Enterprise

Argentor ships enterprise-grade infrastructure — RBAC, multi-tenancy, quotas, compliance modules, and a readiness endpoint — without external plugins.

Readiness Endpoint

A single endpoint reports the complete deployment posture:

GET /api/v1/enterprise/readiness

Example response:

{
  "version": "1.3.0",
  "posture": "ready",
  "score": 71,
  "runtime": {
    "skills_registered": 42,
    "active_connections": 0,
    "active_sessions": 0,
    "uptime_seconds": 60
  },
  "checks": [
    {
      "id": "rest_api",
      "category": "runtime",
      "title": "REST API mounted",
      "status": "active",
      "detail": "REST management endpoints are available under /api/v1."
    }
  ],
  "next_actions": [
    "Wire deployment-specific auth, SSO, rate limits, and approval policy."
  ]
}

Check statuses

RBAC & Permissions

Role-based access control is built into argentor-security. Each agent runs with an explicit PermissionSet that controls which skills it can invoke and which resources it can access.

use argentor_security::{PermissionSet, Capability};

let permissions = PermissionSet::builder()
    .allow(Capability::FileRead)
    .allow(Capability::WebFetch)
    .deny(Capability::ShellExec)
    .build();

let runner = AgentRunner::new(config, registry)
    .with_permissions(permissions);

Multi-Tenancy

The gateway supports multi-tenant deployments where each tenant has isolated sessions, separate rate limits, and audited access logs.

• Per-tenant API keys with independent quotas
• Session isolation — one tenant cannot read another's conversation history
• Per-tenant audit trails exportable to SIEM (JSON/CEF)
• Per-key rate limiting configurable at the gateway level

Compliance Modules

The argentor-compliance crate ships four compliance frameworks out of the box:

Security Guardrails

The 6-layer GuardrailEngine runs on every request and response:

1. PII detection — names, emails, phone numbers, credit cards, SSNs
2. Prompt injection — jailbreak patterns, instruction overrides
3. Toxicity — harmful, violent, or discriminatory content
4. Shell injection — command metacharacters in tool arguments
5. Base64 exfiltration — encoded payloads in outputs
6. Unicode smuggling — homoglyphs, zero-width characters

let guardrails = GuardrailEngine::builder()
    .with_pii_scanning(PiiScanMode::RedactAndLog)
    .with_prompt_injection(InjectionMode::Block)
    .with_toxicity_threshold(0.8)
    .build();

Observability

Argentor exposes first-class observability without external plugins:

• Prometheus metrics at /metrics — latency histograms, skill call counters, guardrail hit rates
• OpenTelemetry traces — every agent turn, skill call, and LLM request is a span
• Structured audit log — JSON/CEF, queryable, SIEM-exportable
• Health endpoint at /health — liveness + readiness signals

Quotas & Rate Limiting

cargo run -p argentor-cli -- serve \
  --bind 0.0.0.0:8080 \
  --rate-limit-per-key 1000/min \
  --max-concurrent-sessions 500 \
  --token-quota-per-day 10000000

Human-in-the-Loop

The orchestrator supports approval gates — pause execution and wait for a human decision before a specific skill executes:

use argentor_orchestrator::ApprovalPolicy;

let policy = ApprovalPolicy::require_approval_for(vec![
    "shell_exec",
    "file_delete",
    "api_key_rotate",
]);

let orchestrator = Orchestrator::new(config)
    .with_approval_policy(policy);

Smoke Test

cargo test -p argentor-gateway --test regression_api enterprise_readiness

Production Deployment

See the full deployment guide for Docker, Kubernetes, Helm, and multi-region setups:

• DEPLOYMENT.md
• ENTERPRISE_GOLDEN_PATH.md