We just released two open projects from ZAQERIS Ai, an independent research group in Cairo, Egypt.
ZAQERIS-7B
A 7B-class multilingual model for coding, reasoning, and tool use, trained from scratch.
- Released in 4-bit NF4 (double-quant), loads in about 4.99 GiB and runs on an 8 GB GPU
- Built-in
<search>tool use, and it declines to guess when search is unavailable - Apache 2.0
Model: https://huggingface.co/ZAQERIS/ZAQERIS-7B
QERLIO
A local-first AI inference runtime.
- Privacy-first and offline by default, with no telemetry or prompt logging without explicit opt-in
- One local server, three APIs: OpenAI, Anthropic, and Ollama
- Adaptive execution on CPU, GPU, and NPU, with automatic CPU fallback
- Phase 0 complete, Phase 1 scaffolded
- Apache 2.0
Code: https://github.com/ZAQERIS/QERLIO
Our other public models, AXIOM and ZQN-12B, are at https://huggingface.co/ZAQERIS. Feedback and issues are welcome.
Top comments (0)