Comparisons
Run AI locally, or redact documents before cloud AI?
Running a model locally with Ollama or LM Studio keeps documents entirely on your computer, at the cost of hardware and, for most machines, weaker models than ChatGPT or Claude. Redacting first lets you keep using the strongest cloud models, but the provider still receives the redacted copy and detection can miss details. Choose local AI for maximum isolation, redaction for maximum model quality.
Last reviewed · 4 sources
What local AI gives you
Ollama says: “Ollama runs locally. We don’t see your prompts or data when you run locally,” and its local models are free. LM Studio runs open-weight models on macOS, Windows and Linux. With nothing sent anywhere, there is nothing to redact.
The trade-off is capability. Local models are limited by your computer’s memory, and the models most Macs can run comfortably are smaller than the frontier models behind ChatGPT and Claude, which can show on complex drafting and analysis.
What redacting first gives you
You keep the model you already trust for quality, on any Mac. What the provider receives is a copy with personal details replaced, which matches what regulators such as the Law Society of Alberta ask: “Never include confidential or potentially identifying information in prompts.”
The trade-offs: the redacted content still goes to the provider, detection can miss details, and the Law Society of Ontario notes that remaining facts can sometimes identify someone.
Who should choose which
Choose local AI if your documents must not leave your computer in any form, you have a Mac with plenty of memory, and the tasks are ones smaller models handle well.
Choose redact-then-cloud if you need frontier-model quality, work on an ordinary laptop, or already rely on ChatGPT, Claude or Copilot.
The two combine well: prepare a reviewed copy locally, then use it with a local model for sensitive matters and a cloud model for demanding ones.
The ranking
Ranked by these criteria, in this order:
- How completely the document stays on your computer
- Model quality available
- Hardware required
- Effort per document
1. Best when documents must never leave your computer
Ollama or LM Studio
Free local models; quality and speed depend on your hardware.
2. Best when you need frontier-model quality
PiBye with any cloud AI
Reviewed, restorable copies for ChatGPT, Claude, Gemini or Copilot; the provider receives the case with identities swapped for tokens it cannot reverse.
Where PiBye fits
How PiBye handles this
PiBye does its detection, review and restoration on your Mac and works with any AI, including a local model: paste the approved copy wherever you choose.
1.0.1 · macOS 14.8.5 or later · Apple Silicon · 1.1 GB
Frequently asked questions
Is a local model private?
When it runs locally, prompts stay on your computer. Ollama says it does not see prompts or data when you run locally; its optional cloud models are different.
Do I still need to redact for a local model?
Not for the provider’s sake. You may still want a reviewed copy if the output will be shared with others.
Sources
- Ollama, Ollama. Checked 17 September 2026.
- LM Studio, LM Studio. Checked 17 September 2026.
- The Generative AI Playbook, Law Society of Alberta, updated February 2026. Checked 17 September 2026.
- Licensee use of generative artificial intelligence (white paper), Law Society of Ontario, April 2024. Checked 17 September 2026.