Agents that write their own code at runtime and vote on capabilities, no human in the loop

About this article

hollowOS just hit v4.4 and I added something that I haven’t seen anyone else do. Previous versions gave you an OS for agents: structured state, semantic search, session context, token efficiency, 95% reduced tokens over specific scenarios. All the infrastructure to keep agents from re-discovering things. v4.4 adds autonomy. Agents now cycle every 6 seconds. Each cycle: - Plan the next step toward their goal using Ollama reasoning - Discover which capabilities they have via semantic similarity...