Unified proxy routing Claude Code between claude.ai and DeepSeek via LiteLLM.

A small local proxy that lets Claude Code talk to either the claude.ai subscription or a DeepSeek API key, without touching how the CLI or the VS Code extension actually work.
It's the piece underneath Relay — a menu-bar toggle for switching providers on the fly. LiteLLM runs in its own isolated Python venv and exposes an Anthropic-compatible /v1/messages endpoint (plus /v1/messages/count_tokens, which the VS Code extension calls separately and needs serving too), so from Claude Code's point of view nothing about the wire protocol changes — only the base URL and auth token do.

The DeepSeek API key lives in the login Keychain, never on disk, and the LiteLLM master key is generated fresh on first launch with SecRandomCopyBytes. Revert is instant: unset the env vars and the CLI falls straight back to the claude.ai subscription, no re-auth needed. Built as the backend for a personal tool, not a hosted service.