Llmproxy

Like LiteLLM, but it gets to the point. One Go binary.

A self-hosted, OpenAI-compatible LLM proxy: one static Go binary with an embedded web UI. Point any OpenAI SDK at it with one line changed and it routes to your vLLM, SGLang, OpenAI or other OpenAI-compatible upstreams, with per-user API keys, a curated model catalog and usage accounting.

No prompt, completion or other request/response content is ever persisted. The database schema has no column that could hold a body, and tests enforce it.