AEGIS Governance

AEGIS evaluates engineering proposals through six gate checks over caller-supplied metrics, returning structured PROCEED / PAUSE / ESCALATE / HALT decisions with a hash-chained audit trail.

Get Started API Reference (not served) Advisor (questionnaire) Product status

How It Works

Proposal → [Risk] [Profit] [Novelty] [Complexity] [Quality] [Utility] → Decision
                                                                          ↓
                                                          PROCEED / PAUSE / HALT / ESCALATE
                                                          + confidence + rationale + next steps

Every authenticated /evaluate call produces a decision_id, timestamp, gate results, rationale, and a hash-chained audit entry (sandbox evaluations are not recorded). The six gates:

• Risk — Is the risk delta acceptable? Bayesian posterior confidence evaluation.
• Profit — Does the expected value justify the change? Bayesian confidence check.
• Novelty — How novel is this approach? Logistic function scoring.
• Complexity — Is the system complexity within bounds? Hard floor enforcement.
• Quality — Does code quality meet standards? Minimum score with no zero subscores.
• Utility — Is the net utility positive? Lower confidence bound verification.

The metrics are supplied by the caller; the gates check them against documented thresholds.

Quick Example

from aegis_governance import AegisConfig, PCWContext, PCWPhase, pcw_decide

config = AegisConfig.default()
evaluator = config.create_gate_evaluator()

decision = pcw_decide(
    PCWContext(
        agent_id="my-agent",
        session_id="session-1",
        phase=PCWPhase.PLAN,
        proposal_summary="Add Redis caching layer",
        estimated_impact="medium",
        risk_proposed=0.15,
        complexity_score=0.7,
    ),
    gate_evaluator=evaluator,
)

print(decision.status.value)  # "proceed", "pause", "halt", or "escalate"

Integration Options

Key Features

• Hash-chained audit trails — decision logs that detect tail deletion and entry edits, verifiable against the stored chain; not independently anchored
• Hybrid post-quantum cryptography — signatures combining classical (Ed25519) with post-quantum (ML-DSA-65, FIPS 204) and encryption (ML-KEM-768, FIPS 203)
• Shadow mode — evaluate without enforcing the decision
• Drift detection — KL-divergence comparison of decision patterns against a baseline
• Configurable thresholds — YAML-driven parameter management with version-controlled schemas

Governance Templates

The repository contains self-authored governance templates and runbooks. They are not certifications, and no third-party compliance audit or certification assessment has been conducted on AEGIS itself. See compliance maturity.