PayPal has developed and deployed the Cosmos LLM Gateway, a policy-driven system designed to manage large-scale enterprise AI traffic across multiple vendors. This gateway, built on Envoy and Rust WebAssembly, handles tasks like authentication, credential brokering, and policy evaluation, while also managing token metering and prompt-caching. It allows for declarative governance through Open Policy Agent, enabling features such as role-scoped authoring, approvals, and shadow-mode rollouts. The system also includes a data-loss-prevention layer and spend scoping for budget management. As of September 2026, the gateway supports over 12,000 developers and hundreds of business use cases, processing approximately 200 billion tokens daily across more than 100 models with high availability. AI
IMPACT This system demonstrates a scalable approach to managing diverse LLM vendor integrations, crucial for enterprises adopting multi-model strategies.
RANK_REASON The item describes the design and operation of a specific system (Cosmos LLM Gateway) for managing LLM traffic, presented as a paper. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →