[Docs][Projects] Add a side-by-side comparison of ChatGPT Web/Cloud Projects, Local Projects, and Codex projects
Summary
OpenAI’s current documentation explains individual parts of the new project system, but it does not provide one authoritative side-by-side comparison of the different project models that now coexist across:
- ChatGPT Web
- the new ChatGPT desktop app
- ChatGPT Work
- Codex Desktop
- Codex CLI
- the Codex IDE extension
This gap became especially visible after the former Codex app became the new ChatGPT desktop app. Existing Codex projects were retained, account-backed ChatGPT Projects were added to the desktop Projects view, Local Projects remained folder-based, and Local/Worktree/Cloud also continued to describe Codex execution environments.
The result is that a user cannot determine from one official page:
What project object am I creating, where is it stored, what does it sync, which product surfaces can use it, and where will a particular chat or task run?
This is separate from my existing issues
This documentation request is intentionally separate from:
- #37573 — ChatGPT official-source retrieval overlooking
learn.chatgpt.com/docs - #37581 — inconsistent ChatGPT/Work/Codex terminology across the app, GitHub, labels, templates, docs, and support
- #35077 — the missing migration/bridge between existing local Codex projects and newer ChatGPT/Web Projects
- #33942 — Chat and Work not sharing task visibility or working context inside the same ChatGPT Project
- #37865 — a specific Local Project identity/discoverability failure and the missing lifecycle documentation exposed by it
This issue does not ask OpenAI to rename the products, fix documentation retrieval, merge every project type, or implement a migration feature.
It asks for one canonical comparison matrix documenting the project and runtime architecture that already exists.
What the current official documentation establishes
The current Projects documentation says that the desktop Projects view includes both ChatGPT projects and local projects. It also explains that:
- the same ChatGPT Project can contain Chat and ChatGPT Work chats;
- a ChatGPT Project carries uploaded files, connected sources, and project instructions but does not directly expose a folder on the computer;
- a Local Project gives chats access to one or more folders on the computer;
- Codex CLI treats its current directory as the project for a chat;
- the Codex IDE extension treats the open folder or workspace as the local project;
- CLI and IDE do not expose the ChatGPT Projects view.
The Codex environments documentation separately defines Local, Worktree, and Cloud as choices for where a Codex chat runs.
The desktop migration documentation says that existing Codex chats and projects remain after the former Codex app updates into the new ChatGPT desktop app, that existing ChatGPT Projects appear in the desktop Projects view, and that Codex retains separate workflows and history.
The Work/Codex documentation further distinguishes Cloud Work, local Work, and Codex Local, including different synchronization and permission behavior.
These facts are useful, but they are scattered across multiple pages. No page maps all project objects and execution modes side by side.
The missing distinction
The documentation needs to separate two independent questions:
- What persistent project or context container does this chat belong to?
- Where does this particular chat or task execute?
Today, the same words—especially Project, Local, and Cloud—can appear at both layers.
Requested Matrix A — persistent project/container types
Please publish one table comparing at least these entries:
| Project/container encountered by the user | Required clarification |
| --- | --- |
| ChatGPT Project created on chatgpt.com | Is this the canonical account-backed ChatGPT Project? |
| ChatGPT Project created or opened in the desktop app | Is it exactly the same persistent object as the web-created Project? |
| Desktop option presented as Cloud or Cloud Project, where shown | Is this a distinct project type, another label for a ChatGPT Project, or only an execution choice? |
| Local Project connected to one or more folders | What creates and identifies the project record? |
| Existing project carried forward from the former Codex app | Is it exactly the current Local Project model, a migrated legacy record, or a separate Codex project type? |
| Codex CLI current-directory project | Is this a persistent project object or only working-directory context? |
| Codex IDE folder/workspace project | Is it shared with Desktop Local Projects or independently inferred from the IDE workspace? |
| Projectless Chat, Work, or Codex conversation | What context and persistence remain without a project? |
For each row, document:
- canonical official name;
- exact UI labels on web, desktop, mobile, CLI, and IDE;
- whether it is a persistent project record, folder registration, directory context, view, mirror, or temporary context;
- where it can be created;
- whether it has an account-backed ID, workspace ID, local UUID, folder-derived identity, or no project ID;
- whether its scope is account, workspace, computer, operating-system user, folder, repository, or chat;
- where project metadata and chat assignments are stored;
- where chat history is stored;
- whether it appears on web, mobile, desktop, Remote, CLI, or IDE;
- what syncs across devices and accounts;
- whether Chat, Work, and Codex can each create or resume chats in it;
- whether it supports uploaded sources, connected sources, direct local folders, and repositories;
- which instructions apply: Project instructions,
AGENTS.md, memories, skills, plugins, or other configuration; - sharing behavior;
- behavior when switching ChatGPT accounts on the same computer;
- behavior on a second computer;
- the exact meaning of Remove, Archive, and Delete;
- whether it can be linked, converted, upgraded, migrated, or handed off to another project type.
Requested Matrix B — chat/task execution environments
Please publish a separate table comparing at least:
| Execution choice |
| --- |
| Chat on the web |
| Chat in the desktop app |
| ChatGPT Work — Cloud |
| ChatGPT Work — Local |
| Codex — Local |
| Codex — Worktree |
| Codex — Cloud |
For each execution choice, document:
- where compute runs;
- which project/container types it can use;
- whether it can directly read local files;
- whether it can modify local files;
- whether it can use uploaded Project Sources and connected sources;
- which instructions and configuration it inherits;
- whether Git operations are supported;
- whether a worktree is created;
- where generated outputs are stored;
- where its conversation history appears;
- whether the task continues after the app closes;
- whether the computer must remain awake;
- whether it can be continued from web or mobile;
- whether Remote is an execution environment or only an access/control surface;
- which plan entitlement, allowance, or credit pool applies.
Questions the canonical page should answer directly
- Is a ChatGPT Project created in a browser exactly the same persistent object as a ChatGPT Project created in the desktop app?
- When the desktop UI presents
LocalandCloud, is that choosing project identity, storage, execution location, or a combination? - Is a desktop
Cloud Projectsimply the desktop representation of an account-backed ChatGPT/Web Project? - Is a current Local Project the same project object as an existing project carried forward from the former Codex app?
- Is Local Project identity based on a project record, internal UUID, Primary folder, all attached folders, or a combination?
- May multiple Local Project records reference the same Primary folder?
- When a ChatGPT Project is used for a local Work chat, does the desktop app preserve one account-backed Project, create a separate Local Project, create a managed mirror, or only copy context into a local runtime?
- Can Chat, Work, and Codex all belong to one persistent project object, or does Codex remain in a separate project/history model?
- What exactly transfers through
Continue in Chat,Continue as task, local/cloud handoff, and project migration? - Which of these are shared across surfaces: transcripts, titles, project instructions, uploaded sources, connected sources, local folders, generated files, skills, plugins, memories,
AGENTS.md, project identity, and permissions? - What happens to each project type when the user switches accounts, uses another computer, removes a folder, removes a project, uninstalls the app, or deletes the project?
- Is a Codex CLI or IDE “project” part of the desktop project registry, or does each surface independently treat a directory/workspace as project context?
- Which differences are intentional architecture and which are temporary migration limitations following the Codex-to-ChatGPT desktop transition?
Steps to reproduce the documentation gap
- Create a ChatGPT Project on
chatgpt.com. - Open the same account in the new ChatGPT desktop app.
- Inspect the desktop Projects view and the project-creation options, including Local and Cloud where presented.
- Compare an account-backed ChatGPT Project with an existing folder-based project retained from the former Codex app.
- Open Chat, Work, and Codex and inspect their available project and execution choices.
- Read the official Projects, Codex environments, glossary, desktop migration, and Work/Codex documentation.
- Attempt to determine from one authoritative source:
- whether the web and desktop Cloud project are the same persistent object;
- whether a retained Codex project and a current Local Project are the same object;
- which histories, instructions, sources, and files are shared;
- what syncs across devices;
- and whether
LocalorCloudchanges project identity or only task execution.
- Observe that the answer must be reconstructed from several pages and remains incomplete for project identity, lifecycle, and handoff behavior.
Expected result
A user should be able to open one official page and determine:
What persistent project object am I using, where is its identity and content stored, which product surfaces can access it, and where will this particular chat or task execute?
Actual result
Users must combine several Help and Learn pages, infer relationships among overlapping project and runtime concepts, and run controlled experiments with their own accounts, computers, folders, and projects.
Suggested acceptance criteria
- One canonical side-by-side comparison page is published.
- Persistent project/container types are documented separately from execution environments.
- Browser-created and desktop-created ChatGPT Projects are expressly identified as either the same or different persistent objects.
- The desktop
Cloud/Cloud Projectoption is mapped to its actual persistent project and execution semantics. - The relationship between current Local Projects and projects retained from the former Codex app is documented.
- Local Project identity, folder relationships, duplicate-folder behavior, removal, and deletion are documented.
- Chat, Work, and Codex compatibility is shown for every project type.
- Local, Worktree, and Cloud are clearly identified as project types, execution modes, or both.
- Synchronization, account scope, machine scope, history location, and cross-device continuation are shown for every supported combination.
- Migration, linking, conversion, mirroring, and handoff limitations are stated explicitly.
- The page is linked from both the Help Center and ChatGPT Learn documentation.
Duplicate search
I searched the public openai/codex tracker for side-by-side comparisons, comparison matrices, ChatGPT/Web Projects, Cloud Projects, Local Projects, and Codex projects. I did not find an exact issue limited to publishing this canonical project/container/runtime comparison.
Closest reports address terminology, product integration, migration, or particular project defects rather than the missing authoritative comparison itself.
Official documentation reviewed
- https://learn.chatgpt.com/docs/projects
- https://learn.chatgpt.com/docs/environments/modes
- https://learn.chatgpt.com/docs/glossary
- https://help.openai.com/en/articles/20001276-moving-to-the-new-chatgpt-desktop-app
- https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex
Feedback / diagnostics
I submitted the same documentation request through the in-product feedback flow before filing this issue.
Feedback ID: 019ff1d2-bfe5-7840-8538-797bae85a3c9
3 Comments
Support independently reached the same conclusion about the documentation gap
After this issue was filed, Support responded in a support conversation by constructing two ad hoc matrices from the currently published Help Center and Learn documentation:
That structure matches the public documentation deliverable requested in this issue. However, Support’s own synthesis identified the central architecture and lifecycle questions as still unspecified in the published documentation:
chatgpt.comand a desktopCloud Projectare the same persistent object;Cloud Project/Cloudidentifies a project type, an execution choice, or both;The synthesis necessarily used qualified inferences such as:
Support later stated that it could not provide a support-case or reference number, could not confirm whether an internal ticket had been created or routed from the conversation, could not guarantee that the conversation would remain open, and could not provide a retention timeline.
This does not establish that no internal escalation exists, and this issue is not requesting proof of internal routing. It establishes that the private support synthesis is not a durable, independently referenceable, publicly maintained documentation artifact.
Requested public deliverable
Publish one canonical documentation page containing:
The page should explicitly define:
It should be maintained as the canonical public source and linked from both the Help Center and ChatGPT Learn documentation.
Related in-product Feedback ID:
019ff1d2-bfe5-7840-8538-797bae85a3c9Mobile Remote can start a remote chat without attaching it to a project, and that chat remains manageable in the Mac app. Desktop has no equivalent: projectless mode removes the Remote target. This looks half-supported already. Feedback:
01a010c5-b699-7701-8dde-88a7da0dc808.@exevil Supported and Documented are 2 things OpenAI is doing horrible right now since the ChatGPT app release, and the confusion in terminology, not even their own chats/work tasks with web search on understand. lol