How to Use Ollama During a Proctored Exam on Mac

Updated 2nd of July 2026 · 3 min read

offline AI · setup

Ollama runs LLMs entirely on your Mac with zero network egress. For proctored exams that block AI endpoints or where you do not want any prompt sent off-device, Ollama plus the LDBypass overlay is the only fully-private path.

The setup takes about 5 minutes total: download Ollama, pull a model, start the server. Mistral 7B fits on 8 GB Macs and gives GPT-3.5-class quality. Llama 3.1 8B is slightly stronger and fits comfortably on 16 GB. Once the model is loaded, your prompts stay on the Mac. The LDBypass overlay points at localhost; no DNS, no TLS, no outbound traffic. This is the configuration to use when the proctor app blocks chat.openai.com and friends, or when your institution has policy concerns about cloud AI.

Key points

1. Install Ollama from ollama.com (one click).
2. `ollama pull mistral` (or `ollama pull llama3.1:8b`).
3. `ollama serve` runs the local API at localhost:11434.
4. LDBypass: configure Ollama mode or URL http://localhost:11434/.
5. Pre-warm the model before the exam by running one prompt; first request after a cold start takes longer.

How it works

Pre-exam:
  $ ollama pull mistral
  $ ollama serve   # leaves running

In LDBypass:
  Settings -> AI provider -> Ollama
  Press Ctrl+Cmd+L to overlay.

During exam:
  Localhost only, no network egress.

Common questions

Will Ollama drain my battery?

Inference is GPU-heavy. Plan to be plugged in for any exam over 60 minutes. Idle Ollama (no prompt running) consumes very little.

Can the proctor detect Ollama running?

They might enumerate processes. Ollama is a known process name. If you are concerned, rename or use the underlying llama.cpp directly with a custom binary name.

Is the answer quality enough for med / law / engineering?

GPT-3.5-class is fine for routine homework. For hard problems (USMLE-style differentials, bar-exam essays), cloud frontier models still win.