Two years ago, useful AI for documents meant sending them to the cloud. In 2026, capable open-weight models run on an ordinary laptop, and some PDF tools let you plug in a model running on your own machine. That gives you a real choice. Which one fits your documents?
What "local AI" means
A local model runs entirely on your computer. Tools such as Ollama and LM Studio make it easy to download an open-weight model (from families such as Llama, Qwen, Gemma, Mistral or gpt-oss) and run it offline. Your document never leaves the device.
Cloud AI means services like ChatGPT, Claude, Gemini or Copilot, where your document is processed on the provider's servers by much larger models.
Where local AI is good enough
- Summarising reports, articles and meeting documents.
- Finding information: "What is the notice period?", "When does this policy expire?"
- Rewriting a paragraph from a PDF in plainer language.
- Classifying documents into folders.
- Working offline, on planes or in secure environments.
For these, a model in the 4 to 14 billion parameter range on a laptop with 16 GB or more of memory performs well.
Where cloud AI is still better
- Very long documents. Large cloud models handle much longer contexts accurately.
- Complex reasoning across many pages, such as comparing a contract with three amendments.
- Tables and charts. Large multimodal models read page images far better than small local ones.
- Scanned PDFs where OCR quality varies.
- Speed on modest hardware. A local model on an older laptop can be slow.
The privacy difference
With local AI, the only copy of your document is the one on your device. That matters for medical records, legal files, HR documents and anything under a confidentiality agreement. Cloud providers' business plans offer strong protections, but data still leaves your control and is subject to their terms and jurisdiction. See risks of using AI on confidential PDFs.
A practical split
| Task | Suggested |
|---|---|
| Summarise a public report | Either |
| Query a confidential contract | Local, or approved enterprise cloud |
| Extract a complex table | Cloud (or convert the table first) |
| Analyse hundreds of pages | Cloud |
| Work offline | Local |
Getting started with local AI
- Install Ollama or LM Studio.
- Download a model sized for your hardware.
- Extract the PDF's text first (or use a tool that does it for you). Clean text matters more for small models. See preparing long PDFs for AI chat.
- Ask focused questions about specific sections rather than the whole document.
Some PDF editors now support bring-your-own-model setups. See Foxit's local AI PDF editor.
Do you need AI at all?
For signing, filling forms, merging, splitting, compressing and redacting, you do not. These are exact operations best done with ordinary tools. A browser editor such as Docento.app does them on your device without any model involved.
Takeaway
Use local AI for private documents and everyday summaries, cloud AI for long, complex or scanned material that is safe to share, and no AI for exact editing tasks. For more, see AI assistants in PDF readers: privacy settings.
