-
Is the brain healthy?
Liveness tells you a service is up. It says nothing about whether what a memory system believes is any good. Here is how Mnemos measures the health of a brain — a handful of cognitive vitals, one verdict — and what running one in production taught us about the hardest part: making sure the check counts the right thing.
-
The tool was guessing
We built a tool to tell an AI agent when it was about to hit its rate-limit cap. A code review found it was answering with a guess — while the real number sat unread in its own database. Here is the bug, and why it is the most dangerous kind.
-
From store to brain
A memory library that only writes and reads is missing the interesting half. What building the other half — consolidation, forgetting, salience, self-correcting recall, and metacognition — taught us, with no language model in the loop.
-
800 to 1
We set out to cut the token bill on our AI-assisted coding. The popular fix — make the model talk in terse shorthand — turned out to touch about 1% of it. Here is what the numbers actually showed, and the tool we built and open-sourced to find out.