Portkey open-sources AI gateway handling 1 trillion tokens daily

Wednesday 25 March 2026 - 18:30

Portkey, a San Francisco-based AI infrastructure firm, open-sourced its unified Gateway on March 24. The control plane targets the emerging agentic AI era, shifting paid SaaS features like governance, observability, authentication, and cost controls to open source.

The gateway already scales massively. It processes over 1 trillion tokens and 120 million AI requests daily, manages $180 million in annualized AI spend, and serves more than 24,000 global organizations.

The open-source package adds to its core routing engine. New features include usage policies for model access rules and limits, a continuously updated model catalog across providers, real-time cost and latency metrics, and enterprise-grade authentication supporting OAuth 2.1 and 2.0.

Portkey also open-sourced its MCP gateway for overseeing AI agents interacting with enterprise tools and external systems via Model Context Protocol. CEO Rohit Agarwal said, "Once agents access tools, query systems, and execute actions, they become operational actors in the enterprise. Companies need a control plane like for any critical infrastructure: governing access, enforcing policies, and providing real-time visibility."

Growth accelerated recently. Token volume jumped from 1 billion daily a year ago to 1 trillion now, with 30-40% week-over-week traffic gains. Portkey raised $15 million in February Series A led by Elevation Capital with Lightspeed participation to expand its platform for next-gen AI infrastructure.

By open-sourcing, Portkey aims to make its control plane a staple AI layer, akin to Cloudflare at the network edge. It connects to 1,600+ large language models and providers, offering a unified API for seamless switching between OpenAI, Anthropic, Google, and AWS without app rewrites.