The llama.cpp project released version b11699 on 2026‑10‑16. It updates the common HTTP client to read the standard proxy environment variables. Downloads via the -hf flag and internal health checks now go through a configured proxy, or skip it for loopback addresses. Malformed proxy URLs are ignored, and a warning is printed. Documentation about proxy usage was moved to the server guide.
Why it matters
Users working behind corporate proxies can now download models and run llama.cpp without the previous “Loading model…” hang.
