The platform

One platform, four capabilities.

Text is available and voice is in private preview; vision and broader multimodal systems are planned in the Amegilla model family.

01 · Language

Arabic as it's actually spoken.

Most models learn Arabic as a translation of English. Amegilla is designed around fuṣḥā, supported dialects, and real code-switching use.

  • Saudi, Gulf, Levantine, and Egyptian dialects, not just fuṣḥā
  • Long-context reasoning and generation
  • Faithful tone - respectful, idiomatic, correct
اكتبلي رسالة شكر بالمصري“Write me a thank-you note in Egyptian.”
من كل قلبي، شكراً على لطفك ووقتك - مقدّرة كل حاجة عملتها.
02 · Voice · private preview

It hears the accent - and answers in kind.

Private-preview speech targets supported regional accents, including speech-to-text, text-to-speech, and low-latency conversation.

  • Speech-to-text in regional accents
  • Natural text-to-speech voices
  • Live conversation, low latency
…يفهم اللهجة الخليجية والمصرية والشامية
03 · Vision · coming soon

It will read the page.

Planned vision capabilities target Arabic script, scans, documents, handwriting, tables, and structured extraction.

  • Arabic OCR for scans and documents
  • Handwriting, even the messy kind
  • Tables and forms as structured output
scan
فاتورة رقم ٤٥٣ - المجموع ٨٤٠ ر.س
invoice_no: 453
total: "840 SAR"
lang: "ar"
04 · Agents · preview and roadmap

From answering to doing.

Tool-calling and retrieval workflows are being tested in preview; broader multi-step and desktop agent experiences remain on the roadmap.

  • Tool calling and function execution
  • Retrieval over your own documents
  • Multi-step workflows with guardrails
agent run · 3 steps
→search_orders(status:"late")
→draft_reply(dialect:"khaleeji")
→send_message()
Done - 4 customers updated in Gulf dialect.
The Amegilla family

Text in products; direct model access in preview.

Everything ships under a single name. Access is invitation-only while we open seats gradually.

ModelModalityBest forStatus
amegillaTextChat, generation, reasoningIn products · model preview
amegilla-voiceSpeechSpeech-to-text, text-to-speechPrivate preview
amegilla-visionVisionArabic OCR, documentsComing soon
amegilla-imageImageImage generation from Arabic promptsComing soon
amegilla-videoVideoText-to-video generationComing soon
amegilla-omniMultimodalVoice, image, and video - combinedOn the roadmap

Arabic AI platform

One Arabic AI platform for language, voice, vision, and agents

Direct answer

The B AI platform combines an Arabic large language model (Arabic LLM) with private-preview Arabic speech and voice AI. Vision, image, video, and broader multimodal capabilities are planned. It is designed for Saudi, Gulf, and wider Arabic use cases rather than translated as an afterthought.

Organizations can evaluate B AI for customer service, document intelligence, and private-preview speech applications, while planned vision and agent automation remain on the roadmap. Regional, private-cloud, and on-premise deployment options are available for qualifying requirements.

Frequently asked questions

What is an Arabic AI platform?

An Arabic AI platform provides models and tools for Arabic text, dialects, speech, vision, and automated workflows through products or developer interfaces.

Does the B AI platform include an Arabic LLM?

Yes. The platform includes the Amegilla Arabic-native model family and private-preview voice; vision and broader multimodal capabilities are planned.

Can the platform run inside Saudi Arabia?

B AI supports deployment choices designed for regional hosting, private cloud, on-premise infrastructure, and isolated enterprise environments.

The B Organization account

Start from one identity.

Create your account, then continue to The B Organization workspace that fits what you're building.