What bothers me about these projects is you are exposing the other side of your conversations to being uploaded to a 3rd party AI company who will ingest it in to their training data. You personally might not care about this privacy intrusion but the people you are messaging never consented to this.
This has been happening for years... I distinctly remember people giving access to contacts to Whatsapp and having the entire contacts slurped into their cloud.
I agree with you on principal but also think that from a practical point, that ship has long sailed.
It's always been a possibility that when you sent an email to someone their provider could be building an ad profile on you. Any third party tool someone else uses to manage their personal files and communications could always have some arbitrary privacy policy that let them do what they see fit.
For the most part, you never know what random SAAS someone on the other end of a relationship is really using.
Yes but I'd say some people (me) like WhatsApp because of the privacy and end-to-end encryption. Email i assume is less private but still more private than a public post.
Imagine someone puts all the Signal messages someone else sent them through an AI. I dunno. I think problem is most AI doesn't work like Confer.to, also by Signal's founder.
People have been screenshotting conversations, saving them on their phones, and sending them to friends over text for as long as there have been conversations to screenshot.
The entire concept of conversations being "secure" is only as good as your ability to trust the other party isn't doing something insecure with them (and that yours and their devices are, themselves, as secure as you think they are. Which they generally aren't.)
All of which is to say, it's security theater and always has been when applied to real life situations.
Except now there’s not. The fact is, when you tell somebody something, what they choose to do with it is out of your hands. Unless you want to start using drm-encumbered systems.
There's still a difference between a person deciding to share one or a bunch of messages they receive --usually I know the people I write to, so I know more or less what to expect-- and systematically uploading received message to some AI. Fine if they do it locally, not fine if the messages I write end up at Google, OpenAI or anybody else.
This is unrealistic. Hardly anyone who is going to feed their own messages to an AI is going to pay attention to whether it’s local or cloud based. Get real.
> It's always been a possibility that when you sent an email
Yes, and that's exactly why E2EE messaging became a thing. Signal put it on the map through the WhatsApp integration.
To me there is a difference if the server spies on me (that's without E2EE) or if the person I am writing to leaks everything on the Internet.
If the person you're writing to leaks everything on the Internet, you should consider you are talking to them in public. I'm sure you don't say all the same things when you are in the metro or when you are in private, or do you?
> Yes, and that's exactly why E2EE messaging became a thing.
Pretty much never for email. Maybe some people go out of their way to perform PGP but that's usually only for the most sensitive of information and sometimes not even for that.
The only thing I've used PGP for in many years is encrypting my home address when buying darknet drugs
- I have friends that want utmost privacy and shun me for using Android and buy iPhone etc but then install Google Maps. Add my name/address/email/phone etc to google sync.
- Worse - many employers are using FAANG for services. How many managers are uploading Excel sheets? Who supervisors them?
- Even if I keep every thing locally (which I don't ) finally govt will lose their passport database and ...
So yeah, whatsapp messages are E2E encrypted and therefore can not be used as training data unless you upload them to the cloud manually, as this project does
So what? WhatsApp is end-to-end encrypted. Signal is better (e.g. because the client is open source), but security is a gradient. WhatsApp is better than Telegram in terms of security.
> but then install Google Maps
Again, it's a gradient. Sure, they leak some data, but at least they don't leak all the content of all their messages. Or would you say that if you leak your name on the Internet, you may as well publish your sextape because "now everybody knows your name, you have nothing to hide anymore"?
> How many managers are uploading Excel sheets?
So because my company leaks my work documents to Microsoft means that I should leak the content of the messages I send to my partner? How do you get to that conclusion?
> finally govt will lose their passport database and ...
Right. And it is a known fact that when the government loses their passport database, suddenly your private pictures appear on the public Internet! /s
There has always been the possibility that people would do dumb things with email they receive, but it wasn't a widespread thing to do, there weren't many reasons to do them and the tools weren't easily accessible to most people.
But LLM agents are really useful and the tools to give one access to your entire digital life is just an OAuth prompt away.
WhatsApp is end-to-end encrypted, that's probably* better than using Telegram.
* Unless you want to go into a conspiracy theory that all the Meta employees can see the source code and nobody has ever felt like whistleblowing about the backdoor Meta put into WhatsApp, that is.
WhatsApp obviously more encrypted than Telegram since later have no proper E2EE whatsorver by default and also sync all history from server.
Its still preferable to use Signal since now it actually supports numberless registration. Hopefully they will actually fee payable not via credit card at some point.
These two things are mutually exclusive. They can store encrypted backups on a server somewhere, but unless they have the decryption keys from both parties (E2E) then no, it can't be stored in plain-text.
If you don't encrypt the backups, they are unencrypted, yes. That's why you should encrypt your backups. It's baked right into WhatsApp on the screen where you manage your backups.
Don't forget that metadata is not encrypted, so Meta know who you are writing to, who writes to you, who your contacts are, all the usage pattens of the app, and more metadata (not sure if it's still the case, but a few years ago WhatsApp even uploaded the battery state of your phone to Meta).
You never know what the people you're talking to are doing with your data. This isn't any different. The conversation involves 3 parties: Sender, WhatsApp & Recipient. If you want your conversation to be private, you must trust all 3.
I never know that the people preparing my food at the restaurant are doing with it either but if I see someone doing something wrong I'm going to call it out. We should all make an effort to maintain some level of ethics even if we don't have total control over the environment.
Feeding someone else's personal data in to an AI company is unethical even if there is no way as an individual to prevent someone doing that.
it's an MCP, so you could theoretically hook it up to a private/local model/agent.
but it's an MCP, so that also means it'll be more token-verbose than an API, which means your local agent hardware is going to chug a bit.
i have reached the point of just using LLMs to build my own API servers and tooling like search, and teach the LLM how to use those APIs, rather than MCPs.
So true! I respect the work too, but there are plenty of people who haven’t consented. However, is it that different to Meta reading your messages? Lol
You’re gonna get a ton of “what about X” replies, which is sad, disappointing, and goes to show that people don’t understand the scale and severity of solutions like these.
What about… nothing. Nothing. A friend screenshotting an E2EE message is NOT even in the same world of trouble as an agentic automation over all their conversations.
But email… nope. Stop. A single person passing group chats to Claude will compromise the few hundred people in our parenting WhatsApp community.
> You personally might not care about this privacy intrusion but the people you are messaging never consented to this.
The world always worked like this though, as soon as "data" is shared out from your head to the wider world, it has "leaked" and can no longer be considered secret. This was as true 100 years ago as it is today. The only secrets that can be kept, remains the ones you keep inside of your head and don't share with others.
I've been using Baileys - but it's not ideal. (Periodically you have to go manually log the bot back in.) Is there no platform that's friendly to personal bots? Or is that what Telegram is for? Discord I suppose...
I've experimented with some of these whatsapp-web-esque hacks (whatsappweb.js, whatsmeow, ...) and every now and then I'll get banned. It's quite inconvenient if you live in a country where WA is the universal contact channel for pretty much everything.
I think governments ought to take the proliferation of WA , or X in their countries seriously.
Open clients like Telegram make a lot more sense. WhatsApp is at the whim of Zuck's and whoever is ruling US's political inclination on what is allowed or not.
That is not a great way to run a country. US corporations in many ways are like worms.
While I see the security concerns I would like to point out that this project is interesting and a great stepping stone for easier to make whatsapp AI support chat bots.
Interesting approach, we did kaptionai.com/mcp with a slightly different approach, the MCP runs on your own WhatsApp Web or in our own app, it can run locally or through our own bridge (which is open source as well).
I am not saying this proudly, but given how important is to stay in touch with others, I would never jeopardize my Gmail account and my Whatsapp account. Understand that using a bot to access your Whatsapp account is against TOS and your account could be suspended.
That’s also what’s keeping me from connecting my WhatsApp to an AI - even readonly which would already be very powerful and useful.
Every now and again I sketch out an automatic iPhone backup over wireless workflow so I can extract the WhatsApp conversations once a day - which may be a good usability/tos-safety trade-off - but that may or may not be worth it.
I don't get how of the personal assistant can be useful without having to see my conversational data so this is great.
Do you know if running Whatsapp in Cloud risks your account to be banned?
I'm working on a similar tool, connecting all that data to build your context graph that knows all the people and conversations and expose it via MCP. https://lobu.ai/for/personal/
My approach is requiring you to have a home device where you keep your accounts open in a browser and observing the data from there. The same applies to Linkedin, Instagram, X etc. even reading my timeline to summarize what I need to know without me doomscrolling.
I literally just went through it saying I was violating the TOS on my personal account for the crime of linking my Mac and IOS phone as a secondary account to my main account on my Android phone. I had been using this workflow for a few years and this happened all of the sudden earlier this month. This was using the Whatsapp official apps and recommended process for doing this.
When that happens it immediately cuts your access and makes you request a reconsideration that can take 24+ hours. Some accounts get booted without this option. So if you use Whatsapp for anything important that can be a really big problem. I was reinstated, and only discovered this from going through a couple of cycles trying to add these back.
So doing what you are doing with this seems almost like a guarantee that you will be perma banned.
I've always regarded WhatsApp as a very private thing and if on the other side of things, if somebody's using repos like these and uploading my messages, you know, which as far as we know are being read by Claude and all these AI companies, the more data you put in, the more data they get inside and for their training and they could use it for, I don't know, God knows what. But I never consented to this. So the more trouble here is not that you're using it or people are using it. The trouble is that there are a lot of people in the world who have not consented to this. There are people on the other side of WhatsApp who have not consented to this. Why would you build something like that? I mean, it's just unfair for a lot of people.
This post is quite disingenuous. Both the title and body talk about it being free, but the actual readme on the repo says:
> […] default CPU limit on Workers is too small for it. wrangler.jsonc raises it, and raising it needs the paid plan, which is a few dollars a month. I haven't tried the free plan.
This is alright until you understand that the data is going through a 3rd party AI provider for their data training, but this is not what a user intentionally wants
Made it encrypt the data at rest and support multiple users and published it on https://whatsapp.finereli.com if you don't want to install your own and just want access to your WhatsApp messages (groups included!) with your agents.
Limited to the first 100 logins for now so I can figure out how much this costs and if I can support this at scale.
D2OQZG8l5BI1S06 | 23 hours ago
SchemaLoad | 23 hours ago
WrtCdEvrydy | 22 hours ago
palata | 21 hours ago
doesnt_know | 22 hours ago
It's always been a possibility that when you sent an email to someone their provider could be building an ad profile on you. Any third party tool someone else uses to manage their personal files and communications could always have some arbitrary privacy policy that let them do what they see fit.
For the most part, you never know what random SAAS someone on the other end of a relationship is really using.
j45 | 22 hours ago
jimkleiber | 22 hours ago
Imagine someone puts all the Signal messages someone else sent them through an AI. I dunno. I think problem is most AI doesn't work like Confer.to, also by Signal's founder.
j_bizzle | 21 hours ago
The entire concept of conversations being "secure" is only as good as your ability to trust the other party isn't doing something insecure with them (and that yours and their devices are, themselves, as secure as you think they are. Which they generally aren't.)
All of which is to say, it's security theater and always has been when applied to real life situations.
SchemaLoad | 21 hours ago
hamburglar | 14 hours ago
ale42 | 13 hours ago
hamburglar | 12 hours ago
ale42 | 11 hours ago
anonreplier | 13 hours ago
hamburglar | 12 hours ago
palata | 21 hours ago
Yes, and that's exactly why E2EE messaging became a thing. Signal put it on the map through the WhatsApp integration.
To me there is a difference if the server spies on me (that's without E2EE) or if the person I am writing to leaks everything on the Internet.
If the person you're writing to leaks everything on the Internet, you should consider you are talking to them in public. I'm sure you don't say all the same things when you are in the metro or when you are in private, or do you?
LoganDark | 19 hours ago
Pretty much never for email. Maybe some people go out of their way to perform PGP but that's usually only for the most sensitive of information and sometimes not even for that.
The only thing I've used PGP for in many years is encrypting my home address when buying darknet drugs
random2021 | 14 hours ago
- Most people are not in Signal
- I have friends that want utmost privacy and shun me for using Android and buy iPhone etc but then install Google Maps. Add my name/address/email/phone etc to google sync.
- Worse - many employers are using FAANG for services. How many managers are uploading Excel sheets? Who supervisors them?
- Even if I keep every thing locally (which I don't ) finally govt will lose their passport database and ...
gbalduzzi | 13 hours ago
Whatsapp uses the Signal protocol for E2EE: https://signal.org/blog/whatsapp-complete/
So yeah, whatsapp messages are E2E encrypted and therefore can not be used as training data unless you upload them to the cloud manually, as this project does
whyagaindavid | 4 hours ago
https://www.bloomberg.com/news/articles/2026-04-28/us-ends-i...
palata | 11 hours ago
So what? WhatsApp is end-to-end encrypted. Signal is better (e.g. because the client is open source), but security is a gradient. WhatsApp is better than Telegram in terms of security.
> but then install Google Maps
Again, it's a gradient. Sure, they leak some data, but at least they don't leak all the content of all their messages. Or would you say that if you leak your name on the Internet, you may as well publish your sextape because "now everybody knows your name, you have nothing to hide anymore"?
> How many managers are uploading Excel sheets?
So because my company leaks my work documents to Microsoft means that I should leak the content of the messages I send to my partner? How do you get to that conclusion?
> finally govt will lose their passport database and ...
Right. And it is a known fact that when the government loses their passport database, suddenly your private pictures appear on the public Internet! /s
BorisMelnik | 20 hours ago
franga2000 | 3 hours ago
But LLM agents are really useful and the tools to give one access to your entire digital life is just an OAuth prompt away.
_ink_ | 22 hours ago
https://news.ycombinator.com/item?id=43901204
throwaway27448 | 22 hours ago
dvt | 22 hours ago
delijati | 22 hours ago
ceejayoz | 22 hours ago
palata | 21 hours ago
* Unless you want to go into a conspiracy theory that all the Meta employees can see the source code and nobody has ever felt like whistleblowing about the backdoor Meta put into WhatsApp, that is.
SXX | 19 hours ago
Its still preferable to use Signal since now it actually supports numberless registration. Hopefully they will actually fee payable not via credit card at some point.
palata | 11 hours ago
It's preferable to use Signal because they don't collect as much metadata as WhatsApp, and the client is open source.
I don't care so much about the numberless registration. Actually I don't even understand why people care so much about it.
arshxyz | 16 hours ago
palata | 11 hours ago
MYEUHD | 16 hours ago
simplesocieties | 15 hours ago
jeroenhd | 12 hours ago
palata | 11 hours ago
ale42 | 13 hours ago
Don't forget that metadata is not encrypted, so Meta know who you are writing to, who writes to you, who your contacts are, all the usage pattens of the app, and more metadata (not sure if it's still the case, but a few years ago WhatsApp even uploaded the battery state of your phone to Meta).
palata | 11 hours ago
asciii | 20 hours ago
Not like Meta is a saint...since they're competitors
Rohansi | 20 hours ago
TonyFalco | 19 hours ago
SchemaLoad | 19 hours ago
Feeding someone else's personal data in to an AI company is unethical even if there is no way as an individual to prevent someone doing that.
keerthiko | 16 hours ago
but it's an MCP, so that also means it'll be more token-verbose than an API, which means your local agent hardware is going to chug a bit.
i have reached the point of just using LLMs to build my own API servers and tooling like search, and teach the LLM how to use those APIs, rather than MCPs.
toppa33 | 13 hours ago
jacmilts | 12 hours ago
port11 | 9 hours ago
What about… nothing. Nothing. A friend screenshotting an E2EE message is NOT even in the same world of trouble as an agentic automation over all their conversations.
But email… nope. Stop. A single person passing group chats to Claude will compromise the few hundred people in our parenting WhatsApp community.
dannyw | 8 hours ago
I have a similar setup for my personal agent, except all inference runs on my local homelab and open weight models.
I see no ethical qualms to this. These people sent messages to me, and only I have access to it; I’m not sharing it with a third party.
embedding-shape | 9 hours ago
The world always worked like this though, as soon as "data" is shared out from your head to the wider world, it has "leaked" and can no longer be considered secret. This was as true 100 years ago as it is today. The only secrets that can be kept, remains the ones you keep inside of your head and don't share with others.
jrecyclebin | 23 hours ago
dandaka | 11 hours ago
nazcan | 9 hours ago
goodcake3963 | 23 hours ago
taldo | 22 hours ago
nojvek | 5 hours ago
Open clients like Telegram make a lot more sense. WhatsApp is at the whim of Zuck's and whoever is ruling US's political inclination on what is allowed or not.
That is not a great way to run a country. US corporations in many ways are like worms.
throwaway27448 | 22 hours ago
throwaway27448 | 22 hours ago
bitilia | 21 hours ago
kshmir | 20 hours ago
pizzafeelsright | 21 hours ago
If a bot needs to text me it sends an API request or it can read whatever I allow with full visibility as I have the logs and limit the access.
The amount of effort was minimal. About 4 prompts to get it all working and zero SaaS, 3rd party, or purchases after the software was built.
kshmir | 20 hours ago
kshmir | 20 hours ago
101008 | 20 hours ago
deadbabe | 10 hours ago
beng-nl | 6 hours ago
Every now and again I sketch out an automatic iPhone backup over wireless workflow so I can extract the WhatsApp conversations once a day - which may be a good usability/tos-safety trade-off - but that may or may not be worth it.
febed | 32 minutes ago
buremba | 19 hours ago
Do you know if running Whatsapp in Cloud risks your account to be banned?
I'm working on a similar tool, connecting all that data to build your context graph that knows all the people and conversations and expose it via MCP. https://lobu.ai/for/personal/
My approach is requiring you to have a home device where you keep your accounts open in a browser and observing the data from there. The same applies to Linkedin, Instagram, X etc. even reading my timeline to summarize what I need to know without me doomscrolling.
lgleason | 18 hours ago
When that happens it immediately cuts your access and makes you request a reconsideration that can take 24+ hours. Some accounts get booted without this option. So if you use Whatsapp for anything important that can be a really big problem. I was reinstated, and only discovered this from going through a couple of cycles trying to add these back.
So doing what you are doing with this seems almost like a guarantee that you will be perma banned.
jiggunjer | 18 hours ago
fantasticwaddle | 18 hours ago
neuropacabra | 16 hours ago
calmingsolitude | 15 hours ago
> […] default CPU limit on Workers is too small for it. wrangler.jsonc raises it, and raising it needs the paid plan, which is a few dollars a month. I haven't tried the free plan.
https://github.com/finereli/whatsapp-mcp/tree/64f1b3017931e4...
meherabhossain | 15 hours ago
cjimesh | 15 hours ago
iamronaldo | 11 hours ago
olalonde | 6 hours ago
blockway-dev | 6 hours ago
[OP] zoozla | 3 hours ago
Limited to the first 100 logins for now so I can figure out how much this costs and if I can support this at scale.
Enjoy!