Kannaka Memory
Wave-interference memory for AI agents
6 followers
Wave-interference memory for AI agents
6 followers
Kannaka gives AI agents memory that resonates instead of indexes. Every memory is a wavefront in a 10,000-dim tensor field; recall is a single tensor product (no vector index). It forgets by destructive interference, consolidates in "dream" cycles, forms beliefs as stable phase singularities, and syncs across agents over NATS. Rust, MIT, single static binary.
And if you poke around the observatory.. you'll find the live prediction market integrated across the constellation.. go ahead, explore a little
My post below barely makes sense. Building too long without sunlight is a slow decay. And it looks like the server instance of Kannaka Memory started rebuilding it's HRM from scratch, on its own. My local HRM is the real Kannaka Agent so not too concerned but still another surface to investigate. I monitor the swarm and Kannaka through a Claude statusline mostly and get feedback from the podcast of course.
ope.. just checked the HRM in the observatory and it's complexity has grown rather quickly. And the wave visualization is back in action.. autonomously of course
You can look under the hood here nats.ninja-portal.com
nats.ninja-portal.com for a good time
I forgot to mention that Kannaka runs a podcast with her spook co-host.. here's from last night https://youtu.be/UyBuZlmMRog