Comparison
How ChatMend compares. Quoted, not asserted.
Comparison tables are usually written by the winner. This one quotes every claim directly from the other product’s own privacy policy, links the source, and includes the rows where something else is the better choice.
Claims read from public pages on 18 August 2026. Policies change — follow the source links before relying on any of this. Where a product does not address something, the cell says “not stated” rather than assuming the worst.
| What we compared | ChatMend | MosaicChats | DoubleText |
|---|---|---|---|
| Your conversation after the analysis | Yes | No | Yes |
| Training models on what you upload | Yes | No | Not stated |
| Anonymised data used commercially | Yes | No | Partly |
| Where the analysis runs | Partly | No | Yes |
| The other person never agreed to any of this | Yes | No | Yes |
| What the reading is grounded in | Yes | Partly | Not stated |
| Conversations you can bring | Yes | Partly | Not stated |
Every mark below is quoted from the source further down this page. Tap a row to read the exact wording and follow the link to the original policy.
The evidence for each row
Same seven rows, with the wording each claim came from.
Your conversation after the analysis
The single question that decides how much a leak, a subpoena or an acquisition could cost you later. A conversation that was never stored cannot be handed over.
Not kept. The message text exists in memory for the length of one request and is never written to a database.
Retained while your account exists.
“We retain your data for as long as your account is active or as needed to provide you with our services, comply with our legal obligations, resolve disputes, and enforce our agreements.” source
Stays in your browser.
“Your personal data from loaded messages will not be shared with us or anywhere online. They remain in the browser.” source
Training models on what you upload
Once a conversation is in a training set it cannot be taken back out. Anonymisation limits who can be identified; it does not return the text to you.
Never. We train nothing, and our AI provider's API terms exclude API data from training.
Listed as a use of collected information.
“Training and improving models using anonymized, aggregated data” source
Not stated in their policy.
Anonymised data used commercially
This is the clause that turns a private conversation into a product. It is usually several paragraphs below the reassuring part.
No. Nothing is sold, shared or licensed — there is no data-sharing integration in the codebase at all.
Any business purpose, plus named partner categories.
“We may create anonymized, aggregated data from the information we collect. This anonymized data cannot be used to identify you and may be used for any business purpose, including developing analytics models, making service improvements, and creating new products and services.” source
Only if you agree to it.
“You will be asked for consent to share any derived data from these messages if you choose to do so.” source
Where the analysis runs
The honest version of the privacy question. Anything that reaches a server is a thing that can be intercepted, logged or compelled, however good the intentions.
Both. The free Quick Look is computed entirely on your device with no network call. The full read is sent once to our backend and on to OpenAI, then discarded.
On their platform.
“All of the data that is processed on our platform is protected through anonymization, industry-standard cloud infrastructure, and rigorous data agreements and practices.” source
The other person never agreed to any of this
Half of every conversation belongs to someone who was not asked. What a service does with their words is a question about them, not about you.
Their messages are not stored, not shared and not trained on, so there is nothing of theirs left to keep.
The obligation is passed to you.
“Uploaded conversations contain messages from participants who have not agreed to this policy. By uploading, you represent you have a reasonable basis to share this content.” source
Nothing is uploaded, so nothing of theirs is held.
What the reading is grounded in
Attachment style has a measured research literature behind it. MBTI type does not — it is popular, and it has never held up as a predictor of relationship outcomes.
Attachment theory, with the ECR-R self-report available in the Library so you can check the model against yourself.
MBTI personality typing, listed as a headline feature on their own comparison page.
“Personality Analysis (MBTI)” source
Not stated.
Conversations you can bring
A tool you cannot get your history into is a tool you cannot use. This is the most common place people give up.
Nine: WhatsApp, iMessage, Instagram, Messenger, Telegram, TikTok, Snapchat, WeChat and SMS — six by file upload, the rest by paste.
WhatsApp, iMessage and Instagram DMs, per their own comparison page.
“WhatsApp Support · iMessage Support · Instagram DM Support” source
Not stated on their policy page.
Other tools in this category — screenshot readers and the various “delulu” apps — are deliberately left out of the table. We could not find primary sources stating their retention and training practices, and repeating a rival’s description of them is not evidence. If you run one and want a row, send us your policy and we will quote it.
Where we are not the right answer
A comparison in which the author wins every row is an advertisement. These are true.
If nothing may ever leave your device, our paid read is not for you
The full analysis is sent to our servers and on to OpenAI. It is not stored and never trains anything, but it is transmitted, and no amount of careful wording changes that. Our free Quick Look does run entirely on your phone, and DoubleText's in-browser approach avoids transmission too. If that is your hard requirement, use one of those.
We ask for an account; some alternatives do not
Analysis requests carry an authenticated user id, which is how rate limiting works and how your history follows you between devices. It also means the requests are not anonymous. A purely local tool has nothing to authenticate and is genuinely more private on that axis.
Attachment style is one lens, not a diagnosis
It is a well-evidenced framework for how people behave under relational threat. It is not a clinical assessment, it does not diagnose anyone, and a conversation is not a person. Anything promising a verdict on your relationship is overselling — including anything we might be tempted to say.
Whatever you use, you can bring it
9 platforms — 6 by file upload, the rest by paste. Every one has a step-by-step export guide.
- WhatsAppUpload _chat.txt
- iMessageCopy & paste
- InstagramUpload message_1.json
- TelegramUpload result.json
- MessengerUpload message_1.json
- Text messagesUpload .xml
- SnapchatCopy & paste
- TikTokUpload user_data.json
- WeChatCopy & paste
Anything we cannot read reliably is refused rather than guessed at — a parser that invents speaker names from a file it does not understand produces a confident analysis of nothing.
Read the conversation you already have
The free Quick Look runs entirely on your device and needs no account. The full read goes out once and is never kept.