Coming soon · for puzzle players, and developers curious how an agent picks its context
One email when it launches. Nothing else, and never shared.
Questions
How does the archivist choose?
It’s Contextrie, an open-source library for choosing an agent’s context. When the case was made, it read every document and wrote a title, a summary and key points. When you ask, it scores all forty against your question and pulls the top three. Each pulled document shows its score and the reason it was picked.
Can I just ask who did it?
You can. The archivist only pulls documents; nothing writes you an answer, so nobody can tell you who did it. Whether the three it pulls help is up to how you ask.
Does the same question always pull the same documents?
Yes. The first time anyone asks a question, its pull is saved, and everyone asking the same thing gets the same three documents.
How many questions did Contextrie need?
Each case is played once by Contextrie with the same eight plain questions. You see how many it needed at the end. Try to beat it.
What do I get for free?
Today’s case, every day. A one-time $0.99 opens the archive of every past case, forever. No subscription.
Is anything about me stored?
Your questions are saved against your case so you can come back to it, and their pulls are shared with other players. We record anonymous counts only.
Can I build something like this?
Yes. It runs on Contextrie (open-source context and state) and Bref UI (open-source interface). Both are on GitHub.