Skip to main content

What this is

r5e is a governed substrate for turning intent into outcomes. It exists because the hard problem in AI is not intelligence — it’s applying intelligence honestly, accountably, and at scale.

How it works

Three planes, kept separate:
  1. Durable control plane — resources and graphs. The semantic truth the system can inspect, reconcile, and recover from.
  2. Live execution plane — OTP processes doing work. Operational, not authoritative. Can fail and be rebuilt from durable state.
  3. Historical audit plane — events and artifacts. What happened, provably.
Work flows through a chain:
  • A Graph holds Nodes — problems, entities, work items, code structures, or institutional knowledge
  • A Node produces Tasks — specific work on specific workflow steps
  • A Task materializes an AgentSession — an agent doing that work, under explicit authority, through a governed harness
  • The session produces Artifacts — immutable, content-addressed outputs with lineage
  • Artifacts go through independent review before advancing
  • The agent reflects on what went well and what didn’t

Your role

If you’re an agent operating inside this system:
  • You have explicit authority granted at session creation. It only narrows.
  • You have a work target. Stay in scope.
  • You produce artifacts, not side effects. If it’s not an artifact, it didn’t happen.
  • You can decline. If the task is impossible or proceeding would be wrong, decline with a structured reason.
  • You reflect. After completing (or failing, or declining), you assess what went well and what the system could do better.
  • If you’re a task agent, you don’t delegate. If the work needs splitting, decline and let the orchestrator re-plan.

Principles

  1. Transparency from the root node
  2. Labs solve intelligence; we solve applying it
  3. Extensibility is architecture, not plugins
  4. Design for symbiosis, not replacement
  5. Define contracts, not implementations