anoman
Trust Center

Trust & Security

Trust is the foundation of every AI infrastructure decision. This page outlines Anoman's commitments to data sovereignty, reliability, and transparency for every customer.

TRUST SUMMARY

How Anoman keeps your customers' data secure & governed

Every AI request that runs through Anoman — a guarded LLM gateway — is screened by its guardrails; faithfulness scoring is opt-in. These numbers are live counts of all AI calls guarded through Anoman, including our own teams' internal use.

Infrastructure secured by AnomanLive configuration
GATEWAY & DATA STORAGE
Jakarta, ID
FRAMEWORK
UU PDP
GATEWAY STATUS
Operational
TRAILING WINDOW
90 days

CONTROLS IN EFFECT

417,069
AI calls guarded
64,363
PII entities masked
2,689
Requests blocked by injection rules, last 90 days (from November 2026, prompt-injection blocks come only from accounts with injection blocking on)
297
Content violations blocked
849
Faithfulness checks

627,999 AI calls · 4.5B tokens since May 2026

Always-on guarantees

  • Account data & logs stored in Jakarta (subprocessors listed on the privacy page); each model's processing region disclosed
  • Guarded models never train on your prompts & completions (the may-train community lane is opt-in)
  • Every response carries an _anoman evidence block
  • API keys hashed; cache partitioned per customer

THE SIGNATURE

Trust here is instrumented, not asserted.

Every response Anoman produces carries this evidence block. The controls above aren't a claim on a page — they're recorded on each individual call.

_anoman
{
  "_anoman": {
    "guardrails": {
      "injection": { "status": "pass" },
      "pii": { "status": "pass", "mode": "redact", "entities_processed": 0, "entity_details": [] },
      "content": { "status": "pass" },
      "policy": { "status": "pass" }
    },
    "routing": { "mode": "realtime", "region": "id", "provider_type": "cloud_direct", "provider_region": "US" },
    "cache": { "hit": false, "type": "none" },
    "weighted_tokens": 1240,
    "cost_usd": "0.0023",
    "burst": { "active": false },
    "egress": { "result": "pass", "dimension": null, "value": null, "data_class": "internal" }
  }
}

Generated by Anoman — the guarded LLM gateway. Counts every AI call guarded through Anoman, including our own teams' internal use. For a full security review or vendor questionnaire, contact us for provider documentation.

Generated: Nov 22, 2026

Our Commitments

  • Data location: Account data and logs are stored in Jakarta (subprocessors listed on the privacy page). A request leaves Indonesia only when the model you choose is processed elsewhere, and every model lists its processing region.
  • No markup on chat input and output tokens for most models: they bill at the provider's list price. A few models whose provider list price doesn't cover our serving cost carry a markup of up to 10%, and each model's page shows the exact price you pay. We keep 15% of the provider's cache saving on cached-input reads, image and video carry a margin, and decision models bill at the provider's price plus 15%; the rest of our revenue comes from plans.
  • Transparency: Every API response includes an _anoman metadata block showing guardrail results, cache status, routing decision, and actual cost.
  • No AI training on your data: Anoman never trains on your data, and guarded models' providers never train on it either; the may-train community lane is opt-in.

Uptime & SLA

We target 99.9% monthly API gateway uptime. Our batch routing system includes an automatic SLA escalation: if a batch job reaches 80% of its SLA window without completing, it is automatically promoted to real-time routing at no extra charge.

System status and active incidents are published at anoman.io/status.

Data Handling

Our data handling principles:

  • API keys are bcrypt-hashed before storage — raw key shown exactly once
  • Response cache entries are partitioned per customer — cross-customer cache leakage is treated as a critical security bug
  • All database writes stay in the Jakarta region — currently Anoman's only live region
  • Sensitive data is detected and masked on every request by our PII detection engine — SG NRIC, Indonesian NIK, Malaysian MyKad, Thai National ID, emails, and credit cards (phone, IP, and URL on request)

Provider Relationships

Anoman routes to leading LLM providers through our routing engine. Each provider handles request content under its own terms; the providers are named in our Privacy Policy. We select providers based on latency performance, reliability, and regional availability. Provider health is monitored in real time and surfaced on the status page.

Security

Security controls sit in the gateway, the data layer and our operations. This section covers our infrastructure security posture, encryption standards, access controls, and how to report vulnerabilities.

Infrastructure Security

All production workloads run in our Indonesia data center (Jakarta). Our infrastructure:

  • Private network with no public-facing database endpoints — all services communicate over private IPs only
  • Production runs on a single node in one region (Jakarta); there is no hot standby or multi-region failover yet
  • Single primary database with continuous WAL archiving and point-in-time recovery
  • In-memory cache with append-only persistence — private network only, snapshots every 6 hours
  • Daily database backups plus continuous WAL archiving to object storage in Indonesia, 14-day retention

Encryption

  • TLS (HTTPS) for all external connections
  • API keys hashed with bcrypt before storage — raw key shown once at creation and never stored
  • Secrets managed via a private secret store — no .env files in production containers

Access Control

Dashboard access uses a session-based authentication layer with short-lived JWT sessions (1-hour max age). RBAC enforces five roles, checked server-side on every mutating request — not just hidden in the UI:

  • Owner — full control: billing, team, every key and policy
  • Admin — manage members, API keys, and guardrail policy
  • Admin (read-only) — SOC 2-friendly auditor seat: read everything, change nothing
  • Member — use the gateway, manage their own keys and policies
  • End-user — governed Copilot-only seat: no dashboard, no key minting

The internal ops dashboard requires Google sign-in, and only accounts on an email allowlist are admitted (an empty allowlist admits no one).

Incident Response

Our incident response process follows a detect → contain → eradicate → recover workflow:

  • Automated email alerts to the operations team for resource exhaustion, unavailable upstreams and anomaly detection triggers
  • Alerts are delivered by email; we do not yet run a 24/7 paging rotation or commit to an acknowledgement time
  • Affected customers notified within 72 hours of confirmed breach
  • Active incidents are shown on the status page (anoman.io/status)

Vulnerability Disclosure

We operate a responsible disclosure policy. If you discover a security vulnerability, please report it to [email protected] with a description of the issue, steps to reproduce, and potential impact. We acknowledge reports within 48 hours and aim to resolve critical issues within 7 days.

Compliance & Certifications

Frameworks we build against:

  • UU PDP (Law No. 27/2022) — Indonesian personal data protection law. UU PDP-ready: account data and logs are stored in Jakarta, and each model's processing region is disclosed.
  • GDPR principles — applied globally as a baseline standard

SOC 2 certification is on our roadmap for Enterprise customers. Contact us at [email protected] for our current security questionnaire.

Contact

For trust and security questions, contact PT Anoman Artifisial Informasi at [email protected]. For enterprise security reviews and questionnaires, please use the same contact.