Documentation.

Install the SDK, register your first agent, and send a trace in minutes.

Run Execlave on your own infrastructure with Docker Compose and a license key. Agent and trace data stay in your network.

19 built-in policy types — injection scanning, access control, cost limits, PII protection, output classification, budget caps, tool integrity, runtime tool guardrails, CEL expressions, response groundedness, OPA Rego policies, agent-lineage controls, and more. Four enforcement modes: block, warn, monitor, require_approval.

Tiered autonomy levels, real-time cost circuit breaker, agent registry & lifecycle, permission-drift detection, eval-to-policy suggestions, and data-access lineage — the AMP control plane.

Intent-aware and multilingual. Tells a definitional question ("what is a system prompt?") from an attack ("reveal your system prompt"), covers 13 languages, and escalates ambiguous cases to an LLM semantic layer.

Checksum-validated national IDs across the US, EU, and India (SSN, cards, Aadhaar, IBAN, NINO, DNI/NIE, Codice Fiscale, IdNr, INSEE), plus opt-in name/address NER via Presidio + GLiNER.

Write custom governance rules using CEL expressions. Access 11 context variables including input, output, cost, tools, environment, and metadata.

Evaluate multiple policies in sequence. Control ordering, short-circuit behavior, and how violations from different policies interact.

Register external HTTP endpoints as policy validators. Execlave calls them during enforcement with HMAC-signed payloads and retries on failure.

Human-in-the-loop approvals with computed risk scores, tier routing, SLA escalation, and a continuous authorization-certificate verification loop.

Issue short-lived RS256 exe_agt_ credentials and verify them offline against a public JWKS endpoint — verifiable identity for machine-to-machine and agent-to-agent calls.

Author policies as JSON/YAML bundles, validate with execlave policies lint, reconcile with sync, and audit every change through tamper-evident version history.

Per-agent tool-invocation allow/deny policy, tool-output PII/injection scanning, and agent-to-agent credential authorization.

Delegation chain-of-custody, capability inheritance, revocation cascade, declared-vs-observed shadow-tool diff, and a non-human-identity inventory.

Run an adversarial probe suite with execlave test; gate promotion to the autonomous tier on a minimum resilience score (0.9 by default).

Attribute spend by cost center, project, and team; export chargeback CSV; set budget caps; and forecast month-end spend.

Detect behavioral drift — new tool usage, denial-rate shifts, new data classes — and automatically downgrade an agent’s autonomy on high-severity drift.

Bind agents to Microsoft Entra workload identities and exchange Entra-issued tokens for short-lived Execlave credentials — no static API keys to rotate or leak.

What happens when an SDK cannot get an enforcement decision, how each ungoverned stretch is recorded as an enforcement.bypassed audit entry, and how it appears in the signed compliance report.

The fourth outcome: the target accepted the request but nobody saw it happen. effectEvidence, the effect_unconfirmed status, resolutions that never rewrite the trace, outcome strictness, the unconfirmed queue, read-back checks, and coverage in the signed report.

Policies in your own words, judged by the deployment’s model after each action: monitor and warn only, every action recorded as violation, no violation or not judged, examples checked before saving, and the model qualified on labelled actions before it may judge.

Capture inputs, outputs, token usage, cost, latency, and policy evaluations. Supports hierarchical spans, trace trees, similarity search, and CSV/JSON export.

Subscribe to 16 event types (policy_violated, kill_switch_triggered, budget_exceeded, and more). Deliveries are HMAC-SHA256 signed and retried with exponential backoff.

Define alert rules on metrics with configurable thresholds, severity levels, and time windows. Supports email, Slack, and webhook notification channels.

ML-powered detection using EWMA with seasonal decomposition. Includes root cause analysis, time-series forecasting, and configurable baselines.

Stream enforcement decisions and traces to Splunk via HTTP Event Collector, with SPL saved searches for violation bursts, cost spikes, and blocked agents.

Route governance events into Sentinel through the Azure Monitor Logs Ingestion API, with KQL analytics rules ready for your SOC queue.

A six-stage runbook for agent incidents: detect, route, triage, contain, collect evidence, and close with a hash-chained audit export.

Authenticate via Logto JWT or environment-scoped API keys (exe_dev_, exe_stg_, exe_prod_). Four-role RBAC hierarchy: Owner > Admin > Developer > Viewer.

Generate digitally signed evidence reports for SOC 2, GDPR, HIPAA, ISO 27001, and EU AI Act. Verify report authenticity via the compliance API.

Tamper-evident, append-only audit trail with cryptographic chain verification. Export as CSV/JSON or as a signed compliance package for auditors.

Human-in-the-loop workflows for require_approval policies. SDKs poll for the decision while admins approve or deny from the dashboard.

Callback handler that instruments chains, LLMs, tools, agents, and retrievers — Python and TypeScript.

Tracing processor that captures agent/tool/generation spans and blocks disallowed tools before they run.

One call to instrument_crew() wires in step and task callbacks, chaining your existing callbacks.

Instrument AutoGen multi-agent conversations with trace capture and policy enforcement per agent turn.

Callback handler for LlamaIndex query engines, retrievers, and synthesis steps.

Gate your n8n AI workflows with Execlave policies and trace logging — no-code templates included.

Govern MCP tool servers — trace tool calls and enforce allow/deny policies on MCP tools before they execute.

Drop-in wrapper for direct OpenAI chat completion calls — traces, cost capture, and policy checks without a framework.

Browse every first-party framework integration and what is on the roadmap.

Full API reference for the TypeScript and Python SDKs — methods, config, error types.

Every endpoint with request/response examples and cURL snippets.

Install the execlave CLI, authenticate via env var, and run all 11 command groups from the terminal.

See how Execlave compares to other AI governance solutions, including Microsoft Agent Governance Toolkit.

A governed two-agent production architecture end to end: stack, autonomy tiers, policy set, measured enforcement latency, and SIEM routing.