Autonome is Ric Richardson's architecture for systems that accumulate judgment, not just knowledge.
It extends the LLM-wiki pattern: a maintained source of truth captures what is known, while a governed decision queue captures what should be done next. Over time, the system learns preferences, priorities, risk boundaries, and operating judgment from human decisions.
Most AI systems answer questions. An Autonome is designed to run work.
The long-term goal is progressive autonomy: the system first observes, then recommends, then drafts, then executes bounded low-risk work as it earns trust. Humans remain sovereign governors over strategy, ethics, risk, money, law, credentials, and external commitments.
- Source of truth — structured files or records that humans and AI agents can both read and maintain.
- Continuous reconciliation — new inputs update the source of truth, strengthen existing claims, and flag contradictions.
- Decision queue — unresolved judgments are surfaced as decisions, not just tasks.
- Authority ladder — autonomy increases only after demonstrated reliability.
- Governance boundary — high-risk or irreversible actions remain human-approved.
- MCP server for agent access to an Autonome source of truth.
- Knowledge graph for people, projects, companies, decisions, and relationships.
- Decision queue interface for review, approval, delegation, and feedback.
- Personal, role-based, and organisational Autonomes.
Personal, educational, research, and non-commercial experimentation is encouraged.
Commercial implementations, hosted services, enterprise deployments, consulting offerings, resale, white-label products, software-as-a-service platforms, managed services, or commercial derivatives may require a licence where they implement patent-backed Autonome concepts.
This repository is a public-facing overview and reference surface. Patent-sensitive claim detail, private operating records, credentials, commercial negotiations, and internal project data do not belong here.