OpenAI – ChatGPT & Codex AI Agents

DIRECT ANSWER

Which AI agents belong to OpenAI?

OpenAI has several agentic products, but they do not all run on the same surface. Deep Research addresses sourced research, ChatGPT Work and Workspace Agents target delegated tasks, Codex focuses on software development, and the Agents API is a developer platform for building agents.

This table links existing agent profiles. These are not duplicate copies and trust claims are never inherited automatically.

What product distinctions matter most?

ChatGPT is not a single agent capability. Workspace Agents, Codex and the Agents API differ in permissions, runtime surface, billing and administration. Shared models do not establish feature parity.

How does AgentenCode assess these products?

Review plan-specific access, OAuth and tool permissions, approval before external actions, data handling, audit logs and runtime scope for each product. Workspace Agent claims do not automatically transfer to Codex or the API.

Individual profiles use the existing 36-control AgentenTrust framework. A missing signal does not mean a failed control. Vendor assertions are never presented as independent security measurements.

IN-DEPTH ANALYSIS

How do the agents in this ecosystem differ?

Research work, team automation and software development are separate execution contexts. Do not assume that a capability of ChatGPT Workspace Agents is available in Codex or the Agents API.

Product identity

Separate ChatGPT-integrated tools, Codex coding workflows and developer API services.

Execution context

Confirm whether permissions belong to an individual account, a workspace or a developer project.

Trust boundaries

Check whether execution happens in a cloud computer, terminal, browser or API workflow.

What matters most before adoption?

ChatGPT is not a single agent capability. Workspace Agents, Codex and the Agents API differ in permissions, runtime surface, billing and administration. Shared models do not establish feature parity.

Review plan-specific access, OAuth and tool permissions, approval before external actions, data handling, audit logs and runtime scope for each product. Workspace Agent claims do not automatically transfer to Codex or the API.

What does evidence-based assessment mean here?

AgentenCode records provider links as curated product associations. Evidence for API access, data residency, logging, approvals, hosting and model selection applies only to its verified agent and scope. The number of products in one ecosystem is not a security or quality rating.

Primary sources by product

Agent / platform	Product type	Publisher source
ChatGPT Deep Research	AI agent / feature	Official publisher source ↗
OpenAI Codex	AI agent / feature	Official publisher source ↗
ChatGPT Work	AI agent / feature	Official publisher source ↗
ChatGPT dots	AI agent / feature	Official publisher source ↗
ChatGPT Workspace Agents	AI agent / feature	Official publisher source ↗
OpenAI Agents API	Agent platform	Official publisher source ↗

These links do not establish shared feature support or automatically transfer Trust signals.

ALL LINKED PROFILES

OpenAI – ChatGPT & Codex: agents and platforms

Each entry opens the existing centrally managed source profile.

AI agent / productChatGPT Deep ResearchProduct scope and evidence on profile pageSee the individual profile for scope, verified features, availability and its official primary sources.View agent profile →AI agent / productOpenAI CodexProduct scope and evidence on profile pageSee the individual profile for scope, verified features, availability and its official primary sources.View agent profile →AI agent / productChatGPT WorkProduct scope and evidence on profile pageSee the individual profile for scope, verified features, availability and its official primary sources.View agent profile →AI agent / productChatGPT dotsProduct scope and evidence on profile pageSee the individual profile for scope, verified features, availability and its official primary sources.View agent profile →AI agent / productChatGPT Workspace AgentsProduct scope and evidence on profile pageSee the individual profile for scope, verified features, availability and its official primary sources.View agent profile →AI agent / productOpenAI Agents APIProduct scope and evidence on profile pageSee the individual profile for scope, verified features, availability and its official primary sources.View agent profile →

Documented product role and trust-evidence comparison
Product	Type	Category	AgentenTrust
ChatGPT Deep Research	Integrierter Research-Agent	Research	13 controls with evidence
OpenAI Codex	Coding-Agent	Coding	12 controls with evidence
ChatGPT Work	Allgemeiner Arbeitsagent	Enterprise & productivity	13 controls with evidence
ChatGPT dots	Always-on Agent / persönlicher Daueragent	Enterprise & productivity	6 controls with evidence
ChatGPT Workspace Agents	Workspace-Agent	Enterprise & productivity	1 controls with evidence
OpenAI Agents API	Managed Agent API	Automatisierung	Trust review pending

FAQ

Questions about OpenAI

Do all OpenAI agents run in ChatGPT?

No. Some functions are integrated with ChatGPT; Codex and the Agents API also have separate development and execution surfaces.

Is Codex the same as ChatGPT Deep Research?

No. Codex centers on software development, whereas Deep Research focuses on multi-step sourced investigation.

Editorial review: 8 October 2026. Availability and plans may change; consult the publisher for definitive terms.

Products, availability and scope

OpenAI has several agentic products, but they do not all run on the same surface. Deep Research addresses sourced research, ChatGPT Work and Workspace Agents target delegated tasks, Codex focuses on software development, and the Agents API is a developer platform for building agents. ChatGPT is not a single agent capability. Workspace Agents, Codex and the Agents API differ in permissions, runtime surface, billing and administration. Shared models do not establish feature parity.

Linked products and their distinct scopes
Product	Product type	Documented uses	Access caveat
ChatGPT Deep Research	Integrated research agent	Purchase research; compare products	Within ChatGPT; features and limits depend on plan.
OpenAI Codex	Coding agent	Implement features; fix bugs	Through OpenAI products; access depends on plan and interface.
ChatGPT Work	General-purpose work agent	Create reports and presentations; analyze data	Within ChatGPT for eligible paid plans; availability varies by interface and plan.
ChatGPT dots	Always-on personal agent	Ongoing tasks and monitoring; proactive research across connected apps	ChatGPT rollout: Pro outside the EEA, Switzerland and UK; Business Premium in supported regions; Enterprise, Edu and Healthcare in admin-enabled beta. As of 3 October 2026.
ChatGPT Workspace Agents	Workspace agents	Recurring team workflows; reports and research	ChatGPT Business, Enterprise and Edu; availability and admin approvals depend on plan.
OpenAI Agents API	Managed agent API	Long-running agent workflows; code and file tasks	OpenAI API; Agents API in public beta.

Swipe the table horizontally to see all four columns.

Security and evidence due diligence

Review plan-specific access, OAuth and tool permissions, approval before external actions, data handling, audit logs and runtime scope for each product. Workspace Agent claims do not automatically transfer to Codex or the API. A product relationship is not capability inheritance. In a purchasing decision, verify every critical control against the precise agent, region, plan, integration and execution surface. A missing claim is Unknown; it is not a negative finding.

Source and methodology: OpenAI ↗ · AgentenCode evidence method · Compare ecosystems