Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/xcoder-es/governance-layer/llms.txt

Use this file to discover all available pages before exploring further.

Intelligence is not solely the ability to optimize decisions — it also requires the ability to govern the process by which decisions are made. The Governance Layer is an architectural specification and reference implementation built on that hypothesis. It introduces three interlocking mechanisms — the Neural Parliament, Ulysses Contracts, and the Identity Layer — that sit above an agent’s raw optimization engine and take responsibility for resolving competing objectives, constraining future choices, and anchoring a persistent identity across time.
Pre-alpha research prototype. The theoretical framework has undergone five rounds of adversarial review by an independent expert panel. The reference implementation is under active construction. Claims about security, formal verification, and production readiness are aspirational.

Why a Governance Layer?

Modern AI has made extraordinary progress on capability — reasoning, planning, tool use, and long-horizon tasks. What it has not solved is the higher-order problem that arises once capability is sufficient: which objectives should guide the decision, and how should conflicts between them be resolved? Current governance mechanisms — system prompts, reward functions, constitutional rules — are predominantly external to the intelligence itself. Engineers define the constraints; the model optimizes within them. As systems become more autonomous, the gap between external governance and internal behavior becomes a structural risk. The Governance Layer addresses that gap by making governance an internal cognitive capability. Rather than asking only “What action should be taken?”, a governed agent also asks:
  • Which of my objectives should take precedence right now?
  • Should I voluntarily restrict my future choices to protect a commitment I made?
  • Does this proposed action remain coherent with who I am?
These are questions of governance, not optimization, and they operate at a higher level of abstraction than any individual decision.

The Four-Layer Architecture

The framework is organized as a four-level stack, with a simulated Trusted Execution Environment (TEE) at the base:
LayerRole
Neural ParliamentSeven specialized members deliberate over proposals, apply budgets and vetoes, and reach a weighted consensus via the Speaker state machine
Ulysses ContractsPre-committed constraints the agent voluntarily locks before entering a high-stakes context, preventing self-amendment under pressure
Identity LayerA four-tier ontology of core commitments, values, and behavioral dispositions that must remain coherent across decisions
TEE EnclaveSimulated hardware isolation that seals governance operations behind a gradient barrier and cryptographic audit log
The separation between capability (what the agent can do) and governance (what it decides to do, and by what process) is the central design principle. Optimization is not replaced — it is governed.

The governance Package

The reference implementation lives in the governance Python package (Python 3.10+). Every layer of the formal specification maps directly to a module:
  • governance.speaker — the Neural Parliament Speaker state machine
  • governance.committee — seven Parliament member implementations
  • governance.contracts — Ulysses Contract lifecycle and enforcement
  • governance.identity — ontology, commitments, tiers, and cryptographic keys
  • governance.tee — simulated enclave, watchdog, and Merkle batching
The full set of public exports from the top-level governance package includes: SpeakerStateMachine, ParliamentMember, the seven example member classes (ExampleRewardMember, ExampleSafetyMember, ExampleIntegrityMember, ExampleCuriosityMember, ExamplePlanningMember, ExampleSocialMember, ExampleMemoryMember), UlyssesContract, ContractRegistry, IdentityCore, CoreCommitment, TieredMutability, MutabilityTier, Proposal, GovernanceDecision, PriorityTag, and Action.

Quickstart

Install the package, run your first governance cycle, and explore the Streamlit dashboard in under five minutes.

Architecture

Deep-dive into the four-layer stack — Speaker, Parliament, Contracts, Identity, and TEE — with annotated diagrams.

Neural Parliament

Seven members, weighted voting, proposal budgets, veto thresholds, and SDoS-resistant agenda management.

Ulysses Contracts

Pre-committed constraints, three κ enforcement modes, and cryptographic sealing against self-amendment.

Build docs developers (and LLMs) love