OpenRelay provides hosted AI model inference and dedicated GPU computing through a unified platform and API. Its inference delivery network automatically routes requests among providers based on latency, cost, availability, and model performance, with failover and load balancing designed to help manage traffic spikes and provider outages. Hosted models run behind a single endpoint, scale with demand, and are billed according to usage.

For teams needing more control, OpenRelay offers dedicated GPU virtual machines with SSH access, customizable runtimes and images, and persistent volumes. Compute is metered by the minute, with no egress fees or commitments. The platform also supports batch jobs for workloads that can run at lower cost than interactive traffic. Customers can manage resources through the dashboard, API, or orl command-line tool; an MCP server lets compatible agents operate the platform with user-granted access.

OpenRelay lists GPU options including RTX 4090, RTX 5090, A100, and H100, alongside inference models from multiple providers.