Launching the NAWA API Platform

Launching the NAWA API Platform

We built NAWA because Arabic creators deserve tools that understand how their audience actually speaks. Not how textbooks say they should.

Today we are making that intelligence available to every developer. The NAWA API is a self-serve platform for Arabic language classification, dialect detection, translation, content moderation, and reply generation.

What we shipped

On March 16, we launched v1.0 with 8 core endpoints. Two days later, we added translate, detect, and moderate. By March 18, automatic Claude fallback was live for all endpoints.

Here is what the platform does today:

Classify comments in one API call

Send any Arabic comment. Get back intent (praise, question, complaint, suggestion), sentiment, dialect (Gulf, Egyptian, Levantine, MSA), and toxicity score. One call. Six credits.

Detect dialect in under 100ms

Lightweight detection using NAGL modules. No AI model call required. Two credits. Returns dialect, confidence, script direction, and language code. This is the fastest Arabic dialect detection API available.

Translate between dialects

Convert between English and any Arabic dialect with tone control. Five credits per call.

Moderate Arabic content

Catches profanity, hate speech, and slang that Google and OpenAI miss. Seven moderation categories with severity scores and flagged terms. Four credits per call.

Generate culturally-aware replies

Automatic dialect-matched replies. If a Gulf Arabic user comments, the reply comes back in Gulf Arabic. Not generic MSA that sounds like a government press release.

Why this matters

Every Arabic NLP API we tested treats Arabic as one language. It is not. Gulf Arabic and Egyptian Arabic are as different as Spanish and Portuguese. A comment that says "وايد حلو" (Gulf: "very nice") gets misclassified as negative by most Western APIs because the vocabulary does not match their training data.

We trained on real social media comments. Gulf, Egyptian, Levantine, MSA. The result: 100% dialect accuracy across a 100-comment test corpus spanning all four dialect groups.

The numbers

Powered by ALLaM

Our NAGL pipeline (NAWA Agentic Gateway Layer) routes Arabic text to ALLaM and English to Claude, with automatic fallback between providers. If ALLaM is unavailable, requests seamlessly fall back to Claude Haiku 4.5 with no user action required.

SDKs and developer experience

TypeScript and Python SDKs are available now. Both include type-safe clients, auto-pagination, retry logic with exponential backoff, and webhook signature verification.

Interactive documentation lives at developers.trynawa.com with full OpenAPI 3.1 spec, code examples in three languages, and a live API playground.

Pricing

Credits never expire. Semantic cache hits are free. If you classify the same comment twice, the second call costs nothing.

Get started

Go to developers.trynawa.com, sign in with your NAWA account (Google or magic link), generate an API key, and make your first classification. Three minutes, no credit card.

We built this for the Arabic-speaking internet. 400 million speakers. One API that actually understands them.

Get Your API Key | Read the Docs | View Pricing