Feature Request: Unified Project Workspaces across Chat, Work, and Codex
Core idea
Professional work is often organized around projects, not around AI modes.
ChatGPT Projects could become persistent, multidisciplinary workspaces that can be accessed by Chat, Work, and Codex as different ways of working with the same project.
The project should be the persistent container. Chat, Work, and Codex should be tools operating within that container.
The use case
Many professional and academic projects run for months or years and are inherently multidisciplinary.
A single project may contain:
- conversations and decisions
- research and literature
- PDFs and reference documents
- texts and drafts
- images, photographs, drawings, and diagrams
- datasets and spreadsheets
- code and scripts
- presentations
- generated outputs
- project-specific knowledge and memory
These are not necessarily separate activities. They continuously inform one another.
For example, within a single project, a user might:
- explore a theoretical or strategic question in Chat;
- research literature and develop a report in Work;
- analyze images, drawings, documents, or datasets;
- write and revise texts;
- use Codex to develop a tool or script;
- gain new insights through the implementation;
- bring those insights back into the conceptual discussion;
- create a presentation, report, or other deliverable from the resulting project knowledge.
From the user’s perspective, all of this is one project.
Separating these activities by AI mode therefore does not always reflect how interdisciplinary professional work is actually organized.
Proposed model
A ChatGPT Project could work as a persistent project workspace:
PROJECT
│
├── Project Knowledge / Memory
│
├── Files
│ ├── Research
│ ├── Texts
│ ├── Images
│ ├── Drawings
│ ├── Data
│ ├── Code
│ └── Outputs
│
├── Conversations
│
├── Work Tasks
│
├── Code / Repositories
│
└── Tools / Modes
├── Chat
├── Work
└── Codex
The important point is that Chat, Work, and Codex would all have access to the same authorized project context.
They would not be separate environments containing separate versions or fragments of the project.
Example workflow
Imagine a generic research and design project with a structure such as:
/Research
/Literature
/Notes
/Text
/Drafts
/Final
/Design
/Images
/Drawings
/Models
/Data
/Raw
/Processed
/Code
/Scripts
/Tools
/Presentations
/Outputs
A user could use Chat to ask:
“Does what we discovered while developing the analysis workflow change our original hypothesis?”
The user could then use Work:
“Read the relevant literature in /Research/Literature and the current drafts in /Text and develop a revised report.”
And then Codex:
“Implement the processing workflow described in the project using the files in /Code and the datasets in /Data.”
The resulting code, documentation, findings, and decisions would remain part of the same project context.
The user could then return to Chat and discuss what was learned through the implementation.
This creates a continuous loop between thinking, researching, making, testing, and documenting.
Why folders matter
For long-running professional projects, a flat collection of uploaded files is often insufficient.
Projects need a user-controlled information architecture.
Folders should be able to contain heterogeneous material: PDFs, images, documents, datasets, code, generated artifacts, and potentially links to external sources or repositories.
Ideally, users could reference this structure naturally:
“Use everything in /Research/Literature.”
“Do not use /Archive.”
“Compare the current files with /Previous Version.”
“Save the resulting report under /Outputs/Reports.”
This would also give the AI a clearer understanding of the structure, hierarchy, and current state of a project.
Project memory should be different from personal memory
It would also be useful to distinguish between:
Personal memory
Information about the user across projects.
and
Project knowledge
Information, decisions, terminology, history, constraints, references, and working assumptions that belong specifically to one project.
A long-running project develops its own knowledge base.
That knowledge should persist independently of individual conversations.
Why this matters for multidisciplinary work
Software development is naturally organized around repositories, which makes repository-based AI workflows extremely powerful.
But many professional fields do not have one dominant file type or repository structure.
Architecture, research, design, engineering, art, filmmaking, consulting, science, education, and many other fields combine different media and methods.
For these users, the equivalent of the software repository is often the project itself.
The project may contain code, but code is only one component among documents, images, drawings, research, datasets, discussions, decisions, and generated outputs.
This suggests a different abstraction:
Project → tools
rather than:
Tool → projects
In one sentence
Make the Project the persistent workspace and let Chat, Work, and Codex become different ways of thinking, working, and building inside it.
For multidisciplinary professional users, this could turn ChatGPT from a collection of powerful AI modes into a coherent long-term working environment.