ChatGPT Library is free now. Your files train the model unless you turn it off.
OpenAI's File Library went free in May 2026, which means anyone can upload a will, a tax slip, or a medical record and ask questions about it. By default, those files improve OpenAI's models. Archevi was built the other way around: your documents never train any model, every family member sees the same vault, and your files live in Canada.
May 14, 2026: ChatGPT File Library expanded to free users
ChatGPT's File Library (previously a paid-tier feature) is now available on the free plan. Anyone can upload family documents and ask GPT questions about them. The setting that controls whether your files are used to train OpenAI's models is buried under "Data Controls" and defaults to ON for free and most consumer accounts.
Archevi documents never train any model. It's not a setting you have to find. It's the architecture.
What you actually pay, and what you pay with
Per user, per month. Free tier exists but throttles file uploads and reasoning. Five family members = $100/month.
- File Library for personal documents, web search, reasoning
- No shared family workspace. Each member uploads separately
- Files train OpenAI models by default unless you opt out
- Hosting in the US. Uploaded files fall under US law
Per family, per month. One vault. Every member gets full AI. Files never train any model.
- Shared family vault. Upload once, every member can ask
- Files never enter model training. By architecture, not by setting
- Hosted in Canada (Toronto), PIPEDA-aligned
- PII redacted before any query touches the model
What Archevi does that ChatGPT Library doesn't
Files never train the model
Your uploaded documents never enter any model training pipeline. This isn't a privacy setting. It's how the system is built. ChatGPT's Library trains on user files by default; opt-out lives in Data Controls.
One shared family vault
Every family member sees the same documents and asks the same questions. ChatGPT's Library is tied to your personal OpenAI account. Your spouse and parents each upload separately into their own accounts.
Canadian hosting (PIPEDA)
Your files live in Toronto on PIPEDA-aligned infrastructure. OpenAI's File Library is hosted on shared US infrastructure. Subject to US laws including the CLOUD Act.
PII redacted before the model sees it
Names, SINs, addresses, and account numbers are stripped before any query touches the language model, then restored in your answer. ChatGPT sees your raw document. Names, SIN, everything.
Feature-by-feature comparison
| Feature | Archevi | ChatGPT |
|---|---|---|
Files never used for AI training By architecture, no training pipeline Opt-out exists but defaults to ON for free + most consumer plans | ||
Shared family vault, one upload, all members can ask File Library tied to individual OpenAI account | ||
PII redaction before queries reach the model Names, SINs, account numbers stripped automatically | ||
Canadian data residency (PIPEDA) DigitalOcean Toronto US-based infrastructure | ||
Per-document Q&A with cited passages | ||
Expiry tracking for passports, insurance, licences | ||
General AI: image generation, code, voice mode What ChatGPT is built for | ||
Flat family pricing (no per-seat) $6 to $24/month for the whole household $20/user. 5 members = $100/month |
The defaults matter
ChatGPT Library is a genuinely useful feature. Free users got it on May 14, 2026, and for casual document Q&A (a recipe, a contract you're trying to understand, a manual you can't find) it's a real upgrade over what was available before.
The problem with using it as a family document vault is the defaults. By default, uploaded files improve OpenAI's models. By default, your spouse can't see your Library. By default, your files live in the US. None of those defaults are hidden, but they're also not what most families would choose if they were asked explicitly.
OpenAI's privacy posture is also a moving target. The policies that govern your File Library today aren't the policies that governed your File Library a year ago. Free-tier features get more usage telemetry than paid tiers. Settings that exist today have been removed or changed in past versions.
Archevi flips every one of those defaults. Files never train any model. There's no setting because there's no training pipeline. Family members share one vault by design. Documents are stored in Canada under PIPEDA. PII is redacted before any query reaches the language model. These aren't features you turn on. They're how the product is built.
Use ChatGPT for the documents you'd happily paste into a search bar. Use Archevi for the documents you'd never paste into a search bar.
What ChatGPT does better, and we'll say so
ChatGPT is a general-purpose AI tool with capabilities that go far beyond document Q&A. Code generation, image creation, web search, voice mode, custom GPTs. None of that is matched by Archevi, and we don't try. For your general AI workflow, ChatGPT Plus or even the free tier is a good tool.
OpenAI also has the deepest model bench in the industry. GPT models reach Plus subscribers within days. Reasoning modes, code interpreters, and connectors evolve constantly. A small team like Archevi will never match that pace of general-AI improvement.
Archevi isn't trying to replace ChatGPT. It's the version you reach for when the files are wills, policies, tax slips, and medical records. Files you'd never want trained on, files your whole family shares, and files you want hosted under Canadian law.
The vault where defaults work in your favour.
Files never train any model. Family members share one vault, hosted in Canada. Start on the Core plan for $6 CAD/month. Every member gets full AI from day one.