Aramm
Now in private beta  ·  ASEAN-first LLM gateway

One API.
Every LLM.
Zero overhead.

Aramm is a Singapore-hosted LLM gateway built for ASEAN engineering teams. Route across Anthropic, OpenAI, Bedrock, and Azure with one API key, unified billing, and built-in PDPA guardrails.

your_app.py
# Before
- client = openai.OpenAI(api_key="sk-...")

# After — swap one line, access every model
+ client = openai.OpenAI(
+     api_key="sk-arm-v1-...",
+     base_url="https://gateway.aramm.com/v1"
+ )

response = client.chat.completions.create(
    model="claude-sonnet",  # or gpt-4o, llama-3...
    messages=[{"role": "user", "content": "Hello"}]
)

Works with

Anthropic
OpenAI
AWS Bedrock
Azure OpenAI
OpenRouter

Features

Everything between your app and the LLM

Aramm handles the infrastructure so your team ships product — not boilerplate.

Multi-provider routing

Automatic failover across Anthropic, OpenAI, Bedrock, Azure, and OpenRouter. 5xx triggers retry with backoff to the next provider.

Credits & billing

Stripe-powered top-ups, real-time balance, per-key spend limits, and monthly budget caps. Auto top-up when balance runs low.

Rate limiting

Redis sliding-window RPM and TPM enforcement per API key. Per-plan limits: Free (10 RPM), Starter (60), Pro (300), Enterprise (custom).

Observability

Token, cost, and latency analytics per model and key. p50/p95/p99 percentiles, error breakdowns, and CSV export from the dashboard.

PDPA guardrails

Singapore NRIC/FIN, phone, email, and credit card detection. Block, redact, or flag per API key. Compliant with PDPA and MAS TRM.

Exact-match cache

Identical prompt + model returns cached response with x-aramm-cache: hit. Zero credit deduction. Configurable TTL per key.

Get started

Live in under five minutes

No infrastructure to manage. No provider contracts to sign. Just an API key and a base URL.

01

Create an account

Sign up with email or Google. An organisation and default workspace are created automatically.

02

Top up credits

Add credits via Stripe. Balance is tracked in real time. Set low-balance alerts or enable auto top-up.

03

Generate an API key

Create a key scoped to your workspace. Set per-key rate limits, model allowlists, and spend caps.

04

Swap your base URL

Point your OpenAI-compatible SDK at gateway.aramm.com/v1. Every model, one key, zero code changes.

Providers

Every provider. One base URL.

Aramm routes to the provider that matches your model alias. Failover is automatic — no client changes needed.

Anthropic

Claude family — best for reasoning, coding, and long-context tasks.

claude-sonnet-4-6claude-haiku-4-5
OpenAI

GPT-4o family — strong general-purpose and multimodal tasks.

gpt-4ogpt-4o-mini
AWS Bedrock
In-region

ap-southeast-1 deployment. SigV4 auth handled automatically.

Claude on BedrockNova Lite/Pro
Azure OpenAI
In-region

Southeast Asia endpoints with per-deployment key management.

gpt-4o (deployment)gpt-4o-mini
OpenRouter

Access 200+ open-source and proprietary models via one endpoint.

Llama 3.1 70BGemini Flash 1.5
Vertex AI · vLLM

Gemini Pro/Flash on Google Cloud and self-hosted vLLM — coming in Phase 3.

Integration

Works with every OpenAI-compatible SDK

Change two lines. Keep your existing SDK, tooling, and workflow — Aramm speaks OpenAI format.

from openai import OpenAI

client = OpenAI(
    api_key="sk-arm-v1-YOUR_KEY",
    base_url="https://gateway.aramm.com/v1",
)

response = client.chat.completions.create(
    model="claude-sonnet",
    messages=[
        {"role": "user", "content": "Explain neural networks simply."}
    ],
    stream=True,
)

for chunk in response:
    print(chunk.choices[0].delta.content or "", end="")

Observability

Catch problems before your users do

Real-time visibility into every LLM call. Spot slowdowns, cost spikes, and error patterns the moment they happen — not after a postmortem.

Know where every dollar goes

Costs attributed per request, model, and team. No more end-of-month guesswork.

Know when it's the model, not your app

Provider-level latency data makes root-cause faster — and SLA conversations easier.

Fits the stack you already have

Send traces to Langfuse, your own webhook, or any OTel-compatible tool. No new dashboards required.

Your app always knows what happened

Model used, cache result, latency, and credits remaining in every response header. Build tooling that doesn't break.

Usage · Last 7 days

Requests

12,847

Total cost

$24.13

Avg TTFT

410ms

ModelStatusTokensCostTTFT
claude-sonnetok1,247$0.0037312ms
gpt-4o-miniok843$0.0002198ms
claude-haikucache$0.0011ms
gpt-4oerror

Guardrails

Compliance enforced, not just documented

Every request screened on the hot path. PII detected, handled, and logged — before your users' data ever leaves your perimeter.

Built for regulated markets

Goes beyond generic detection with Singapore-specific patterns — NRIC checksum validation, local phone formats, and postal codes included.

One policy, three outcomes

Block, redact, or flag — per API key. Dev teams use flag mode; customer-facing keys get block. No separate deployment needed.

Keep data in the right region automatically

Requests blocked if the resolved provider sits outside your allowed jurisdictions. A runtime control, not a policy document.

gateway.aramm.com
POST /v1/chat/completions

→ PII detected: NRIC (S●●●●567D)
→ Action:       REDACT
→ Forwarding scrubbed request…

────────────────────────────────

x-aramm-pii-detected: true
x-aramm-pii-action  : redact
x-aramm-pii-types   : nric

HTTP 200 OK

Governance

Control who uses what — and how much

Set model access, spend limits, and team roles once. Aramm enforces them on every request — no one can route around them.

Audit Log
AllSecurityConfigAccess

Updated guardrail

Toxicity Shield threshold: 0.7 → 0.85

sarah@acme.com

2 min ago

PII blocked

SSN detected in req_d4f1, scrubbed before model

system

8 min ago

API key created

Key rq_live_****7f2a for Engineering team

mike@acme.com

15 min ago

Rate limit triggered

Marketing team at 100% budget ($1,500/$1,500)

system

22 min ago

Role changed

jake@acme.com: Developer → Admin

sarah@acme.com

41 min ago

FAQ

Common questions

Aramm is a managed LLM gateway for engineering teams in Singapore and ASEAN. You point your existing OpenAI-compatible SDK at our endpoint, and we handle provider routing, billing, rate limiting, and data compliance — so you ship product instead of gateway infrastructure.

No seat fees, no subscriptions. Top up credits in USD and pay only for what you use. Aramm adds a 5% service fee on each top-up; inference runs at provider cost. Credits never expire.

Content logging is off by default. You can opt in per API key, and when you do, PII redaction runs before anything is stored. You control the retention period — 30 days on Free, configurable up to custom retention on Enterprise.

Anthropic, OpenAI, AWS Bedrock (ap-southeast-1), Azure OpenAI (Southeast Asia endpoints), and OpenRouter (200+ models). Provider failover is automatic — a 5xx triggers a retry with backoff to the next available provider.

Aramm runs in AWS ap-southeast-1 (Singapore), with PII detection tuned for Singapore patterns — NRIC/FIN, local phone numbers, postal codes — validated against PDPA and MAS TRM requirements. We don't claim blanket compliance on your behalf; what we guarantee is what the infrastructure actually enforces.

Block returns a 422 without forwarding the request. Redact replaces the detected value with a placeholder and lets the scrubbed request through. Flag logs the detection and allows the full request through unchanged. Each API key can be configured independently, so dev and production keys can have different enforcement modes.

Yes. Any framework that accepts a base URL and an OpenAI-compatible API key works — LangChain, LangGraph, LlamaIndex, n8n, Dify, and custom agent loops. Aramm governs and bills the calls; it does not orchestrate or execute tools.

Request early access, top up credits, generate an API key, and change your SDK's base_url to gateway.aramm.com/v1. No provider contracts to sign, no infrastructure to provision. Most teams are live in under five minutes.

Pricing

No seats. No subscriptions.

Top up credits in USD and pay only for what you use. We charge a 5% service fee on each top-up — inference runs at provider cost.

Free

Explore Aramm

$0

No credit card

  • 10 RPM / 50K TPM
  • 2 API keys
  • All providers
  • Dashboard & usage
Request access
Starter

Small teams

Pay per use

Credits from $10

  • 60 RPM / 500K TPM
  • 5 API keys
  • All providers
  • PDPA guardrails
  • CSV export
Request access
Most popular
Pro

Growing teams

Pay per use

Credits from $50

  • 300 RPM / 2M TPM
  • 20 API keys
  • Workspace management
  • Exact-match cache
  • Audit logs
  • Auto top-up
Request access
Enterprise

Custom scale

Custom

Talk to us

  • Custom limits
  • Unlimited keys
  • SSO / SAML
  • Regional residency
  • SOC 2
  • Dedicated support
Contact us

All plans include streaming SSE, Prometheus metrics, and response headers (x-aramm-*). Credits never expire.

Early access

Built for ASEAN engineering teams

Aramm is in private beta. Join the waitlist and we'll reach out when access opens. Singapore-hosted, PDPA-ready, and production-grade from day one.

No spam. No marketing emails. Just a heads-up when you're in.