Why we do not use AI to process email — and what it means for your privacy
No language model reads your mail in eml-pdf.com: a deterministic parser follows fixed rules. How the processing works and what that does — and does not — give you for privacy.
What we mean by AI — and what is not here
By AI people usually mean large language models: they read text, fill in what is missing and phrase answers by probability. eml-pdf.com does not use them.
- A parser reads the MIME structure, headers, body and attachments exactly as they are written — no guesses and no understanding of meaning.
How a message becomes a PDF
The pipeline is linear: parse the message, sanitize HTML, disable scripts, download remote images and block the remaining remote resources, lay out PDF pages, package attachments and the index.
- The same input produces the same output — it can be recomputed and reconciled with checksums.
- Nothing is improvised on the fly: what you receive is fully explained by code, not by a model's mood.
What this means for privacy
Message contents are not passed to third-party models and your data is not used for training.
- Files are deleted after the archive is packaged successfully; access to the result ends an hour after it is ready or sooner when you choose Delete.
- Files travel over HTTPS and default AES256 encryption is confirmed in the live object storage.
- This is not an audit of the whole infrastructure or a guarantee that any correspondence is safe: the Russian and English versions share one infrastructure, and we do not promise a processing country yet — see the Data handling page.
The limits of deterministic parsing
And the honest flip side: code does not guess.
- A non-standard, damaged or oddly encoded message may fail — visibly, in errors.csv, rather than being hidden.
- Format support does not guarantee correct parsing of every message.
- That is exactly why the archive ships an index and a report: they show what worked and what did not.
Convenience without a neural net
The CSV and XLSX index, the chronological merged PDF, page numbering, checksums, the free daily allowance and credits — all of it is plain code behaving predictably.
- If you need summaries, translations or classification, do them yourself on your own copies in your own tools: that keeps the contents under your control.
Who cares about this
Legal teams, because correspondence becomes something people examine.
- Finance, because exports end up in audits.
- IT, because processing rules must be predictable.
- And an international audience: the rules are the same in Russian and English, and the service boundaries are written plainly on the Data handling and Terms pages.
Check it yourself
Start with the five free messages a day: upload a small batch, review the preflight inventory, take the archive and inspect the index.
- Questions about data handling belong on the security page, questions about features in the FAQ, personal ones to support.