As enterprises scramble to deploy AI, they’re creating a Wild West of unmanaged API keys, spiraling costs, and glaring security holes. It’s a chaotic new frontier, and London-based startup Requesty just raised $3 million in a seed round led by 20VC to bring some order to it. The company is building what it calls an LLM gateway, a unified control plane for the messy business of talking to large language models.
The problem Requesty is tackling is becoming painfully familiar. A developer leaves the company, but their OpenAI API key with access to sensitive data lives on. A single provider like Anthropic has an outage, and suddenly every AI-powered feature in your application breaks. Costs balloon without anyone noticing. Requesty’s platform sits between a company’s developers and the AI models they use, acting as an intelligent traffic cop for every single API request.
The Cloudflare for AI
Requesty’s founders, Thibault Jaigu and Daniel Trugman, are pitching their LLM gateway as the “Cloudflare for AI,” an essential infrastructure layer for reliability and security. The platform allows companies to set spending limits, enforce security policies (like blocking sensitive documents from being uploaded), and manage model access from a central dashboard.
One of its cleverest features is prompt caching. By storing and reusing the answers to frequent requests, Requesty claims it can significantly cut both latency and costs, with some customers reporting savings of up to 40 percent. This combination of routing, governance, and optimization is clearly resonating. Since pivoting from data analytics earlier this year, the company says it has onboarded over 25,000 developers and is generating more than $1.5 million in annual recurring revenue.
With fresh capital, Requesty plans to grow its small team and double down on its mission to provide the foundational plumbing for enterprise AI. It’s a crucial, if unglamorous, part of the stack that could determine whether corporate AI initiatives succeed securely or collapse under their own weight.
