---
## About Univars AI
Univars AI operates the premier **Sovereign GPU Network**, bridging enterprise AI needs with high-performance, locally-hosted infrastructure. As an official Hugging Face Inference Provider, we offer highly optimized, zero-latency inference routing designed specifically for financial institutions, governments, and privacy-conscious enterprises.
We believe that intelligence should be sovereign. Our custom infrastructure operates seamlessly across top-tier data centers globally and bare-metal nodes across Africa (Nairobi, Lagos, Johannesburg).
## Hugging Face Inference Provider
We provide **serverless and dedicated inference endpoints** directly integrated into the Hugging Face ecosystem.
* **Zero Token Tax:** Pay only for raw compute. No platform markups on open-weights inference.
* **Sovereign Data Residency:** Choose your execution region to comply with local data protection laws (GDPR, CBK, NDPR).
* **Ultra-Low Latency:** Optimized vLLM and TGI routing for sub-second Time To First Token (TTFT).
* **Multi-Currency Billing:** Settle inference compute natively in USD, KES, NGN, and EURC via ShujaaPay integration.
---
## Our Fleet Capabilities
Through our proprietary **Jenga GPU Orchestrator**, Univars AI dynamically routes Hugging Face requests across:
* **NVIDIA H100s & A100s** for massive LLM training and batch inference.
* **NVIDIA L4s & T4s** for high-throughput, low-cost conversational inference.
* **Google TPU v5e** for specialized tensor operations.
---
## Integrating with Univars AI
Getting started with our Hugging Face endpoints is seamless. Just select **Univars AI** from the Inference Providers dropdown on any supported model card!
```python
from huggingface_hub import InferenceClient
# Use Univars AI as your Inference Provider
client = InferenceClient(
model="meta-llama/Llama-3-70b-chat-hf",
provider="univars_ai",
api_key="hf_..."
)
response = client.chat_completion(
messages=[{"role": "user", "content": "Explain sovereign AI in one sentence."}],
max_tokens=100
)
print(response.choices[0].message.content)
```
## Trust & Security
- **SOC2 Type II Compliant** (Pending)
- **End-to-End Encryption** (TLS 1.3)
- **Zero-Logging Policy** on inference payloads (Enterprise Tier)
Building the intelligence layer for the Global South and beyond.
