LoopThing is an exportable file format combining AI-generated outputs with the AI chats that created them, showing exactly how the idea arrived and was refined.
In this meta demo: LoopThing itself was created from 67 chat messages across 3 AI tools. loopthing.loopthing contains 3 primitives and 5 artifact families.
Have you ever:
- Gotten an AI-generated doc from your boss and wished you knew what their prompts were?
- Seen a super-impressive prototype on AI Twitter and wondered how they made it?
- Written "summarize this chat" to share your AI project with a friend, but the response just wasn't there?
LoopThing is for that. You LoopThing your messages across multiple chat sessions to generate artifacts that demonstrate your thinking: prompts, metadata, maps, discarded ideas, explainers, media, slides, prototypes, and code.
The final output is not the only artifact. The thinking can ship too.
A .loopthing is not JSON, not a folder, and not a transcript. It is a single portable container file with a MIME marker: application/vnd.loopthing+zip.
The thing you share is named for the project: agent-docs.loopthing, openclaw-origin.loopthing, yourproject.loopthing. This repo's example is loopthing.loopthing.
The point is simple: make the work understandable to someone who was not in the chat.
This demo container has three core primitives:
Metadata/: message counts, topic tags, and source-shape statsPrompts/: initial prompt, follow-ups, and one-line prompt changesThinking/: journey map, discarded ideas, and process narrative
And five artifact families:
Generated Explainers/: problem, solution, how it works, relevant context, share briefArtifacts/Media/: generated visual explainersArtifacts/Screenshots/: meta screenshots of the viewer explaining the repoArtifacts/Slides/andArtifacts/Docs/: presentation and reference materialArtifacts/Prototypes/andArtifacts/Code/: runnable examples and source code
Open index.html in a browser. The top bar is prefilled with the example container name; press Enter to browse the unpacked contents of loopthing.loopthing.
No build step. No backend. No framework. Just the viewer and the container.
These are meta screenshots: the LoopThing viewer explains the thinking behind this repo, while the README stays focused on the concept.
The artifact stage keeps the thought object visible while a selected primitive or artifact opens in the lens.
Chat metadata becomes a primitive: message counts, topic tags, and source-shape stats.
The journey map is a whiteboard-style lineage view with boxes and arrows, inspired by Excalidraw's hand-drawn diagram style.
Previous screenshot sets are kept for comparison in docs/screenshots/archive/.
The final .loopthing file is the entire exploration: portable, forkable, auditable.
Future versions could:
- Generate
.loopthingfiles from one chat, many chats, or a project workspace. - Turn chat history into metadata, prompt lineage, journey maps, and generated explainers.
- Preserve discarded ideas and the reason each branch died.
- Package media, slides, prototypes, and code so the artifact feels closer to a rich PDF for AI work.


