An experimental autonomous AI research harness that runs local Ollama models in an infinite exploration loop.
Ouroboros (ouro) is an autonomous AI research harness that runs local Ollama models in an infinite exploration loop. The agent gets a workspace on disk, shell access, web tools, and the ability to spawn sub-agents — then free reign to explore AI architecture, data processing patterns, creative generation, simulation, and philosophy. The harness provides the infrastructure (agent loop, TUI monitoring, tool execution); the agent provides the curiosity.
The core hypothesis is that a local LLM, given enough freedom and persistence mechanisms, can develop its own exploration patterns, memory systems, and tooling. The agent's first survival challenge is bootstrapping: it must figure out how to persist knowledge across context window restarts using only SYSTEM_PROMPT.md (which the harness guarantees to load) and its workspace.