openai / openai/codex

[ChatGPT Projects] Add account-wide inventory, cross-Project search, and read-only assistant access

Open
#44,345 0 comments 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

enhancement Linux
Dominant language
Rust
Stars
125k
Forks
19.4k
PR merge metrics
PR metrics pending

Description

Summary

ChatGPT Projects needs an account-wide, read-only inventory and retrieval layer.

Once a user has accumulated dozens of Projects, there is no practical way to answer basic questions such as:

  • How many Projects do I have?
  • Which Projects are empty, dormant, recently active, or overlapping?
  • How many conversations and uploaded files does each Project contain?
  • Where is the old conversation in which I already discussed a topic?
  • Which Projects depend on unique instructions or uploaded context and should be preserved?

ChatGPT/Codex also cannot currently perform this audit on the user's behalf through a safe account-scoped read-only interface. The only workaround is brittle browser automation that manually opens every Project.

User impact

In the observed account, 29 Projects had accumulated across research, coursework, writing, planning, and temporary questions.

The user wanted a read-only audit before reorganizing anything. Even with explicit permission, the assistant could not reliably:

  • enumerate all Projects without expanding the sidebar UI;
  • count conversations per Project;
  • obtain oldest/latest activity;
  • identify empty or dormant Projects;
  • search conversations across all Projects;
  • inspect metadata about Project instructions and uploaded sources;
  • suggest overlap or archive candidates from complete evidence.

The result is predictable: users cannot find prior work, repeat questions they have already asked, lose the value of accumulated research context, and hesitate to clean up because they cannot see what would be lost.

This is especially painful for long-running research workflows, where Project instructions, uploaded files, and multi-month conversation history may be difficult to reconstruct.

Steps to reproduce

  1. Accumulate many ChatGPT Projects over several months.
  2. Open the Projects sidebar on chatgpt.com.
  3. Try to obtain an account-wide table containing Project name, conversation count, earliest/latest activity, file count, and instruction presence.
  4. Try to search for a past discussion without knowing which Project contains it.
  5. Ask ChatGPT or Codex to perform the inventory with explicit read-only authorization.
  6. Observe that there is no account-level Project inventory/search tool available to the assistant.
  7. The remaining workaround is to use browser control and manually visit every Project. This is slow, sensitive to lazy loading, and dependent on browser-extension and OS support.

In this case, the Chrome side panel could read the currently visible page text, but full browser control failed in the Linux desktop preview environment. The account data itself remained inaccessible through any direct read-only Projects API.

Expected behavior

1. Projects management dashboard

Provide an account-wide table with at least:

  • Project name;
  • conversation count;
  • earliest and latest activity;
  • uploaded-source count and broad file types;
  • whether Project instructions are present;
  • storage/ownership/workspace scope;
  • sorting and filtering by activity, name, size, and emptiness;
  • explicit indicators for empty and dormant Projects.

No destructive operation should occur from this view without separate confirmation.

2. Global cross-Project search

Search all Project conversations and ordinary chats, with every result labeled by its containing Project. Search should work even when the user does not remember the conversation title or Project name.

3. User-consented read-only assistant tools

Expose safe, non-mutating tools such as:

  • list_projects
  • list_project_chats
  • get_project_metadata
  • search_project_chats

Read access should be explicitly authorized and separable from rename, move, archive, and delete permissions. Metadata-only enumeration could be the default, with conversation contents requiring a stronger opt-in.

4. Organization assistance

After a complete inventory, ChatGPT should be able to suggest—but never automatically execute—possible merge, archive, export, or knowledge-base candidates. It should preserve clear boundaries between unrelated people, subjects, and instruction sets.

5. Exportable index

Allow exporting a Project/chat index in CSV, JSON, or Markdown so users can maintain a durable catalog in tools such as Obsidian or Notion.

Actual behavior

Project information is fragmented across individual pages and the sidebar. The assistant normally sees only its current conversation/Project context. There is no supported account-wide read-only inventory capability, and browser automation is an unnecessarily fragile substitute for basic information architecture.

This is not primarily a model-quality problem. The model can analyze the inventory; the product does not expose the inventory safely.

Why this matters

Projects become less useful as their number increases—the opposite of what a long-term workspace should do. The longer a user relies on Projects, the harder prior knowledge becomes to retrieve and the more likely they are to repeat work.

A read-only inventory and global retrieval layer would improve trust, reduce duplicated conversations, make cleanup safer, and make Projects viable for multi-year research and professional use.

Related issues

This request is related to but distinct from:

  • #41819 — lightweight chat groups/collections;
  • #29548 — history search missing text in long Project threads;
  • #37791 — functional fragmentation across ChatGPT, Work, and Codex;
  • #38042 — documentation comparison for different Project/container types.

Those issues cover grouping, a search defect, product fragmentation, or documentation. This request specifically asks for account-wide Project observability, cross-Project retrieval, and consent-based read-only assistant access.

Environment

  • ChatGPT web and desktop
  • Account with 29 ChatGPT Projects
  • Linux desktop app preview with Chrome extension
  • No Project or conversation content was modified during the attempted audit

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No files, tests, or implementation entry points are identified in the issue. First determine whether account-wide Projects data and assistant tools belong in this repository, then narrow the request to a concrete component with an explicit read-only scope and acceptance criteria.

Written by the indexing model from the issue text.

Assessment

Domain
backend-api-design, search
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.