Question 1

What is an LLM gateway?

Accepted Answer

An LLM gateway is a proxy between your application and the upstream model providers. It routes requests, enforces security policies, meters usage, and gives you observability — through one OpenAI-compatible endpoint. Anoman is the guarded LLM gateway: prompt-injection defense, PII masking, and policy control run on every call.

Question 2

Is Anoman a drop-in replacement for OpenAI?

Accepted Answer

Yes. Anoman exposes a fully OpenAI-compatible API — point your base_url at Anoman and nothing else changes. Any SDK that works with OpenAI (Python, TypeScript, Go, and others) works with Anoman out of the box.

Question 3

How does Anoman stop prompt-injection attacks?

Accepted Answer

Every request passes an ML classifier trained for prompt-injection detection (configurable threshold) plus heuristic and content-moderation checks, blocking malicious prompts before they ever reach a model. Injection detection is mandatory — only the sensitivity is adjustable, never the check itself.

Question 4

What latency does the gateway add?

Accepted Answer

Under ~30ms at p50 and ~80ms at p95, including the full guardrail pipeline (injection detection, PII masking, content moderation, policy). A typical model call takes 500ms–3s, so the overhead is negligible.

Question 5

Is my data safe from Anoman itself?

Accepted Answer

Your prompts and completions are never used to train any model. Cache is partitioned per customer so no one sees another customer's data, API keys are bcrypt-hashed, and PII is detected and masked on every request before it reaches any model.

Question 6

Where does my data actually go?

Accepted Answer

It's processed and stored in your configured region. Indonesian data stays in Jakarta and only crosses a border if a specific model you choose declares it — and that's shown on the model's detail page. No silent cross-border transfers.

Question 7

Is there any lock-in?

Accepted Answer

No. Anoman is OpenAI-compatible — you changed two lines of config to come in, and you can change them back to leave. There's no proprietary SDK to rip out. Start free, cancel anytime.

Question 8

Why trust a newer, Indonesia-based vendor?

Accepted Answer

Because trust here is instrumented, not asserted: every response returns a verifiable receipt of exactly what we did (region, guardrail results, routing, cost), we run a public status page and the Founding 50 program, and Anoman is built by security experts with more than 15 years in enterprise cybersecurity.

Question 9

Do you have SOC 2 or ISO 27001?

Accepted Answer

Formal certification is planned. In the meantime we handle enterprise security reviews and vendor questionnaires directly, and every call already returns machine-readable evidence of the controls applied — contact us for provider documentation and a Data Processing Agreement.

Question 10

What's your uptime and SLA?

Accepted Answer

We target 99.9% monthly on the gateway and publish live status at anoman.io/status. A contractual SLA is available on the Enterprise plan.

Question 11

Why not just use LiteLLM or build this myself?

Accepted Answer

You could — and then you'd own building and maintaining guardrails, PII masking, data residency, batch routing, observability, and audit logging yourself, forever. Anoman ships all of them on every call, maintained, behind one OpenAI-compatible endpoint.

Cloud Direct (OpenAI, Anthropic, Google)	1×
Bedrock / self-hosted	0.5×
Local ID (Jakarta edge)	0.3×

Budget (Gemini Flash, Llama, Haiku)	1×
Mid (Claude Sonnet)	4×
Premium (Claude Opus)	17×
Ultra (largest models)	42×

Cache hit (semantic)	0% of quota used
Provider cache (prefix)	10% discount
Batch routing	50% discount
Real-time	100% (full cost)

Model	Tier	Weight	Starter (500K wt)	Pro (20M wt)
Gemini 2.5 Flash	Budget	1×	500K calls	20M calls
Llama 3.3 70B	Budget	1×	500K calls	20M calls
Claude Haiku 4.5	Budget	1×	500K calls	20M calls
Claude Sonnet 4.6	Mid	4×	125K calls	5M calls
Claude Opus 4.7	Premium	17×	~29K calls	~1.2M calls

Monthly calls	Markup	Note
0–1,000	0%	Pure pass-through
1,001–5,000	1%	Near pass-through
5,001–20,000	2%	Below OpenRouter (10–25%)
20,001–50,000	3%	3–5× cheaper than OpenRouter
50,001+	5% cap	Still 2–5× cheaper than OpenRouter

Simple, transparent pricing.

Guardrails on every call — even the free ones

Guarded, not raw

Best-effort, honest

Pay in Rupiah when you grow

Not all tokens are equal

Provider Multiplier

Model Tier Multiplier

Quota Discounts

How far do your tokens go?

Volume-based progressive markup

Pick where your data is processed

Everything you get

Ready to secure every AI call?

Frequently Asked Questions

Feature	Free	Starter	Pro	Pay-As-You-Go	Enterprise
Weighted tokens/mo	2M/day, best-effort	500K	20M	Unlimited	Custom
Overage	Hard cap	Hard cap	$1.20/1M wt	0–5% markup	Negotiated
Token cap (budget, mo)	2M/day	25M	120M	No cap	No cap
Token cap (mid, mo)	—	—	24M	No cap	No cap
Token cap (premium, mo)	Trial	—	4M	No cap	No cap
Ultra-class token cap	—	—	—	No cap	No cap
Requests / min (rate)	20	30	120	60	30,000
Tokens / min (rate)	40K	60K	500K	200K	100M
Max concurrent requests	25	60	150	125	500
Budget models
Mid models
Premium models	Trial
Ultra models
Community models (opt-in)
API keys	1	1	5	5	Unlimited
Prompt injection detection
PII masking (4 modes)
Content moderation
Conversational guardrails
Agent policy control
MCP governance
Batch routing (50% savings)
Token packs
Semantic cache
Team management + RBAC
Organization workspaces
Member groups + entitlements
SSO/SAML
Audit log export
Custom rate limits
SLA guarantee
Net-30 invoicing
IDR billing
Support	Community	Community	Email	Email	Dedicated