A proxy on Janitor AI is a setting that sends your chats to an outside model, such as DeepSeek, instead of the built-in JLLM.
It is not a VPN: you enter a model name, a proxy URL and an API key, and the key, URL and model must come from the same provider.
The default JLLM is free with about 8,000 to 9,000 tokens of memory; a proxy costs per token but offers any compatible model.
At an assumed 8,000 input and 300 output tokens, a message costs about $0.95 per 1,000 on GPT-6 Luna and $38 on Claude Opus 5.5.
Use only trusted endpoints and a dedicated, low-balance key; most errors come from a wrong URL, model name or missing refresh.