mental ledge
( i put this on the dev community and the bot says i was lost, oops! )
**TL;DR: Projects need shared context across chats (optionally). Starting a new chat within the same project should not require re-explaining core canon or guessing what the model remembers. Optional cross-chat awareness, or optional access to project-level files, would turn Projects from organizational folders into true working spaces.
Potentially this is resolved via implementation of an explicit, project-level canonical memory, distinct from global or implicit memory systems.**
I use Projects primarily for long-form creative writing. Within a project, I define characters, canon rules, worldbuilding, and narrative direction, while ChatGPT helps elevate prose, dialogue, pacing, and cinematic structure.
Over time, individual chats become extremely dense. Performance degrades, responses slow, and eventually details like chapter titles and established references begin to fall out of active context. At that point, starting a fresh chat within the same project becomes necessary to maintain usability and momentum.
The core issue is continuity.
Currently, moving to a new chat requires heavy manual reseeding of canon: character rosters, relationships, organizational structures, timelines, and hard rules like “no fabrication” or “earliest-established details are canon.” While ChatGPT sometimes states that certain details do not need to be re-provided, this is not reliably true upon verification. This creates a trust gap and breaks creative flow.
Projects currently feel like folders, not shared contexts.
There is no reliable way for a new chat to reference prior chats in the same project, distinguish canon from exploratory discussion, or confirm whether specific information is actually available versus assumed.
Even uploaded project files are not accessible across chats. For example, when attempting to reference a project file (e.g., Chapters-1-3-Summary.txt), the model responds that it does not have ambient access to files unless they are re-uploaded or pasted into the current chat.
Requested improvements (optional and user-controlled):
A project-level “canon memory” or pinned reference (Project instructions is limited)
Optional ability for a chat to reference other chats within the same project
Optional read-access to project-level files
Clear, explicit signaling of whether the model does or does not have access to specific information.
These changes would significantly reduce manual repetition, improve trust, and support long-running creative and technical projects without forcing unwanted memory persistence. - In other words, I don’t want it to remember everything by default, I would like to choose what carries forward. - like a **project-level Memory.
In creative terms: drafts stay drafts, exploratory chats stay messy, and when canon is declared (“lock it in”, “add to project memory”, etc), it gets put on the memory of the project to be referenced going forward.**