What is litellm-relay?

LiteLLM Relay is a company IT tool installed on employee laptops through a device management system. It gives a company two things at once: an easy way to set up AI coding tools for developers, and a way to catch AI usage that IT does not already know about. On the setup side, when Relay is installed it scans the machine for AI tools it recognizes, currently Claude Desktop, Claude Code, and Codex in its CLI, VS Code, and macOS app forms, and automatically points each one at the company's LiteLLM AI Gateway. Developers do not need their own provider API key or any manual configuration. They just open the tool and sign in with their normal company login. Relay also rechecks the machine on a schedule, so if a developer installs one of these tools later, it gets configured automatically without anyone running Relay again by hand. On the monitoring side, Relay also detects AI traffic from tools that were not officially sanctioned, such as Notion AI or Perplexity, sometimes called shadow AI, and routes that traffic through the same Gateway. That gives the company a single place to see every request, response, and usage event from AI tools across the organization, rather than a blind spot for tools nobody signed off on. There is also an optional menu bar app for macOS that lets individual developers see their own AI spending locally on their machine, broken down by tool, without needing to log into a separate dashboard. It shows things like monthly spend, daily spend, which models are being used, and how often responses are served from cache. Relay says it does not log cookies or authorization headers, and that captured payloads are truncated with sensitive headers removed. It currently supports macOS, works with common device management systems like Jamf, Microsoft Intune, and Kandji, and is installed by building it from source in Rust, since it has not yet published a tagged release.