▸ Sovereign Mesh · Live model fabric

One sovereign mesh.
Every model, on demand.

The Kawkav mesh is a pooled, private inference fabric spread across EU and Gulf regions. It powers every Kawkav AI product — agents, voice, sovereign models, intelligence — through one gateway, with no dependence on US hyperscalers.

Mesh online · serving live traffic
EU + Gulf sovereign regions · one routed gateway · capacity scales on demand
12
Models online
4
Sovereign regions
7
Capabilities
2.4B
Tokens served
updated continuously · snapshot · token counter is illustrative
Model catalogue · by capability

Every capability, one fabric

Models are grouped by what they do, not where they run. Pick a capability; the gateway routes to live capacity in the nearest sovereign region.

Chat · compact

Live

Fast, low-cost conversational models for everyday assistants, drafting, and high-volume automation.

7B / 8B class · low latency

Chat · balanced

Live

Mid-size models that balance reasoning quality with speed — the default for most agent workloads.

14B class · general purpose

Flagship reasoning

Live

Large models for hard reasoning, long context, and complex multi-step tool use where quality matters most.

Large · deep reasoning

Arabic-native

Live

Models trained for Arabic-first dialogue and Gulf-region use cases — fluent MSA and regional nuance.

Arabic + English · MENA

Vision-language

Live

Multimodal models that read images, documents, and manuscripts — OCR/HTR, charts, and visual Q&A.

32B class · image + text

Embeddings

Available

Dense vector embeddings for semantic search and retrieval-augmented generation across private corpora.

Multilingual · RAG

Speech · ASR/TTS

Live

Speech-to-text and text-to-speech for real-time voice agents and call handling in multiple languages.

EN / ES / AR · real-time

Full-duplex voice

Live

End-to-end conversational voice that listens and speaks at the same time — natural, interruptible dialogue.

Real-time · low-latency

Image & video

Available

Generative image and short-form video capacity for media and creative workloads, spun up on demand.

Diffusion · on demand
How the mesh works

Pooled, private, sovereign

A single fabric that pools capacity, keeps inference private, and scales elastically — without ever leaving sovereign ground.

Pooled private inference

Capacity from across the fabric is pooled into one private engine. Your prompts and data are never used to train shared models and never leave the mesh.

EU + Gulf sovereign regions

Inference runs in EU and Gulf regions under our own control — EU data residency by default, with no dependence on US hyperscalers.

One gateway, one API

Every product talks to a single routed gateway. Pick a capability, not a provider — the gateway selects a live model and the nearest region for you.

Scales on demand

Heavier capabilities — flagship reasoning, vision, image and video — spin up capacity when demand arrives and release it when idle, so you pay for what you use.

What the mesh powers

From capability to product

The same fabric powers everything Kawkav builds. Here is how mesh capabilities map to the products you can use today.

Chat · reasoning · function callingCompact, balanced and flagship models with tool use
AI Agents
Speech ASR/TTS · full-duplex voiceReal-time, multilingual conversational voice
Voice / PBX
Arabic-native · private single-tenantSovereign models on customer-controlled terms
Sovereign Models
Vision-language · embeddings · RAGDocument, image and signal understanding at scale
Media Intelligence

Build on the sovereign mesh

Try the models in the playground, talk to a live voice agent, or wire the mesh into your own product through one gateway.

Try the playground → Try voice AI → Build on the mesh →