Hugging Face Inference Endpoints

What it does

Access requirement

Access to the Inference Endpoints web application requires a valid payment method on the Hugging Face account or organization.checked 2 Oct 2026

API access

Endpoints can be called through the UI, cURL, the Hugging Face inference libraries, or another REST client.checked 2 Oct 2026

Autoscaling

Users can configure minimum and maximum replicas and enable scale-to-zero when inactive.checked 2 Oct 2026

Cloud providers

The configuration guide lists AWS, Microsoft Azure, and Google Cloud Platform as hosting providers.checked 2 Oct 2026

Compliance

Hugging Face states that the Hub and Inference Endpoints are SOC 2 Type 2 certified.checked 2 Oct 2026

Hub integration

Endpoints use model weights and artifacts from Hugging Face Hub repositories.checked 2 Oct 2026

Inference engines

Native engine support includes vLLM, Text Generation Inference (TGI), SGLang, llama.cpp, and Text Embeddings Inference (TEI), with custom containers also supported.checked 2 Oct 2026

Monitoring

The web application provides endpoint logs and a metrics dashboard for monitoring deployments.checked 2 Oct 2026

Notable limits

Available instance types may require a quota request, and paused endpoints do not count against used quota while scale-to-zero endpoints do.checked 2 Oct 2026

Private networking

Private Endpoints are accessible only through an intra-region AWS or Azure PrivateLink connection and are not internet-accessible.checked 2 Oct 2026

Purpose

Inference Endpoints is a managed service for deploying AI models to production on infrastructure Hugging Face manages.checked 2 Oct 2026

Security

Hugging Face says endpoint payloads and tokens are not stored, logs are retained for 30 days, and traffic is encrypted in transit with TLS/SSL.checked 2 Oct 2026

Support

The product page lists email support for Self-Serve and dedicated support and SLAs for Enterprise.checked 2 Oct 2026