krummholzQuestions worth
working on.
We are conducting pure research into LLM memory and approaches to artificial intelligence beyond language models.
Learning that lasts.
We study how language models can retain and use knowledge over time, with user control at the center.
Other paths to intelligence.
We use Meta’s V-JEPA 2 as a starting point for research into joint-embedding predictive architectures. This work explores non-LLM approaches to artificial intelligence, separate from our current model offerings.
These are active research areas, not released products or promises of a particular result. We are keeping technical details private while the work develops.
Explore our open-weight models →