I built a neat tool that turns years of my LLM chat history into one 3D context map!

I built a neat tool that turns years of my LLM chat history into one 3D context map! 图片 1

I built a pretty neat tool/app thing that turns years of LLM chat history into one 3D context map. I've open-sourced the code now so anyone can run it locally on their computer. I call it Constellate though (for obvious reasons, and I thought it sounded cool) Basically, it pulls together every conversation I've had with any AI chatbot, across every provider, and turns the whole thing into one 3D map I can move through. I drop my ChatGPT, Claude, Gemini, or whatever else export into it, and it parses the messages, figures out which conversations are about the same things, and places each one as a point in space. There are two ways to look at it. One clusters related chats together so proximity tells me what's similar. The other assigns coordinates, time along one axis, and the two strongest topic contrasts pulled out of my own history along the other two, labeled with the terms driving them, so I can watch how what I talk about has shifted over time. The whole thing is built to stay current without much upkeep. If I point it at a folder, any export drop there later gets picked up automatically, and a scheduled task refreshes my Claude history every morning so the map never goes stale. I can also export the whole thing as an Obsidian vault, with links between related conversations already drawn in, or as a JSON snapshot I can hand off or reload somewhere else later. I am disappointed to find that my years of ChatGPT history have been lost (not sure if I swapped accounts and deleted an old one at one point). I was originally an OpenAI user before AI checker came around. But I do have around a year and a half of decent history (>2,000 chats & >18,000 messages) across all the providers I use. Oh also! It has Apple intelligence integrated so I can get decent summaries of various context clusters and other cool stuff. And , I can export specific related context queries as a context pack if I ever want to upload context into a fresh LLM session across any provider. ---------- If you want to check it out, the repository is up on GitHub at github.com/JCarterJohnson/constellate . To get the basic unbundled command-line version running, you just need to drop into the directory and run cd ConstellateKit && swift test to run the core verification tests, then head back to the root folder and run swift build -c release to compile the release binary. To get the exact native application setup like I have on my Mac (w/ the floating search panels, drag-and-drop import overlays, and system Spotlight deep-linking) you just need to open the canonical project file in Xcode 15 or newer to run a sandboxed app build, and make sure you have Apple Intelligence turned on in your macOS 26 system settings so the offline FoundationModels layer can handle the summaries, gap explanations, and grounded answer features.

添加评论
点赞收藏
点踩分享查看原文
评论
?
参与讨论