Local models and large downloads

Ollama

Models are downloaded by the ollama serve process, so that process needs the proxy.

• macOS app:

launchctl setenv HTTPS_PROXY http://127.0.0.1:7898
# then quit and reopen Ollama

• Linux (systemd):

sudo systemctl edit ollama
# in the editor that opens, add:
# [Service]
# Environment="HTTPS_PROXY=http://127.0.0.1:7898"
sudo systemctl restart ollama

• Running it by hand: HTTPS_PROXY=http://127.0.0.1:7898 ollama serve

LM Studio

Settings → Proxy, enter http://127.0.0.1:7898 (menus vary by version); in PAC mode it usually works automatically.

Tips for large downloads

• Prefer tools that can resume: huggingface-cli download, aria2c, curl -C -.
• Watch your traffic: the weights of a 70B model can exceed 100 GB, so check your remaining traffic first and add a data pack if needed.