changes response configuration, currently reasoning effort, for all subsequent responses until replaced. Context Compaction Reduces context-window usage via server-side compaction or a standalone compact endpoint. Conversation State Management Persists multi-turn conversation state via a durable Conversations API object or by chaining previous_response_id. Custom Tools (Free-Form Input) Declares a tool that receives an arbitrary text string rather than a JSON-schema argument object, optionally constrained by a Lark or regex context-free grammar. Deno SDK Support Provides official OpenAI SDK support for the Deno runtime. Embeddings API Generates vector embeddings from text for search, clustering, and retrieval. Go SDK Provides an official OpenAI SDK for Go. Java SDK Provides an official OpenAI SDK for Java. Mid-Turn Steering Lets a client send additional instructions over a WebSocket connection to a response that is still generating, so the model can incorporate corrections or changed requirements without waiting for the turn to finish. Parallel Function Calling with Built-in Tools Lets GPT-5-and-later models call custom functions in parallel in the same turn that built-in tools are available, though built-in tools themselves cannot be included in the parallel batch. Predicted Outputs Reduces generation latency by supplying an expected output the model can diff against, such as for code edits. Programmatic Tool Calling Lets the model call tools by writing and running code that invokes them programmatically. Prompt Cache Diagnostics Compares prompt-cache reuse against a previous response and identifies reasons for cache misses in the Responses API. Prompt Caching Automatically caches repeated prompt prefixes to reduce latency and cost on later requests; GPT-5.6+ models support an explicit prompt_cache_breakpoint/prompt_cache_key pairing and an explicit-only caching mode. Python SDK Provides an official Python client library, generated from the OpenAI OpenAPI spec, with synchronous and asynchronous clients for calling the REST API. Reasoning Effort Control Lets a request set how much reasoning effort a reasoning model spends before answering. Responses API Creates model responses over a single stateful endpoint that unifies text generation, function calling, hosted tools, and conversation state, and is the API OpenAI directs new integrations to. Responses API Phase Field Adds a phase field to Responses API streaming events to indicate the current generation stage. Responses API Tool Search Lets the Responses API select from a large tool set by searching tools at request time. Responses API WebSocket Mode Lets clients stream a Responses API session over a WebSocket connection, with native client helpers in the Python, JavaScript and Ruby SDKs. Reusable Prompts Lets developers create, version, and reuse named prompt templates across requests. Ruby SDK Provides an official Ruby client library, generated from the OpenAI OpenAPI spec, for calling the REST API, retrying only requests whose bodies are replayable and, after dispatch, only idempotent methods or requests carrying an Idempotency-Key. SDK Amazon Bedrock Provider Routes OpenAI SDK requests through Amazon Bedrock via a pluggable provider that derives the regional Mantle endpoint and authenticates with a Bedrock API key or the AWS credential chain with SigV4 signing. Streaming Usage Stats Includes token usage statistics in the final chunk of a streamed Chat Completions response. Structured Outputs Constrains model output to a supplied JSON Schema so responses always match the structure. Tool Call Output Media Types Lets tool call outputs return typed media such as images in addition to text. TypeScript / JavaScript SDK Provides an official TypeScript/JavaScript client library, generated from the OpenAI OpenAPI spec, for calling the REST API from Node.js or the browser.
Agentic Execution Agent Builder Provides a visual canvas to build, connect, and deploy multi-step agent workflows. Agent Orchestration & Handoffs Lets an Agents SDK application orchestrate multiple agents through handoffs and agents-as-tools patterns. Agent Skills (Hosted & Local Bundles) Lets a caller upload and reuse versioned skill bundles, each a SKILL.md manifest plus files, in hosted container-based or local shell execution environments. Agents API Runs managed Codex-harness agent sessions with durable state, streaming and webhook progress, context compaction, multi-agent delegation, and hosted or self-hosted sandbox environments. Agents API Agent CRUD Creates, lists, retrieves, updates and deletes reusable project-scoped agents whose persisted configuration covers model, instructions, reasoning, service tier, text format, metadata, multi-agent settings and tool definitions. Agents API Environment Templates Stores reusable environment templates that provision a fresh OpenAI-hosted environment for each agent session, with full CRUD on the template resource. Agents API Environments Provisions and manages first-class execution environments for agent sessions, either OpenAI-hosted or self-hosted, with files, packages, plugins, skills and network policy. Agents API Session Artifacts Publishes immutable artifacts from completed hosted session turns and exposes list, retrieve, content download and delete endpoints for them. Agents API Session Settings Update Updates a saved agent's configuration and changes model, reasoning effort, or service tier for an existing session in place via POST /v1/agents/sessions/{session_id}, preserving conversation history. Agents API Session Webhooks Delivers signed webhooks for agent session state changes — created, action required, in progress, idle, and failed — so applications can react without holding an event stream open. Agents API Sessions Runs long-lived agent sessions that keep configuration, conversation, and saved work across turns, with streaming events, mid-turn steering, cancellation, and item retrieval. Agents API Subagents Spawns, coordinates, interrupts, resumes and closes subagents within an agent session, with list/retrieve endpoints for subagents, their turns and their items. Agents API Turns & Event Streaming Represents and lists agent session turns as first-class resources with status, usage and error fields, plus streaming session events for turn lifecycle and output deltas. Agents SDK Provides an open-source Python and TypeScript framework for building agent loops with tools, handoffs, guardrails, sandboxes, memory controls, and tracing, over an inspectable harness. Agents SDK Sandboxes Gives an Agents SDK agent an isolated Unix-like execution environment with a filesystem, shell, installed packages, mounted data, and controlled external access. Apply Patch Tool Lets a model create, update, and delete files by emitting structured diff operations that the calling application applies. Async Tool Calling Lets a function or custom tool declared with async: true keep running after the model's turn continues, so the application supplies the tool's output in a later request instead of blocking the turn on it. Code Interpreter Tool Provides a hosted code interpreter tool that runs Python in a sandbox from the Responses API. Computer Use Tool Lets a model control a computer or browser by emitting mouse and keyboard actions via the computer-use tool. Deep Research Runs agentic models that browse, search files or MCP sources, and synthesize analyst-grade research reports. File Search Tool Provides a hosted file-search tool that retrieves relevant chunks from uploaded documents for grounding. Hosted Shell Tool Runs model-issued shell commands in a terminal environment inside an OpenAI-provisioned container reached through the Responses API, with optional container networking. Local Shell Tool Lets an agent loop run shell commands on a caller-controlled machine. Responses API Multi-agent Lets a Responses API model spin up and coordinate parallel subagents and synthesize their work into one final response. Web Search Tool Lets models search the live web as a built-in tool and return cited results.
Access & Governance Admin API Hosted Tool Permissions Lets admins control which hosted tools projects may use via the Admin API. Admin API Key Rotation Lets an admin programmatically rotate organization and project API keys. Admin API Model Allowlists Lets admins restrict which models a project or organization can use via the Admin API. API Key Expiration Lets administrators set expiration dates on project API keys and enforce a maximum key lifetime at the organization or project level. API Key Governance Controls Lets organization and project administrators restrict new API key creation to service-account keys or user-owned project keys, or disable new key creation entirely, with organization restrictions taking precedence. API Key Permissions Sets an API key to All, Read Only, or Restricted with per-resource scopes at creation or afterwards, so a key can be denied read or write access to specific endpoints. Audit Log API Exposes an API to retrieve organization audit-log events. Fine-Tuned Checkpoint Sharing Permissions Lets an organization grant or revoke other projects' access to a fine-tuned model checkpoint. Human-in-the-Loop Tool Approvals Pauses an agent run before a sensitive tool call until a human approves or rejects it. Organization Member Invites Lets organization admins create, list, and revoke pending invitations for new organization members via the Admin API. Organization Verification Gates certain products, features, models and programs behind business verification, identity verification, or both for an API organization. Project Rate Limit Configuration Lets organization admins view and modify per-project request- and token-per-minute rate limits via the Admin API. Project Service Tier Restrictions Restricts which API service tiers a project may use, rejecting a request whose explicit or auto-resolved service_tier is not on the project's allowed list. Projects & Organization Hierarchy Organizes API usage into a hierarchy of organizations and projects with scoped keys and settings. RBAC & SCIM Group Sync Provides organization and project role-based access control with custom roles and identity-provider-synced groups. Self-Serve SSO Configuration Lets an admin configure SSO for their organization without contacting support. Service Accounts Lets admins create, update, and manage project-scoped service accounts as non-human identities with their own API keys, optionally expiring, separate from individual user accounts. Terraform Provider Manages OpenAI organization resources -- projects, users, groups, roles, service accounts, rate limits, and spend alerts -- as infrastructure-as-code through an official Terraform provider backed by the Administration API. Usage Tiers & Rate Limits Places an organization in a usage tier from its cumulative spend, which sets per-model request- and token-per-minute ceilings and the monthly usage cap, with headroom visible per model.
Platform Trust & Compliance API IP Allowlisting Restricts which source IP addresses or ranges may call the OpenAI API, applied per project or across the organization, rejecting other requests even when they carry a valid API key. Azure Private Link Lets Azure workloads reach regional OpenAI API endpoints through a private Azure Private Link endpoint instead of the public internet. Certificate Management (mTLS) Lets organizations upload, activate, deactivate, and manage TLS client certificates at the organization or project level for mutual TLS authentication to the API. Cybersecurity Capability Safeguards Monitors and can temporarily restrict access to high-cyber-capability models when suspicious activity is detected. Data Residency Controls Pins storage and/or processing of customer content to a selected region such as the US, EU, UK, Australia, Canada, Japan, India, Singapore, South Korea, or UAE. Data Retention Controls Lets approved organizations choose Zero Data Retention, Modified Abuse Monitoring, Eyes Off, or Safety Retention per organization or project. Data Sharing Program Lets an organization or project opt in, through data controls settings, to sharing feedback, evaluation and fine-tuning data, or API inputs and outputs, with OpenAI; sharing is off by default. Daybreak trusted access Gates reduced-refusal cyber model access through the Daybreak Blue and Red program, with reduced refusals on GPT-6 Astra for Daybreak Red customers and onboarding-confirmation-based provisioning. Enterprise Key Management (EKM) Encrypts customer content using customer-managed keys held in an external key management service. Fine-Tuning Safety Checks Automatically evaluates fine-tuned models across multiple safety categories and blocks deployment on failure. HIPAA Business Associate Agreement Provides a Business Associate Agreement, reviewed case by case, that lets a covered entity or business associate process protected health information through the API platform. Image Content Moderation Controls Lets developers configure moderation strictness for generated images and returns structured error codes for blocked generations. IP Egress Allowlisting Publishes OpenAI's outbound IP ranges for ChatGPT integrations and Codex cloud so customers can allowlist traffic. Moderation API Classifies text and image inputs for harmful content using the omni-moderation model, either via a standalone moderation endpoint or as inline moderation scores on a Responses/Chat Completions generation. Safety Identifier Accepts a stable hashed per-end-user string on requests so policy violations can be traced to one of a customer's users, superseding the earlier `user` parameter. Workload Identity Federation Exchanges a short-lived subject token from a supported external identity provider (AWS, Google Cloud, Microsoft Azure, GitHub Actions, SPIFFE, Oracle Cloud Infrastructure, or an X.509 client certificate) for an OpenAI access token without requiring static API keys.
Developer Tooling & Ergonomics ChatKit Provides an embeddable, customizable chat UI toolkit for adding an agent chat interface to an app. Direct Preference Optimization (DPO) Fine-Tuning Fine-tunes a model from pairs of preferred and non-preferred responses. Fine-Tuning Checkpoints Exposes intermediate training checkpoints from a fine-tuning job for evaluation and use. Fine-Tuning for Function Calling Fine-tunes a model on function-calling examples to improve tool selection. Fine-Tuning Seed Lets a fine-tuning job set a seed for reproducible training runs. GPT-4.1 nano Fine-Tuning Lets developers fine-tune the GPT-4.1 nano model on their own data. GPT-4o Fine-Tuning Lets developers fine-tune GPT-4o on their own data. Input Token Counting API Returns the exact input token count a request will consume, including images, files, and tools. Model Distillation Uses stored completions from a larger model to fine-tune a smaller model. Model/API Deprecation Policy Publishes advance-notice timelines and replacement guidance for retiring models and endpoints. OpenAI CLI Provides an official command-line tool for calling the OpenAI REST API from the terminal, distributed via Homebrew or Go install. OpenAI Developers Plugin Bundles the OpenAI Docs MCP server, an OpenAI API Platform connector, and guided API key setup into one plugin listed for ChatGPT, Codex, Claude Code, and Cursor. Playground Runs prompts, model settings and function calls against models in a browser console, with project-level prompts carrying version history, one-click rollback and prompt variables. Reinforcement Fine-Tuning Fine-tunes a reasoning model with reinforcement learning against a grader. Structured API Error Codes Documents standardized HTTP error codes and SDK exception types with remediation guidance.
Evaluation & Observability Agent Workflow Evaluation (Trace Grading) Grades captured agent run traces, including model calls, tool calls, and handoffs, against structured criteria. Agents API Trace Export Exports session traces as paginated OpenTelemetry Protocol (OTLP) JSON through GET /v1/agents/sessions/{session_id}/traces for ingestion into external tracing tools. Agents API Tracing Records session traces in the Logs dashboard with agent, generation, and tool spans — including MCP calls and subagent delegation — showing inputs, outputs, durations, and token usage. Agents SDK Tracing Provides built-in structured tracing of model calls, tool calls, handoffs, and guardrails, viewable in a traces dashboard. Evals Platform Provides a structured framework of datasets, metrics, and LLM-as-judge scoring for testing AI system performance. Evaluate External Models Runs evaluations against third-party models or custom model endpoints inside the OpenAI eval platform. Graders Provides a scoring API with string check, text similarity, score-model, code, and multigrader methods for evals and fine-tuning. Red Teaming Documents adversarial red-team testing, via open-source Promptfoo or OpenAI Red Teaming for enterprise, to surface unsafe or policy-violating model behavior before deployment. Safety Usage Dashboard Provides a dashboard of safety-related usage and moderation signals for an organization. Service Health Dashboard Publishes a live status dashboard of API and product service health and incidents. Usage Dashboard Reports token and request usage and cost per project and model in the platform console, including a Scale Tier detail page, restricted to organization owners and users granted the Usage Dashboard permission.
Billing & Monetization Admin API Spend Alerts Lets admins configure programmatic spend alerts via the Admin API. Budget Alerts & Auto-Recharge Sends budget threshold alerts and can auto-recharge prepaid credits. Fast Mode (formerly Priority Processing) Offers a higher-cost processing tier -- renamed from Priority Processing to Fast Mode on July 30, 2026 (service_tier: "priority" still accepted) -- for faster, more consistent latency, including up to a 2.5x speed increase for gpt-5.6-sol. Flex Processing Offers lower-cost inference for Chat Completions and Responses in exchange for slower, less-guaranteed response times. Hard Spend Limits Lets an organization or project enforce a hard monthly spend limit that returns a 429 error with a specific organization_spend_limit_exceeded, project_spend_limit_exceeded, organization_usage_limit_exceeded, or credit_balance_exhausted code once reached. Realtime API Cost Management Manages Realtime API session cost via automatic prompt caching, configurable truncation and retention ratio, and a cheaper mini model tier. Scale Tier Reserves a purchased number of input and output token units per minute against one dedicated model snapshot for a minimum of 30 days, added and removed from organization settings. Usage API Exposes an API to programmatically retrieve API usage and spend data. Usage Credit Purchases Lets an organization buy prepaid API usage credits.
Integrations & Extensibility Agents API MCP Connections Connects MCP servers to agent sessions over HTTP from OpenAI or from the session environment, or starts stdio servers in the environment, with vault or inline authentication and tool allowlists. Agents API Plugins Packages skills and MCP server configuration into plugin archives that load into self-hosted or OpenAI-hosted agent environments and reuse via environment templates. Agents API Vaults Stores MCP connection credentials — bearer tokens and OAuth grants with refresh — in vaults attached to sessions so agents use authenticated tools without receiving secret values. Agents API Web Search Gives agent sessions a built-in web search tool with live, cached, and disabled modes plus context size, domain allowlist, and location settings. Agents SDK MCP Integration Wires hosted remote or SDK-managed local MCP servers into an Agents SDK tool loop. MCP and Connectors Lets models call remote MCP servers or OpenAI-maintained OAuth connector wrappers for third-party apps. OpenAI Docs MCP Server Hosts a public read-only MCP server exposing search and content from OpenAI developer documentation for use as agent context. Secure MCP Tunnel Connects a private or on-prem MCP server to OpenAI products without exposing it to the public internet; covers private developer-mode plugin testing only, and can be selected directly as a 'Tunnel' connection method in ChatGPT's plugin developer mode.
Content & Creative Generation Chat Completions Audio I/O Lets Chat Completions accept audio input and return audio output. Image Generation (GPT Image) Generates and edits images from text or image prompts via the Image API or Responses API, including multi-turn editing. Realtime API (Voice Conversations) Provides a low-latency WebSocket API for building live, bidirectional voice conversation and transcription sessions, authenticated via short-lived ephemeral client secrets for browser clients. Realtime Speech Translation API Provides real-time speech-to-speech translation via the Realtime API. Speech-to-Text API Transcribes audio into text via the transcription API, with a documented streaming-event reference for incremental transcription results. Text-to-Speech API Generates spoken audio from text via the audio speech API. Vision / Image Analysis Input Lets models analyze image inputs supplied as a URL, base64 data, or file ID with configurable detail levels.
Voice & Realtime Custom Voices Creates custom voice likenesses from consent and sample recordings for use in text-to-speech, Realtime, and GPT-Live sessions. GPT-Live API Runs live full-duplex voice sessions (gpt-live-1) over WebRTC or WebSocket that delegate task work to a Responses backend model or a client-run agent, with session forking, stored recordings, and sideband server controls. GPT-Live SIP Telephony Accepts inbound phone calls over SIP for GPT-Live, with webhook call notification, accept/reject/refer/hangup endpoints, DTMF events, and SRTP media. Live Session Fork Starts a new Live session from a stored session's conversation state over a dedicated fork WebSocket, inheriting configuration and receiving a new session ID. Live Sideband WebSocket Attaches a backend server to an existing Live session over a sideband WebSocket to control the session and observe events while audio stays on the primary connection.
Data Movement File Inputs Accepts PDFs, Office documents, spreadsheets, and text/code as model input, with type-specific processing. File Storage Quota Caps stored files at 2.5 TB per project and returns a storage-quota error on requests that would exceed it. Large File Uploads API Uploads large files in multiple parts via the Uploads API.
Agent Safety Controls Agent Guardrails Automatically validates agent input, output, or tool-call behavior and can block a run. Misalignment Monitoring Asynchronously reviews an agent's reasoning and actions in supported Responses API requests to catch it misinterpreting instructions in consequential contexts, and can stop a conversation or send a safety.alert.created webhook when it flags a concern.
Content Safety & Provenance Content Provenance API Checks an uploaded image or audio file for supported OpenAI provenance signals (C2PA Content Credentials and SynthID watermarking) and returns verification results synchronously.
OpenAI products and featuresopenai.com