Memory that survives and stays useful
Memory should persist across sessions, stay structured, and help with understanding—not just store fragments.
We’re exploring the intersection of memory, verification, and agent learning.
We’re focused on making those three things practical in real agent work.
What we’re focused on
Verification that creates proof
The system should generate and use evidence for work done, so important claims can be checked.
Agent learning that compounds
Learnings should be reusable and shared across runs so agents improve over time instead of restarting from scratch.
Why this matters
Today, agents often lose context, make unsupported claims, and repeat the same mistakes.
Better memory, proof-based verification, and compounding learning are how we make them more useful over time.