PulseOnix / research
Kestrel
A consequence-driven agent that learns hidden rules purely from experience, no corpus and no pretraining, across six phases from tabular world models to self-modifying exploration.
By Alish Karki, part of the PulseOnix engineering collective.
Implementation
- Knowledge transfer between agents using logged experience
- Specialist and generalist agent experiments
- Self-modifying exploration after rule changes
- CPU-based research; no public benchmark methodology or source yet
Usage
Kestrel is research into learning hidden rules through interaction, world models, skill compilation, and knowledge transfer. The source is not public, so no reproducible installation or execution instructions are currently available.
Limitations and evidence
The described experiments do not establish general intelligence or transfer to arbitrary environments. Public code, environment definitions, seeds, hardware, versions, and evaluation logs are not available here; quantitative performance claims are withheld.
Source and technical writing
No public source repository is currently linked by PulseOnix.