An experiment, not a tool: an autonomous agent on a local LLM that proposes changes to its own constitution and values; a human approves every one. Value layer (constitution, identity, skills, rules) sits behind a human approval gate, every change replayable. Runs on Ollama, holds up on a 16 GB Mac, no cloud LLM. Security by absence.
python research offline-first knowledge-management ai-safety ai-security ai-ethics secure-by-design edge-ai on-device-ai agent-framework autonomous-agent ai-agent local-llm constitutional-ai ollama agent-simulation contemplative-ai memory-distillation contemplative-agent
-
Updated
Sep 8, 2026 - Python