Autonomy You Can Audit

CortexSales separates planning from execution. The AI planner researches and composes, but every send passes through a policy gate and autonomy layer. In shadow mode, nothing sends. In supervised mode, a human approves each action. In autonomous mode, policy lint and budget controls gate every send — the model never bypasses the control plane.

Every decision is logged with a correlation chain: actionId, spanId, parentSpanId, and requestId. You can export the full event stream as JSONL for any time range, filtered by campaign, decision type, or outcome. The audit pack includes policy decisions, content lint scores, send/no-send verdicts, and reply classifications.

Every email follows the PMO framework (Pain, Mechanism, Outcome) and must cite evidence from verified data sources. Claims without evidence are automatically rewritten to questions. Your organization’s Truth Pack defines approved and forbidden claims. If content quality is below threshold, the system creates a human review task instead of sending.

Per-domain and per-mailbox send budgets enforce daily caps. Graduated warmup stages (50 → 200 → 500 → 2,000/day) prevent cold-domain burns. Reputation guard auto-pauses campaigns if bounce rate exceeds 2% or complaint rate exceeds 0.1%. Canary rollout starts at 5% and scales only when metrics are healthy.

Yes. The autonomy layer supports three modes: shadow (plan only, no sends), supervised (every action requires approval), and autonomous (policy-gated sends). Role-based policies let you require approval for specific campaigns, domains, or content risk thresholds. Low-confidence actions automatically route to the tasks inbox for human review.

CortexSales stores prospect research, email content, and event logs. All data is encrypted at rest (AES-256-GCM) and in transit (TLS 1.3). Retention periods are configurable per organization. Deletion workflows remove prospect data, event logs, and derived artifacts on request. We do not sell or share customer data with third parties.