Portraitor vs WhatsAnalyze: which one leaves your device
WhatsAnalyze reads your chat in the browser and sends nothing. Portraitor sends masked text to a model and keeps none of it. What each trade buys you.
WhatsAnalyze is more private than Portraitor, and says so accurately: it reads your chat entirely in your browser, sends nothing to a server, and publishes its source code so you can check. It returns counts and charts. Portraitor returns a reading of one person, which cannot be done in a browser, so masked text has to leave your device. Choose by which of those two things you want.
Most comparisons in this category are written to make the writer win. This one cannot be, because on the single question people ask first, where does my conversation go, WhatsAnalyze has the better answer. What follows is why that is true, what it costs you, and when it stops being the right trade. Claims about WhatsAnalyze were read on its own live site on 8 October 2026; claims about Portraitor come from its home page and privacy policy, read the same day.
What each one produces
WhatsAnalyze’s home page describes dropping in a chat export and seeing who talks most, when you write and what you write about. It lists messages per hour, share of speech, word clouds and fun facts, and a PDF export that puts the statistics up front with every message behind them. These are counts of things that are already in the file.
Portraitor’s privacy policy describes something else: inferences about one named participant, covering personality traits, attachment and conflict style, emotional regulation, motivations and communication style, each carrying an estimate of confidence. These are not counts. They are judgements produced by a language model reading the text.
That distinction decides everything else on this page. Counting can happen on your own machine. Reading, as things stand, cannot.
Where your chat goes, in each case
WhatsAnalyze states on its home page that everything is read in your browser and that nothing is uploaded, and its own FAQ answers the question directly: all chat processing occurs locally in your browser, no data is ever sent to a server, and you can inspect the open source code on GitHub to verify. That last clause is the part that matters. A privacy claim you can check against published source is a different kind of claim from one you have to take on trust.
Portraitor cannot make that claim and does not. Its policy describes masking first: names, email addresses, telephone numbers, postal addresses and credentials are removed on your device before anything is transmitted, and the list matching each placeholder back to a real detail never leaves your device. Photographs, videos and voice recordings are never opened and never transmitted. Only the masked written text is sent, and it goes to a third party artificial intelligence provider on that provider’s paid service tier, with three limits on what the provider may do with it stated expressly in the policy. The conversation is not stored, is held only for as long as generation requires, and is never written to a database, to backups or to logs.
So the honest summary is that one tool sends nothing and the other sends masked text to a model and keeps none of it. Those are not the same promise, and if sending nothing is your requirement, the comparison is already over.
A second honest point, since this page is about privacy. The Portraitor website and this blog run Google Analytics, which sets cookies, and the privacy policy says so and offers an opt out. A tool that does its work in your browser can make a cleaner claim about tracking than Portraitor currently can.
Why Portraitor cannot simply do it locally too
The obvious question is why the masking can happen on your device but the analysis cannot. The answer is size. Masking is pattern work and ships as a filter; the policy describes the iPhone app downloading that filter once, at about 175 MB, and masking on the phone. A model capable of producing the inferences described above is orders of magnitude larger than that, and does not run in a browser tab on a phone.
That is the real trade on this page. The local tool is local because counting is small. The remote tool is remote because reading is large. Anyone promising both a language model reading of your conversation and that nothing leaves your device is worth reading very carefully.
Where WhatsAnalyze is the better choice
If you want to know who sends more messages, what hours you both write, which words one of you uses and the other does not, or you want a printable record of the conversation, it does that and nothing has to leave your browser to get it.
If your requirement is absolute, that no part of the conversation is transmitted anywhere under any circumstances, it meets that requirement and Portraitor does not.
If you want to verify the claim rather than believe it, its source is published and you can read it. Portraitor’s masking happens in your browser too, but its policy is a description of what the code does rather than the code itself.
And it is free to use from the home page, with no file leaving your machine and no account step before you see output.
Whether Portraitor is worth it for you
Buy it only if the thing you want is a reading of one person rather than a count of messages, and you accept that masked text reaches a model to produce it. Portraitor’s home page lists $10 for a single portrait with no account and no renewal, and $19 a month for ten that refill and can be shared from one pool.
Do not buy it if counts answer your question, if your rule is that nothing may be transmitted, or if you want to audit the implementation rather than read a policy about it.
If you want the full data path written out before deciding, whether it is safe to upload your chat history sets it out, and what a chat analysis cannot tell you about someone covers the limits of the output itself. For a comparison against a tool that does keep what you upload, see Portraitor against Mosaic. The file formats each tool accepts are in which chat export files can be analysed, and Portraitor shows the flow itself.
Frequently asked questions
Frequently asked questions
Does WhatsAnalyze upload my chat?
Its home page states that everything is read in your browser and nothing is uploaded, and its FAQ states that all chat processing occurs locally, that no data is ever sent to a server, and that the source code is published on GitHub so the claim can be checked.
Is WhatsAnalyze more private than Portraitor?
On transmission, yes. WhatsAnalyze sends nothing. Portraitor sends masked text to an analysis provider after removing identifiers on your device. Portraitor keeps no conversation and no account, but it does transmit, and its site runs Google Analytics.
Why can Portraitor not run in my browser?
Masking is small enough to ship as a filter; the policy describes the iPhone app downloading one of about 175 MB. The model that produces the inferences is far larger than that and does not run in a browser tab.
Do they do the same thing?
No. WhatsAnalyze returns statistics drawn from the file, such as message counts, timing and word frequency. Portraitor returns inferences about one named participant, each with an estimate of confidence.
Can I use both?
Yes, and for many people that is the sensible order. Run the free local tool first to see the shape of the conversation, and pay only if the question you are left with is about a person rather than a count.
One of these never lets your conversation leave the room. The other tells you something a count cannot. Decide which question you came with, and be suspicious of anything claiming both.
If a count is not the question
Portraitor reads one named participant, masks identifiers on your device first, and keeps no conversation.
See how Portraitor works →