Hosted in the UAE · Serving the GCC

Open-model inference, hosted in the UAE.

An OpenAI-compatible API — change one line. Low latency across the GCC, priced per token. Private by default: zero retention, data stays in-region. No GPUs to rent, run or babysit.

Get API key Talk to an engineer

Early accessFree credits*, dedicated capacity on demand

One APIRequests in

Open modelsYour choice

Rihstal AIRoutes, batches, servesfalcon-h1 → uae

Rihstal AI cloudUAE region · in-country

Your cloudDedicated or on-prem

OpenAI-compatible Zero data retention UAE data residency Pay per token Dedicated GPUs on demand

01Familiar API

Keep your SDK. Change one line.

Rihstal AI speaks the OpenAI Chat Completions format, so the clients, frameworks and agents you already use work as they are. Point them at our base URL and pick an open model.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.rihstal.ai/v1",  # ← the only change
    api_key=os.environ["RIHSTAL_API_KEY"],
)

stream = client.chat.completions.create(
    model="falcon-h1-34b-instruct",
    messages=[{"role": "user", "content": "لخّص هذا العقد في ثلاث نقاط"}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.rihstal.ai/v1", // ← the only change
  apiKey: process.env.RIHSTAL_API_KEY,
});

const res = await client.chat.completions.create({
  model: "qwen3-235b-a22b",
  messages: [{ role: "user", content: "Draft a bilingual reply to this customer." }],
  tools,                                  // tool calling works too
});

console.log(res.choices[0].message);

curl https://api.rihstal.ai/v1/chat/completions \
  -H "Authorization: Bearer $RIHSTAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama-3.3-70b-instruct",
    "messages": [{"role": "user", "content": "Marhaba from Dubai"}],
    "stream": true
  }'

Chat completions & streamingServer-sent events, token by token, exactly where your client expects them.
Tool calling & structured outputFunction calls and JSON schemas for agents and back-office automation.
Arabic-first embeddingsMultilingual retrieval for Arabic, English and code-switched text.
Batch APILower-cost async jobs for evaluations, classification and back-fills.

Works withOpenAI SDKLangChainLlamaIndexVercel AI SDKLiteLLMn8n

02Open models

Open weights, served close to home.

Leading open models from around the world — and from right here in the UAE. Switch models with a string, not a migration.

Falcon H1 34B
UAE-built

TII · Abu Dhabi
falcon-h1-34b-instruct
chatarabiclong context
Jais 30B
UAE-built

Inception · MBZUAI
jais-30b-chat
chatarabic
K2 Think
UAE-built

MBZUAI · G42
k2-think
reasoningmath
Qwen3 235B A22B

Alibaba Qwen
qwen3-235b-a22b
reasoningmultilingual
DeepSeek V3.1

DeepSeek
deepseek-v3.1
reasoningcode
gpt-oss-120b

OpenAI
gpt-oss-120b
reasoningtools
Llama 3.3 70B

Meta
llama-3.3-70b-instruct
chattools
Qwen3 Coder 480B

Alibaba Qwen
qwen3-coder-480b-a35b
codeagents
Kimi K2

Moonshot AI
kimi-k2-instruct
agentscode
Gemma 3 27B

Google
gemma-3-27b-it
visionmultilingual
Mistral Small 3.2

Mistral AI
mistral-small-3.2-24b
visionchat
Qwen3 Embedding 8B

Alibaba Qwen
qwen3-embedding-8b
embeddingsmultilingual

Need a model that isn't listed, or your own fine-tuned weights? We'll bring it up on dedicated capacity →

03Why Rihstal AI

Fast because it's near. Private because it stays.

Milliseconds matter.

Your users are in Dubai, Riyadh, Doha and Muscat. Your model should be too. Serving from the UAE removes the long haul to Europe or the US before a single token is generated — and keeps every byte in-country.

Network round-trip from Dubai

Rihstal AI · UAE~4 ms
Mumbai~38 ms
Frankfurt~118 ms
N. Virginia~215 ms

Illustrative typical round-trip times, before any model compute. Actual latency varies by network and workload.

Priced per token

Pay for the tokens you use — no GPU reservations, no minimums. Continuous batching, prompt caching and right-sized hardware keep unit costs low, and we pass that on.
Low latency, in region

GPUs in the UAE mean fewer milliseconds on the wire and a faster first token for users across the Gulf. Built for chat, voice and agent loops where every hop counts.
Private by default

Zero data retention. Prompts and completions are processed in memory and are never stored, logged for review, or used to train anything.
Regional control

Your data is processed and stays in the UAE — designed around residency expectations in government, financial services and healthcare across the GCC.
No GPUs to operate

We run the fleet — drivers, scheduling, autoscaling, failover and model upgrades. You call an endpoint and ship your product.
Dedicated when you need it

Reserve isolated capacity for steady workloads, serve your fine-tuned weights, or deploy the same stack inside your own UAE cloud.

04The name

Rihstal AI /riːħ·stæl eɪ·aɪ/ n.

From rīḥ (ريح), Arabic for wind, and crystal. Movement given structure.

Wind is pure movement — restless, unstructured, everywhere. A crystal is matter that has found its order. Inference is both: a constant stream of requests that has to be shaped into something precise, reliable and fast. We give it that structure, close to home.

Build on open models,
served from the UAE.

Get an API key in minutes, or talk to us about dedicated capacity and on-prem deployments.

Get API key Talk to an engineer

or email hello@rihstal.ai

Open-model inference, hosted in the UAE.

Keep your SDK. Change one line.

Open weights, served close to home.

Falcon H1 34B

Jais 30B

K2 Think

Qwen3 235B A22B

DeepSeek V3.1

gpt-oss-120b

Llama 3.3 70B

Qwen3 Coder 480B

Kimi K2

Gemma 3 27B

Mistral Small 3.2

Qwen3 Embedding 8B

Fast because it's near. Private because it stays.

Milliseconds matter.

Priced per token

Low latency, in region

Private by default

Regional control

No GPUs to operate

Dedicated when you need it

Build on open models, served from the UAE.

Build on open models,
served from the UAE.