Frequently asked questions.

Server-side policy checks measure p50 2.1 ms and p99 3.9 ms, including the database round-trip, against local Postgres in a single sequential worker. That excludes HTTP authentication, network time, and model-backed semantic checks — see /benchmarks for the method. Over the network your client adds round-trip time, and your agent's user-facing latency is usually dominated by model inference.

Yes. Self-hosted deployment is available from the Starter plan upward; the Free plan is non-commercial cloud evaluation only. You get the same Docker images we run in our cloud, plus a JWT license key. Agent and trace data stay on your infrastructure; the license check sends usage metrics only, and air-gapped mode sends nothing. See the self-hosted page for details.

SOC 2 Type II, ISO 27001, EU AI Act, GDPR, HIPAA, PCI DSS, and NIST AI RMF. Compliance reports are generated from your actual audit log — not from a questionnaire — and map to the specific controls each framework requires (CC6.1, A.9.1, Art. 14, Art. 30, §164.308, etc.).

Those tools focus on input/output validation at the prompt layer. Execlave is the governance layer one level deeper: it sits between the agent and the systems it can act on (tools, APIs, databases, files). Prompt validation catches some attacks; tool-level enforcement blocks unauthorized actions routed through the gate regardless of which prompt produced them, and produces the audit trail your compliance team needs. They are complementary, not competing.

Yes. Beyond runtime enforcement, Execlave provides the full agent-management control plane: tiered autonomy governance, a real-time cost circuit breaker, an agent registry with lifecycle and versioning, permission-drift detection, eval-to-policy suggestions, and data-access lineage. All are on by default in both the managed cloud and self-hosted deployments.

Yes. Execlave baselines each agent's tools, data sources, and permissions, then continuously detects drift: privilege escalation (new tools or domains it was never granted), anomalous access to sensitive or PII data, and unused over-privileged permissions it no longer needs. High-severity drift fires a notification through your configured channels, and the permission posture is included in compliance reports.

Yes. The real-time cost circuit breaker enforces spend caps synchronously in the policy path — per organization, agent, user, or workspace, across 1-minute, 1-hour, 1-day, and 1-month windows. It can warn, block, or pause the agent when a cap is hit, and burn-rate alerting warns you when projected spend will breach a budget before it actually does, rather than reconciling after the bill.

Yes to both. You assign each agent an autonomy level — observe, advise, act-with-approval, or autonomous — and Execlave applies the recommended policy bundle for that tier, flagging drift when an agent's policies no longer match its level. Separately, eval-to-policy suggestions turn detected anomalies, violation patterns, and eval failures into proposed policies with confidence scoring; you accept, modify-and-accept, or reject — nothing is auto-enforced.

The Execlave audit chain is where every block, violation, and approval on your agents’ customer-data actions is recorded in a tamper-evident, auditor-acceptable format. After 90 days of use, removing Execlave doesn't just remove a tool — it removes your compliance evidence. Most customers treat this as a feature. We treat it as honesty.

You do. In cloud mode, your data is encrypted at rest, isolated by organization via Postgres Row-Level Security, and never used to train any model. In self-hosted mode, we never see your data — it lives entirely on your infrastructure. We can prove this because the only thing that crosses the boundary in self-hosted is a license heartbeat (license key, instance fingerprint, hostname, version, and trace/agent/user counts — no trace content or customer records; air-gapped mode sends nothing).