A ChatGPT user proposes turning chat history into a zoomable city map

A forum post with a big idea

A ChatGPT user has put a product idea in front of the OpenAI community, as community.openai.com shows. Nikolai Bakke, who posts from Norway, calls it "ChatGPT Knowledge City." It is a proposal from a customer. Nothing in the post says OpenAI is building it, has read it or has replied.

The problem he describes will be familiar to heavy users. He has been using ChatGPT for several long-term projects, and his history has grown to hundreds of conversations. He often remembers discussing something but cannot find where. The sidebar, which is a plain list of past chats, works poorly at that size.

His answer is to swap the list for an optional map. You could zoom out to see everything, or zoom in until you reach a single paragraph inside one conversation.

How the city would be laid out

The post borrows the vocabulary of a city to describe how the pieces would fit together:

• Districts would hold major projects and areas of interest.
• Streets would be individual conversations and related topics.
• Addresses would mark specific messages, documents, decisions and milestones.
• Junctions would show connections between conversations, including branches that lead into new threads.

The idea is that one view could carry you from the big picture of your work down to a single detail, without typing a search term or scrolling a long list.

Why a map instead of a list

The author's argument rests on how memory works. He says you might forget a numerical access code yet still remember where the buttons are and the movements needed to press them. In the same way, you may forget what a conversation was titled but recall that the answer sat in a particular corner of a project.

That is why he insists that places on the map should stay put. If the layout kept rearranging itself, the familiarity you build up by returning to the same spots would be lost. He wants the city to grow gradually without reshuffling what already exists.

He also wants it built jointly. ChatGPT could propose districts, streets and links based on what the conversations contain. The user, though, would keep control of the structure, including the names, the positions and how everything is organised.

Branches and a modest first version

The post notes that ChatGPT already supports conversation branching in certain interfaces. Branching means starting a new thread from a point in an existing chat. Knowledge City would make those relationships visible. If a side topic deserves its own conversation, you could spin it off while keeping the relevant context and a visible link back to the original. He compares it to a railway junction, where one line leads into another without abandoning the first track.

He does not ask for a full virtual city on day one. A first version, he suggests, could offer five things:

• A zoomable visual overview of conversations and projects.
• Organisation that the user defines.
• Visible links between related conversations.
• Direct jumps to specific messages and documents.
• Stable locations that support spatial memory.

Over time, he says, this could grow into a richer personal knowledge environment.

What this is, and what it is not

It is worth being clear about the limits. The post is a concept the author himself describes as not a finished technical proposal. It includes no mockups in the material we have, no account of how a map would be generated from hundreds of chats and no data on how many other users share the problem. The author simply asks whether others struggle with long histories and whether a spatial interface would help.

Even so, the underlying complaint is worth taking seriously. As assistants take on long projects, the value of a conversation depends on whether you can find it again. Search, folders and a list sorted by date are all attempts to solve that, and this post argues that none of them uses the way people actually remember. Whether a map would do better, or just become one more thing to tidy, is an open question that only a real prototype could answer.