Egypt has opened Karnak, a state-backed artificial intelligence chatbot, to public testing.
The communications ministry said on August 23 that Karnak Chat was available through a web service and Android and iOS apps. The assistant was developed by engineers at the ministry-affiliated Applied Innovation Center and supports Modern Standard Arabic and colloquial Egyptian, according to the ministry.
The announcement also directs developers to a public Karnak model repository. The 40bn-parameter version identifies Alibaba’s Qwen3-30B-A3B-Instruct-2507 as its base. The ministry has not said whether it powers the chatbot or identified the model that does.
AIC says it continued pre-training, applied supervised fine-tuning and built an Arabic-optimised tokenizer. Public configuration files show 64 layers, compared with 48 in the base model, while retaining Qwen3’s mixture-of-experts architecture. The released materials do not include the full training data or an independent evaluation.
Users can download Karnak-40B’s weights under an Apache 2.0 licence and deploy or fine-tune the model, according to its model card. That gives organisations more control than a closed overseas API, but is not evidence of an entirely domestic technology stack.
Egypt’s 2025–2030 AI strategy calls both for developing an indigenous LLM and for selecting local or global models to fine-tune with local data. It leaves room for Karnak’s hybrid approach: a foreign base model adapted locally.
The February launch positioned Karnak as a base for government, business and education applications. The August pilot is a separate step, making a general-purpose conversational interface available to the public.
MCIT says software and applications built on Karnak can run securely in local data centres “when required”. The announcement does not say where the public service is hosted or whether users’ prompts and responses remain inside Egypt, leaving the extent of Cairo’s control over the live service unresolved.
