Feature Request: Decouple Chat/Agent Modes from Cloud/Local Projects — We Need Local Chat

I’ve been using ChatGPT, Work, and Codex extensively for research, learning, and software development.

While trying to build a consistent AI workflow, I noticed what seems to be a fundamental issue with how ChatGPT organizes projects and interaction modes.

I believe ChatGPT should separate two independent dimensions:

1. WHERE — Execution Environment

  • Cloud: Cloud-based projects and resources.
  • Local: Local folders, files, and development environments.

2. HOW — Interaction Mode

  • Chat: Conversational learning, reasoning, exploration, and discussion.
  • Agent: Autonomous task execution, file modification, and multi-step workflows.

Conceptually, these should be independent choices.

1. The Missing Combination: Local + Chat

Ideally, all four combinations should be available:

Environment Chat Agent
Cloud :white_check_mark: Supported :white_check_mark: Supported
Local :cross_mark: Not supported in my current app :white_check_mark: Supported

Currently, local projects primarily support Work and Codex workflows.

However, when I switch to native Chat mode and attempt to select a local project, the desktop application explicitly displays:

“Local projects do not support chat.”

Chat offers a different interaction experience, particularly for iterative learning, conceptual understanding, research discussions, and exploratory reasoning.

I want to preserve that experience while accessing my local files.

2. Why This Matters: A Research Workflow

As a researcher, I often work with a local project containing:

  • Research papers and technical notes.
  • Source code and configuration files.
  • Experimental results and logs.
  • Project documentation and research decisions.

My ideal workflow is straightforward:

Chat → Work → Codex → Chat

  1. Use Chat to read a paper, understand an algorithm, and discuss research ideas using my local project context.
  2. Use Work to organize notes, prepare documentation, or conduct additional research.
  3. Use Codex to implement experiments, modify code, and run tests.
  4. Return to Chat to analyze experimental results and explore the next research direction.

All of this should happen within the same logical project.

I shouldn’t need to maintain separate Cloud and Local projects or repeatedly transfer context just because I want to switch between discussion and execution.

3. Proposed Improvements

Allow ChatGPT’s native conversational mode to access user-authorized local folders.

Let users independently select:

  • Environment: Cloud / Local
  • Interaction: Chat / Agent

Work and Codex can remain specialized Agent experiences.