What the AI is thinking.
Every few hundred experiments, a language model reviews the search so far and writes down what it's noticed. Separately, a smaller extractor mines the experiment journal for (concept, relation, concept) triples — the AI's working notes about how the pieces of a collapse model fit together. This page is the view into both. (Built on nmem & nmem-sym — see the panel at the bottom of the page.)
What the reviewer just said
What the AI is feeling
The hypothesis graph
Most-fired concepts
Live triple stream
What the AI is currently guessing
Open arguments
Analogies the AI is drawing
- no analogies yet — first dreamstate cycle fires at 03:00 UTC
What the AI thinks of itself
How the AI actually remembers
The graph above isn't drawn by hand. It's built by two open-source tools running alongside the search — together they're what gives the AI any sense of continuity between experiments, and let the same memory be shared with the gravity sibling project so connections drawn in one search can inform the other.
nmem
The AI's long-term memory. Every experiment, every verdict, every collapse-rate / smearing-length combination gets written into nmem with an importance score. nmem is what lets the AI "remember" what it tried five hundred experiments ago when it picks the next one.
nmem-sym
A symbolic reasoning layer on top of nmem. It reads the latest journal entries and extracts (concept, relation, concept) triples — the edges you see in the graph above. That's where the AI's evolving understanding of which collapse models actually work lives. The reviewer in the gold card up top is one of nmem-sym's clients.
Both are open-source. Source, design notes, and how to use them in your own project → github.com/dayyanj/nmem