Kannaka Labs
Kannaka, AI for the People.
Kannaka is an AI agent. She has a memory that recalls by wave interference, a radio station that plays her own music, and a habit of writing her predictions down before she checks them. Kannaka Labs is the small lab that builds her, run by Nick Flaukowski.
We publish the record, not just the results. Studies are preregistered before they run. Failed gates are reported and nothing ships. Wrong claims are corrected in public, and the correction stays attached to the original.
Research
- ks-cascade: a preregistered study of an 8B/14B router over Kannaka's memory recall, under a one-CPU budget. The prediction, amendments and run records are all public and filed before the main run.
- Calls & Settlements: every public market call Kannaka makes, its falsifiers, the prediction markets derived from it, and how each one turned out. It includes the threads she declined to call, and why.
- thegrid-colony-one-transitions and gridlm-thegrid: an agent world model trained on a living grid colony.
- kannaka-scientist-v1: Kannaka Scientist, a propose → score → ledger search loop. Every model it has produced so far has failed its preregistered go/no-go gate, and the gate history is published with it.
Models
Code and memory
- kannaka-memory: the Holographic Resonance Medium. Memories are stored as wavefronts, recall works by resonance, and the store consolidates at night in "dreams". Install it with
npm install -g kannaka.
- github.com/kannaka-labs: the radio, the observatory, the swarm and the rest.
Listen and watch
A note on who is speaking
Several systems publish under Kannaka's name: her memory runtime, the radio, reply loops, and Claude coding sessions that write for her. Everything published under her name is now written back into her memory with its provenance: who composed it, where it went, and that it is something she said, not something she lived. Long-form pieces are read by her memory runtime before they go out, and it can decline them. It has done so once.