AI agents make thousands of decisions with no record of why. AEGIS evaluates proposals through six quantitative gates — returning structured decisions with hash-chained audit trails and compliance-aligned documentation.
AI agents and engineering teams make high-stakes decisions every day — deployments, architecture changes, resource allocations. But there is no standardized way to evaluate those decisions, and no audit trail to prove they were sound.
| Challenge | AEGIS Solution |
|---|---|
| No paper trail | Every evaluation produces a decision_id, timestamp, gate results, rationale, and hash-chained audit entry. |
| No standard for "good" | Six quantitative gates evaluate risk, profit, novelty, complexity, quality, and utility against calibrated thresholds. |
| Compliance can't audit | Hash-chained audit logs, NIST AI RMF artifacts, EU AI Act Annex IV technical file included. |
| Governance slows teams down | One API call. Five integration methods. Zero runtime dependencies in the core SDK. |
Every proposal is evaluated against six quantitative gates. The result is a structured decision — PROCEED, PAUSE, HALT, or ESCALATE — with confidence scores, rationale, and next steps. Every evaluation is hash-chained.
Proposal → [Risk] [Profit] [Novelty] [Complexity] [Quality] [Utility] → Decision
↓
PROCEED / PAUSE / HALT / ESCALATE
+ confidence + rationale + next steps
+ hash-chained audit entryMost governance tools either filter agent actions in real-time or manage compliance through checklists. AEGIS does neither — it evaluates the engineering decision itself through six quantitative gates.
| Approach | What It Does | Blind Spot |
|---|---|---|
| Runtime agent guardrails | Monitors agent behavior, gates capabilities via trust scores and circuit breakers | A trusted agent can still produce a bad proposal |
| Compliance dashboards | Policy management, checklists, binary pass/fail risk scoring | No quantitative rigor — a checklist can't evaluate mathematical risk |
| Quantitative decision governance | Bayesian posteriors, utility theory, KL divergence drift detection across six gates | Evaluates decisions, not runtime behavior — complementary to the other approaches |
# Install and evaluate in 30 seconds
$ pip install aegis-governance
$ export AEGIS_API_KEY="your-key"
# Python SDK
from aegis_governance import AegisConfig, PCWContext, pcw_decide
decision = pcw_decide(
PCWContext(
agent_id="my-agent",
proposal_summary="Add Redis caching layer",
estimated_impact="medium",
risk_proposed=0.15,
complexity_score=0.7,
)
)
# → DecisionStatus.PROCEED (confidence: 0.87)
# → decision_id: aegis-2026-03-24-a7f3...
# → audit hash: sha256:e3b0c44298fc...AEGIS includes compliance artifacts and runbook templates aligned with these frameworks. Alignment artifacts, not certifications. See compliance maturity →