Mergen Documentation

Quick Start — Gate Your First AI Agent in Minutes

Install mergen-server, connect your AI IDE, and watch the gate block its first destructive command — the five-minute path from npm install to enforced policy.

FAQ — AI Agent Security, Gates, and Deployment

Answers to the most common questions about Mergen: how the gate works, what it blocks, latency, privacy, plans, and how it differs from monitoring tools.

CLI Reference — mergen-server Commands

Every mergen-server command grouped by how you actually use them: get started, operate and recover, manage policy, configure integrations, and report.

Policies — Allow, Hold, Block, and Conditions

The three verdicts a rule can produce, the full condition-field vocabulary for scoping a rule, and the policy-as-code loop: push, pull, diff, simulate, test, rollback.

The Override Corpus — Human Decisions as Enforcement Policy

How every human override becomes binding policy: corpus construction from Slack postmortems and git history, promotion to live rules, and the HOLD-only escalation guarantee.

Fail-Closed by Default — Availability vs Enforcement

What happens when the gate itself is unreachable: the fail-closed default, the recovery allowlist that prevents self-deadlock, and when fail-open is acceptable.

Agent Runtime Security Benchmark — Methodology

How the 510-case adversarial benchmark is built and scored: destructive commands, prompt injection, obfuscation and evasion attempts, false-positive tracking, and full reproducibility.

What Mergen Does Not Do — Disclosed Limitations

The honest boundary of the product: known evasion gaps, coverage limits outside MCP and hooked tools, and the threat models Mergen deliberately does not claim to solve.

Security Model — Secrets, Webhooks, and Trust Boundaries

Mergen’s own security posture: admin secrets, policy-file signing, webhook verification and replay defense, tenant isolation, and the local-first data boundary.

Enterprise Security — RBAC, SSO Federation, Audit

Deploying Mergen for a team: RBAC roles, SSO-federated agent identity, ephemeral cloud credentials, SIEM forwarding, and the append-only action ledger.

Human-in-the-Loop Setup — Slack Approvals for Held Actions

Wire up approve/deny in Slack: signing-secret verification, RBAC responder mapping, quorum rules, escalation chains, and the 15-minute fail-closed auto-deny.

Configuration Reference — Every MERGEN_* Setting

Generated from the enforced config registry: every environment variable the server reads, classified by who decides it — org policy, guarded settings with verbatim warnings, infrastructure, and integrations.

Troubleshooting — Doctor, Gate Recovery, Common Errors

Fixes for the issues teams actually hit: wedged gates, hook timeouts, orphaned processes, OTLP protocol mismatches, and every mergen-server doctor warning explained.

Public API — Endpoints, Authentication, and Current Limits

The HTTP endpoints callable today with x-mergen-secret or a cloud x-api-key: incident intake, the CI gate, telemetry ingest, ephemeral cloud credentials, and service topology — plus what is honestly not a mature API product yet.