{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreieufyig2ubsspdluu46tpcd3wl7grn626inrhpkcbxkvbiu4k233a",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mpemgruzfc42"
  },
  "path": "/t/feature-request-make-project-memory-transparent-searchable-and-user-controlled/1385159#post_1",
  "publishedAt": "2026-06-28T18:25:51.000Z",
  "site": "https://community.openai.com",
  "textContent": "**Feature Request: Make Project Memory Transparent, Searchable, and User-Controlled**\n\nI use ChatGPT Projects extensively for long-term work. Over time, I have realized that Projects are becoming much more than simple folders for organizing chats. For many users, they are becoming long-term collaborative workspaces.\n\nA single project may contain many conversations over weeks or months, covering troubleshooting, research, decisions, failed attempts, conclusions, documents, and evolving plans .\n\nThis creates a new need: not only project-scoped retrieval, but **transparent and user-controllable project continuit** **y**.\n\n**Current is sue**\n\nWhen working inside a Project, users often do not know whether ChatGPT is actually using previous conversations from that Proj ect.\n\nSometimes, important context already exists in older project conversations, but the user still has to manually summarize or paste it a gain.\n\nThis creates uncertainty:\n\n  * Did ChatGPT search the current Project?\n  * Did it only use the current conversation?\n  * Did it use uploaded files but not past conversations?\n  * Did it miss an important previous decision?\n  * Is it relying on global memory instead of project-specific knowledge?\n\n\n\nFor long-term projects, this uncertainty becomes a real limitation.\n\n**The m issing piece**\n\nProject knowledge is not only stored in upload ed documents.\n\nIt is also stored in the conversatio ns themselves.\n\nFor example, a long-term proje ct may contain:\n\n  * decisions that were made months ago,\n  * assumptions that were tes ted and rejected,\n  * troubleshooting steps that worked,\n  * troubleshooting steps that failed,\n  * technical configurations,\n  * writing direction,\n  * research conclusions,\n  * user preferences specifi c to that project,\n  * explanations built over many conversations.\n\n\n\nThis is not just “chat history.”\n\nIt is accumulat ed project knowledge.\n\n**Proposed improvements**\n\nI would like to suggest making Project Memory and Project Retrieval more vi s ible and controllabl e.\n\nPossible features:\n\n**1. “Search this Project before answering”**\n\nAdd a button or command that lets the user explicitly ask ChatGPT to search the current P roject bef ore responding.\n\nExample:\n\nSearch th is Project before answering.\n\nThis would be very useful when the user knows the answer may already exist somewhere in pr e vious project conversat ions.\n\n**2. Show what was used**\n\nAfter answering, ChatGPT could indicate what kind of c ontext was used.\n\nFor example:\n\nUsed context from: current conversation, 3 previous project conversations, 2 uploaded files.\n\nThis would build trust and help users understand whether the answer was properly grounded in the Project.\n\n**3. Let users inspect retrieved context**\n\nUsers could optionally expand a small section showing which previous convers ations or documents were referenced.\n\nThis would make the system more transparent witho u t cluttering the default experie nce.\n\n**4. Add project retrieval modes**\n\nEac h Project could have a setting such as:\n\n  * Current conversation only\n  * Current conversation + uploaded files\n  * Current conversation + previous project conversations\n  * Full Project context when relevant\n  * Ask each time\n\n\n\nThis would give users more control depending on the se nsitivity and complexity of the project.\n\n**5. Dist inguish global memory from project memory**\n\nGlobal mem ory should store general user preferences.\n\nProject memo ry should store project-specific knowledge.\n\nThese two layers should remain clearly separated, so unr e lated projects do not influence each other.\n\n**Why this matters**\n\nFor s i mple conversations, this may not matter muc h.\n\nBut for long-term work, it matters a lot.\n\nExamples include:\n\n  * software development,\n  * technical troubleshooting,\n  * research,\n  * engineering,\n  * legal or administrative work,\n  * writing projects,\n  * education,\n  * investment research,\n  * creative worldbuilding,\n  * personal knowledge management.\n\n\n\nIn these case s , users are not just asking isolated que s tions.\n\nThey are b uilding knowledge over time.\n\n**The broader vision**\n\nProjects should not only organize conversations.\n\nThey should help preserve and reuse the reasoning, decisi ons, and knowledge built inside those conversations.\n\nIn other words, Projects could evolve f rom folders into persistent collaborative workspaces.\n\nProject Memory should not feel mysterious or passive.\n\nIt should be:\n\n  * visible,\n  * searchable,\n  * controllable,\n  * project-specific,\n  * and trustworthy.\n\n\n\nI believe this would make ChatGPT much more powerful for users w ho rely on it as a long-term research and work partner.\n\nAs more people use ChatGPT for months or years instead of isolated conversations, Projects are evolving into long-term collaborative workspaces. Making Project Knowledge transparent and controllable would help bu il d the trust required for that kind of colla b ora tion.\n\nThank you for considering this suggestion.",
  "title": "Feature Request: Make Project Memory Transparent, Searchable, and User-Controlled"
}