I’d like ChatGPT Projects to support selecting a user-approved local directory.
The goal is to let a ChatGPT Project use a local working directory as part of the project workspace, so ChatGPT can act as the personalized planning and review layer for local work.
The key value is not only that ChatGPT can see the chats inside a project. ChatGPT can also use personalized memory and long-term context from previous interactions to understand the user’s preferences, constraints, working style, recurring decisions, and what kinds of outputs are actually useful for that user.
Many important requirements are created through conversation before anything is formally specified: goals, preferences, constraints, rejected ideas, UX decisions, and the user’s actual working style. In addition, ChatGPT often already knows user-specific context that is not contained in a single project folder or chat thread.
Desired workflow:
In a ChatGPT Project, I specify a local directory.
While brainstorming or planning in the project, ChatGPT can read and write files in that directory, especially docs/spec.md, docs/plan.md, and review notes.
I can open the same local directory in Codex or an IDE.
Codex can implement based on the files ChatGPT created.
ChatGPT can later review Codex plans, diffs, or notes using the same project context, local files, and personalized understanding of the user.
This would make ChatGPT the main agent that understands the user, while Codex or another coding agent acts as the implementation agent.
Minimum useful version:
Allow each ChatGPT Project to select a user-approved local directory.
Let ChatGPT read and write files within that directory, with clear user permission.
Make it easy to write specs, plans, and review notes under docs/.
Let Codex or an IDE open the same directory separately.
Let ChatGPT review files, plans, or diffs from that directory using Project context and personalized memory.
This would be useful for individual creators, product managers, designers, writers, and small developers who use ChatGPT for early-stage brainstorming before implementation.
In short:
Let ChatGPT Projects become the personalized brain for a local workspace.
A similar workflow already exists in some AI coding tools, where chat-based planning can work with local project files and the coding agent can continue from those files.
My request is for ChatGPT Projects to support this kind of local-directory workflow, with ChatGPT’s personalized memory and long-term user context as the main advantage.
In other words, the important part is not just local file access. It is connecting a local workspace to a personalized main agent that already understands the user’s preferences, constraints, working style, and past decisions.
Hey @NeMab4! I get what you’re going for here, and it feels really practical. Having ChatGPT help shape specs, plans, and review notes in a local workspace, while Codex or an IDE picks things up from there, would make that collaboration loop feel a lot tighter.
Appreciate the detail here, especially the extra context. I can’t share a timeline right now, but I’ll pass it along internally.
this is how ai coding AI has evolved. Question is how advanced of a harness ChatGPT app should be, given broad mainstream use. The simple decision to expand its sandbox to reading and writing files opens any entirely new dimension for errors, a new attack surface, and changes the perception of a safe app.
OpenAI_Support, not sure this solves that, but what about a limited sandboxed directory when only .md files can be accessed. Doesn’t do everything asked but gets part way without creating our own harness in Python etc?
