> No channel modes and no channel operators, which is a decision rather than a gap: a global channel is owned by nobody, so there is nobody to be an operator of it. Blocking is per person and per instance instead
That's completely unworkable. Lets say I create #some_minority, and people from many different servers join. Some shithead comes and starts screaming slurs at us. Now the admin for every server must block that person. Multiply that by every channel, and every admin is responsible for moderating the entire ecosystem. The only way they can do this is with shared blocklists, which have been a nightmare on mastodon.
If this is really distributed, then off the top of my head I imagine you could possibly make some sort of reputation-based filter list wherein trusted admins can subscrbe to each other's filter lists such that when one admin blocks are user, it quickly propagates to other admins' lists. Somewhat similar to how email domain spam block lists work.
Ownership of namespaces is such a hard problem, our society has had to form global organizations and some of our most secure distributed systems to make it happen.
Working in the p2p space, scalable namespace ownership that doesn't get instantly ruined by perverse incentives is the holy grail. I don't think it is actually tractable in any "central authority free" system. Don't get me started on blockchains that tether names to the most perverse systems available and call it victory.
I get the impression that every attempt to make block lists ends up getting used to target people an admin of the block list doesn't like. I think the root of it is that the admins consuming the blocklist don't want to have to spend time on moderation and are remote from communities where the blocklist is happening. Also the scope of the blocklist is the instance. Compare that to community internal moderation (e.g. channel mods), where someone close to the community is invested in the health of the community and the scope of the block is the relevant community.
So rooms are "global" between whatever hosts your host happens to know about? So it's one giant netsplit party forever and only your server admin can ban someone?
I am not the owner of the project, I just use it. Federated rooms have no owners, there will be no collisions (they would simply merge). Nicks are tied to their own servers, and hence unique and will not collide.
This has been an issue for as long as IRC itself has existed. There are solutions to resolve such "netsplits" and "netjoins", but they all rely on the servers in the network being trusted and not malicious.
With Parley, where anyone can bring their own instance, and they are not guaranteed to not to be malicious, that could be an interesting issue to solve.
Is anyone leveraging IRC/XMPP/etc for agents or A2A communication? It seems like a natural fit; all the "human agents communicating" abstractions and underlying technology pieces are in place and mature, as are methods for dealing with unruly actors.
Every person (or team) runs a small instance for their own domain. Instances find each other through DNS and well-known identity documents, exchange signed messages over HTTPS, and present the whole federated network to ordinary IRC clients such as Lurker, Mango, mIRC, WeeChat, Textual, etc., without the need of any plugins.
Hacker News marketing skills tell people to leave a context comment on Hacker News. Don't complain about it too much, that is our way of figuring out who is lazy in an obvious way!
When someone makes a submission, the text field becomes a comment when the URL field is not empty. (but sometimes not?) It's a poorly designed form that should be redone.
Presumably if the protocol becomes popular you will have public instances that anyone* can sign up to, the same way other federated services work now.
(It's approximately as hard for a layman user to sign up to a Mastodon instance as it is to sign up for X. The fact that most layman users don't do that is a separate issue.)
There’s a specific downside to LLM-based development that people can get neck deep in a new project without interfacing with the existing work in the field.
In this case, this is basically a poorly specified implementation of half of XMPP. Of course, I half expect the LLM would have mentioned that at some point, but the repository does not.
Unfortunately XMPP is an absolutely terrible protocol. IRC's not much good either, but at least it has the excuse of being ancient and limited in what it wanted to achieve.
I don't know how it just happens that sending text messages to people can manage to result in specs that are painful to implement.
IRC is good in that it is open, durable, and slow moving. All of those points are intentional features that make it a stable cockroach in a nuclear world.
I briefly entertained building a modern chat experience on top of IRC when I came to this realization.
IRC really is good. It's reliable too, but it really fell behind the times and never pushed forward with additions to the spec and such. When the clients and libraries tapered off too it just felt like it froze.
>Unfortunately XMPP is an absolutely terrible protocol
Can you provide more details here? I've never seen an issue with the protocol at all, more issues with feature difference between servers depending on what they have decided to implement or not.
Most times I've seen complaints about it, it boils down to "XML being yucky" and the extension component model being difficult to write clients for, which is true but not really a fault of the protocol, it's more of an inherent tradeoff of supporting opt-in extensions at all
I don't have that many problems with XML myself, but XMPP is weird even in XML land, because it tries to do XML all the way.
For instance, it works as a constant, uninterrupted stream. You never close the document until the disconnection. Which means you absolutely have to parse it with a stream parser.
It could have been something sensible, like a document per message: "Here's 300 characters of XML document: <xml>....". But nope.
What’s so complex about it? From what I gather (after skimming the document just now), you attach an extra element to a message which says how to style which parts of the message, from codepoint n to codepoint m, essentially. At least conceptually, that seems really simple and straightforward.
Are there more elegant or natural ways to do it? Probably. But when you say ‘extremely complex’, saying ‘codepoints 7 to 15 should be bold’ is not what comes to my mind.
* It's not standard. You can't plug in some normal formatting library there and be done with it.
* You'll probably end up having to write conversions back and forth.
* It's fragile -- if anything gets misaligned it'll break
* It's tooling unfriendly. Think things like Nagios, Grafana, etc sending formatted notifications. Everything can do this or <b>this</b>, but practically nothing is set up to accommodate this format.
* It leaks into other subsystems. You can't eg, just search/replace/insert/delete words if you need to for any reason without breaking this.
Is it the worst thing ever? No, but it seems to go with the pattern that nothing in XMPP is normal or comfortable. Everything has this weird particular flavor to it and more complexity than necessary.
If it were up to me, my option would be to support 2 things: Markdown for simple cases, and embedded HTML for when you really have to get fancy. Each of those can be handed out to existing code and you don't ever have to keep adding extensions because what if people want colors or something now.
None of this seems particularly different from doing <i>spans</i> inline (what if you forget to close one? or they overlap?), but it has a major benefit of being very backwards compatible, supporting future variants in parallel for gradual migration, and doesn't require any changes to search tools. And separating markup instructions from content is generally a very good idea.
And you could use markdown. Just get the rendered spanned text result and transpile it. This would be true for any editor as well, just get convert it to spanned text, edit, convert back.
And with HTML you'll still have to specify what subset you support, and many tools won't support that either. Markdown suffers from this too, by supporting embedded HTML, though at least commonmark is pretty baseline and well supported.
There is zero baggage on NOSTR regarding identities. An npub became the non-ambiguous way to reference anyone regardless of where they are writing from.
Thank you for the reference to that project called anproto, never had heard of it and I'm still without understanding what problem it really solves. Seems to encrypt texts but then contains zero outside metadata to route it somewhere. The website is very scarce on details for someone unfamiliar the scuttle-something they mention.
* It's hard to even parse, XMPP uses an uninterrupted XML stream.
* The contents are often baroque and complex
* Standards are a mess, and stuff that should be in core isn't
* Data loss is possible
* Protocol wasn't made for mobile devices
* Multiple devices are terribly supported
* Data loss is possible
From my attempts long ago, and other testimonials, writing an XMPP client is a full time job of solving weird problems that shouldn't exist in something better designed.
> We hear this too often: “XMPP uses XML. It should use JSON—it’s more modern.”
I can't take this write-up seriously if it starts like that. I still read the whole thing though and I couldn't find any solid argument as to why someone would prefer XML to JSON for XMPP.
> This is especially true in browser environments, where XMPP streams run over WebSockets, which naturally frames the XMPP protocol. That’s why you are never actually working with XML trees consuming large chunks of memory. Modern implementations like XMPP.js go further and use LTX—a lightweight parser built specifically for XMPP’s streaming model—rather than the browser’s DOM parser. The result: developers work with JSON-like objects anyway. The wire format becomes invisible to your application code.
That to me, is an argument for using JSON, not XML. XML is strong when you have elements referencing other elements in your document structure or DOCTYPE for grammar defs. I might be missing something but I don't get how streaming XML is to be preferred over JSON for XMPP.
If it was actually using XML as a markup language then that would be a reason to stick with it. But XML seems to be complicating things much more than it simplifies.
>> XML remains the best format for representing trees—deep hierarchies of nested data. JSON handles flatter structures well, but good messaging protocols are extensible: extensions can be embedded at different levels and composed together, like Lego bricks. That’s where XML shines.
Anything reasonable you'd be doing with messaging protocol is pretty flat.
I once implemented the presence and basic messaging functionality of XMPP for a web site, using a braindead bridge (that I also wrote; so, the meaningful logic lived in the browser).
Considering I'd never hosted an XMPP daemon and didn't know anything about the protocol (I'd used XMPP clients a little bit, but had never looked at the protocol) and got a server (that part, I didn't write), an auth connector for our website's authentication system (so the daemon would authenticate against that instead), prod-ready and the features I wanted all working smoothly and reliably in maybe three weeks of very part-time work (this'd be, like, 3-4 part-time days with LLMs now, tops, from the same starting point)... seems decent to me? I mean I did direct work with the protocol, didn't just glue together libraries, and it was pretty damn good. Also (and I know browsers seem to be retreating on this front, which sucks) being XML made it very nice to work with in a Web context, since you can just ask the browser to turn ~any XML into a DOM for you, and get a bunch of functionality for free.
> I don't know how it just happens that sending text messages to people can manage to result in specs that are painful to implement.
This is what boggles my mind. We're talking: Text. Over the Internet. It should not be a complex, difficult problem! We've been sending text over the Internet from the moment the Internet went online. So how is it that 10 companies have managed to find 30 different ways to do it, which are all incompatible with each other? You have to TRY to fail this badly.
This is what boggles my mind. We're talking decades of arguments and discussions, dozens of different ways to do it, with dozens of tradeoffs that you could start listing with 20 seconds of thought, and still people just pop in like "why isn't it simple??" as if that's some award winning insight.
Come on, try! Start listing some of the core features of the different protocols/clients and answer your own question as to why it isn't simple! Presence. Chat history while offline. Threading. Mobile devices. Intermittent/mobile/cellular/wifi/CGNAT connectivity. Multiple clients on the same identity and presence. Identity establishing and confirming. Encryption. End-to-end encryption. Backing up and restoring chats, encrypted chats, group chats. Federation. Friends/groups/permissions/trust boundaries. Markup, markdown, formatting, images, emoji, unicode, code blocks. Audio, video, broadcast, multicast. Discovery. Extensibility. Federation. Store-and-forward.
> "We've been sending text over the Internet from the moment the Internet went online"
And IRC has RFC 1459, RFC 2810, RFC 2811, RFC 2812, RFC 2813, RFC 7194 and it does almost none of those things.
And you can still use IRC. But IRC is not good. At this point it's either wilfull ignorance of what people use computers for, or it's indistinguishable from deliberate trolling.
Threading? God, I hate threads in a live chat. Is it never not awkward to use? Slack is the example I'm thinking of, though I'm sure it's shit everywhere else too. In Slack, rather than display the one threaded reply just below (maybe indented), it makes me go to another window, but then someone else is replying non-threaded, so I have to bounce back and forth for the next 10 minutes.
As an anecdotal counter-point, I love threads in a live chat. When used correctly, it means when my idiot friends start having a heated discussion about some random topic while I'm at work, I get one 'ding' from my phone and then never again until I start interacting or they @ me. As opposed to the 427 new notifications I get on my phone about whether Golden Eye was better than Mario Kart 64.
Ultimately the tech is only as good as the humans that use it. If you're chatting with folks that can't understand the difference between a thread or not, that's a meat-bag issue and not anything an RFC can fix.
I completely agree. Threads in Slack should be a great way to make conversations easy to navigate but damn did they fsck up ever part of the UX:
- opening threads in a new window is in some hard to find and hard to remember location
- threads can’t be inlined (like you’ve described). I actually like how they’re in a pane to the right but sometimes that causes issues (as you’ve described). Sometimes it’s just that the view is too narrow.
- there’s no way to shift non-threaded comments into the thread. I’ve lost count of the number of times conversations have happened over multiple threads and/or non-threaded replies.
The way they implemented it is a complete clusterfuck.
I have half-wondered many times whether it would be possible/practical to build a chat system on top of MQTT or similar. Chat seems like mostly a pub/sub problem from where I sit.
or maybe they know about XMPP and prefer IRC. I have always wanted to build one of these on top of IRC. I grew up on IRC and sometimes just the nostalgia of familiar tech is what drives us to build towards it not how good it is.
That is currently still the differentiator between people producing code and shipping something and people who care and actually have years of experience shipping. At the current state how well you steer the LLM and which questions you ask still matters, maybe in the future "build me an app" will give amazing results, but right now you still need to know what you are doing.
In this case, it's just a waste of tokens as something not very unique was generated that doesn't have a real use case, or solves anyone's problem. As with many AI generated projects, I'm willing to bet that OP themselves will not using it any more in a month.
You guys are way too critical and about the wrong things. "Waste of tokens"? Perhaps the author know of the tradeoffs and just wanted to build something.
How you went from seeing somebody's post to deciding they don't care is a pretty big jump, one that isn't warranted. This kind of post seems like the old but now more elaborated form of hating on something that you don't even know what it is.
It just isn't fair to the work that has been put in.
I agree. This is one thing that's gotten me about the new wave of poo-pooing projects made since coding agents have gotten big. Five years ago it was 'just build something!'. I understand the market is more saturated now so building something isn't as valuable as it used to be, but it still shouldn't be treated as a negative. I don't know anything about the creator's background, but even if this project is ultimately redundant and unnecessary, it's still pretty cool that they built something.
I've done this a few times. If I had to gesture at the root cause I'd say: LLMs lack curiosity and are trained to complete assignments, not to question their validity, so both the research phase and the questioning of intent tend to get short shrift prior to building anything.
I'm not sure I agree. Seems to me to depend a lot on how one interacts with an LLM.
If you tell it to "build me an application X to do Y using programming language Z", it will comply.
If you interactively ask "I need X, can you suggest some options? use an existing product ? or build something using a library ? or build entirely myself ? can you suggest alternatives with pro's and con's ? Anything I should think of before deciding ?" then you will get an entirely different answer.
Maybe we need additional modes, aside from thinking mode, agentic mode etc. we need e.g. "sparring mode" ? or does such a thing already exist ?
Agreed, you have to ask them to look for prior work. But then they will do a good job finding it in most cases. I think it’s partly their universal tunnel vision and partly sycophancy.
> I half expect the LLM would have mentioned that at some point, but the repository does not.
The repository does not even mention this software is LLM-generated.
I wonder when/if the collective realization will land that LLM's failure to retain provenance during training means the code it generates exists in an indeterminate state of public domain / IP trap.
But it's not just for their own purposes. They shared it with the public which means they will get feedback, good and bad. No one forced them to post to HN.
”There’s a specific downside to LLM-based development that people can get neck deep in a new project without interfacing with the existing work in the field.”
That’s because it’s a paradigm shift, not meant to be planned or organized by humans. The prior paradigm was that there were subject-matter experts who were human. The new paradigm is that the only author of all code is to be a machine. This new paradigm requires removing all prior authors (who happen to be human), and removing them simply requires overcrowding their output such that it’s not locatable or trustworthy.
Ya, ive stopped interacting with all new projects, the amount of slop written by people who have no business remotely putting something out there for others to use is untenable.
Curious, why the interest in building your own ntfy client (instead of using the one already built/available)?
By the way, for years now, i have used a little python (cli) app to send messages into a private matrix room. This room has 2 people in it: me and a bot. (The bot is simply the matrix account that sends the messages.) I went this route because i did not want to install another app - such as ntfy - on my mobile simply to receive server notifications, etc. And, since i had ben on matrix already chatting with friends, it happened to be a low-effort approach. To clarify, i am no longer self-hosting my own synapse home server...so i'm simply piggybacking on the matrix.org one...but that has helped at least unify my notifications and comms to a good degree.
Now, if you were to aske me to implement my "notifications system" via a different approach, funny enough, i would gauge between XMPP and ntfy to see which is lighterweight....because i believe them to both be quite light on the need for resources. I don't know which is lighter (sorry, i know i'm not answering your question). however, i *think* the ntfy server is less hasle....Only because xmpp takes a little more work to setup...at least that was my experience a couple of years ago. I cast no shade on either xmpp nor ntfy - because to me boh are awesome...i' m merely saying that resources needs aside, i wonder if ntfy might need less time from YOU.
I'd build my own NTFY client to enable two-way chat. It's not something I'm seriously considering, because it's obviously a bad idea, but that's about where I am.
>I'm searching for a simple, self-hosted chat system that allows me to host unlimited bots and send messages (with notifications) to my iPhone easily.
I do this with IRC. I used to have a blog post up about it, back when blogs weren't just training data providers for LLMs, but I'm afraid it's been deleted.
tl;dr IRC works great, but you need to wire up push notifications. For android I used pushbullet, for iOS I am still evaluating the alternatives but pushover looks okay. IRC itself is nice because the protocol is shit simple, to the point that if you can type fast enough to respond to the pingpongs you can cosplay as a bot using nothing but telnet. As a text protocol it's easy to inspect and doesn't require binary decoding of packets for debugging.
I recommend the Ergo ircd. It ships as a single binary (golang), can support tens of thousands of clients, is at the forefront of the ircv3 modernization initiative, and is under active development.
Interesting. Why not use an open link network? These will result in around the same state. Spam _will_ be an issue here, in general. And with a different backend you're going to struggle to use preexisting tooling to handle it.
Using & channels is fun, I'd be interested to see how many bots and clients fall over dead when faced with that particular bit of IRC history.
IRC linking is a mess, in part due to the spanning tree requirement. Also the server to server protocol in the RFC is spoken by zero servers so you have to pick which unofficial protocol you like best. May as well invent your own that actually fits your use case.
Except that now there's yet another standard, which nothing but this thing speaks. TS6 is very well used at the moment as everything in the Charybdis line speaks it, etc.
I had a pseudo-plan for something like this but for reasons related to privacy I had two types of chatrooms:
* regular chats, existing only on the server, no leaking the chat transcript except via users, but never via server to server.
* chambers: a global chatroom, located at a server, maybe with a MQTT anyone could subscribe to.
This never left the early planning stages though, but I thought the segregation between federated chatrooms and regular chatrooms was of interest to keep in sync with IRC open but closed nature of chats.
As someone very critical of mindlessly vibecoded software, I've worked with James (@prologic) years ago and very much respect his craft. AI assisted or not, I wouldn't describe him as a vibecoder. He's a great engineer.
Instead of a separate federated network for a single app, why not implement the app on top of a pre-existing federated network, so that nobody has to create yet another account?
Like, on top of atproto or activitypods or...?
To be fair, if i recall correctly, some of the folks who started up matrix protocol (and associated project) came from/and loved the IRC world (e.g. the rooms concept, etc.)...so it stands to reason that matrix and irc have at least some *conceptual* similarities, and hence other stuff that spins off from or leverages IRC may also appear similar to matrix.
Cool. I did something like that so agents and people could communicate with a Jira instance using Websocket-aware IRC and a browser-based client based on superchat.
Put the issue context in a #channel and invite external parties/entities, bring the results back to the work.
I have tested it but the complexity didn't justify the results in my experiment. I couldn't get it accepted into Atlassian Marketplace, either.
Besides not having C runtime linked (which is great for scratch/distroless Docker images) - any other advantage of using a pure Go SQLite instead of the C binding?
I used Bitlbee until not that many earth rotations ago - it's an IRC daemon that provides bridging of various other chat protocols. It worked way better than you would expect, especially back in the day when you could use XMPP to connect to the various proprietary chat services.
I didn't stop using it until I stopped caring about those chat services.
I wish them the best of luck. Federated systems struggle to achieve critical mass, and if they do, large corporations tend to snuff it out. I distinctly remember Google committing to The Microsoft EEE strategy with XMPP. We've left behind lots of great ideas in the last 20 years
Google helped make XMPP quite popular for awhile, then killed their own product and left us with a lot of useful specs they had contributed. So I'm not sure I'd call that EEE
> No channel modes and no channel operators, which is a decision rather than a gap: a global channel is owned by nobody, so there is nobody to be an operator of it. Blocking is per person and per instance instead
That's completely unworkable. Lets say I create #some_minority, and people from many different servers join. Some shithead comes and starts screaming slurs at us. Now the admin for every server must block that person. Multiply that by every channel, and every admin is responsible for moderating the entire ecosystem. The only way they can do this is with shared blocklists, which have been a nightmare on mastodon.
If this is really distributed, then off the top of my head I imagine you could possibly make some sort of reputation-based filter list wherein trusted admins can subscrbe to each other's filter lists such that when one admin blocks are user, it quickly propagates to other admins' lists. Somewhat similar to how email domain spam block lists work.
[seemingly negative thing] which is a decision (after hindsight), rather than a gap [scope creep]
Ownership of namespaces is such a hard problem, our society has had to form global organizations and some of our most secure distributed systems to make it happen.
Working in the p2p space, scalable namespace ownership that doesn't get instantly ruined by perverse incentives is the holy grail. I don't think it is actually tractable in any "central authority free" system. Don't get me started on blockchains that tether names to the most perverse systems available and call it victory.
> which have been a nightmare on mastodon
Why?
I get the impression that every attempt to make block lists ends up getting used to target people an admin of the block list doesn't like. I think the root of it is that the admins consuming the blocklist don't want to have to spend time on moderation and are remote from communities where the blocklist is happening. Also the scope of the blocklist is the instance. Compare that to community internal moderation (e.g. channel mods), where someone close to the community is invested in the health of the community and the scope of the block is the relevant community.
gestures broadly is going to be one of the common answers from anyone trying to ward off spam
How do you plan to handle bad actors creating biblical amounts of servers dynamically and then spamming at line rate from all of those servers?
That sounds pretty load-bearing.
Actually it’s a footgun gate
An honest gap
And then, all galaxies move away from us.
This is why we can no longer have nice things...
Online communities are meant to be pharmed..
Haha so this project speaks IRC, no less :) This is surely relevant then: https://www.mirc.com/split.html
Internet drama was so much better back then.
So rooms are "global" between whatever hosts your host happens to know about? So it's one giant netsplit party forever and only your server admin can ban someone?
Bans are per server (for what I have seeing). Federated rooms continue to exist if a server drops off, as no server "owns" them.
> Federated rooms continue to exist if a server drops off
That doesn't answer the question.
How are you resolving things like... user name collisions when the servers reconnect?
I am not the owner of the project, I just use it. Federated rooms have no owners, there will be no collisions (they would simply merge). Nicks are tied to their own servers, and hence unique and will not collide.
This has been an issue for as long as IRC itself has existed. There are solutions to resolve such "netsplits" and "netjoins", but they all rely on the servers in the network being trusted and not malicious.
With Parley, where anyone can bring their own instance, and they are not guaranteed to not to be malicious, that could be an interesting issue to solve.
CAP theorem suggests it is not solvable.
Relearning the lessons from the old days, I see. This situation is what created the Eris-Free Network.
Is anyone leveraging IRC/XMPP/etc for agents or A2A communication? It seems like a natural fit; all the "human agents communicating" abstractions and underlying technology pieces are in place and mature, as are methods for dealing with unruly actors.
I've wondered why I haven't seen this everywhere.
Oh my pi does it, not sure if that’s something from upstream or just a fork feature but from my experience it seems to work well
Parley is a chat network with no centre.
Every person (or team) runs a small instance for their own domain. Instances find each other through DNS and well-known identity documents, exchange signed messages over HTTPS, and present the whole federated network to ordinary IRC clients such as Lurker, Mango, mIRC, WeeChat, Textual, etc., without the need of any plugins.
This is an AI-generated summary of TFA?
It's a copy paste of the readme intro. Don't know why it's left as a comment here
Hacker News marketing skills tell people to leave a context comment on Hacker News. Don't complain about it too much, that is our way of figuring out who is lazy in an obvious way!
When someone makes a submission, the text field becomes a comment when the URL field is not empty. (but sometimes not?) It's a poorly designed form that should be redone.
can any layman user simply use it without running their own instance?
Yes, you will need to know someone running an instance to create an account for you.
So... no? :)
Presumably if the protocol becomes popular you will have public instances that anyone* can sign up to, the same way other federated services work now.
(It's approximately as hard for a layman user to sign up to a Mastodon instance as it is to sign up for X. The fact that most layman users don't do that is a separate issue.)
Yes, you can try it without running your own instance. You will need to use someone else's instance.
If you (top and child commentator) want to try without running your own, please send me an email: david dot collantes at gmail dot com.
There’s a specific downside to LLM-based development that people can get neck deep in a new project without interfacing with the existing work in the field.
In this case, this is basically a poorly specified implementation of half of XMPP. Of course, I half expect the LLM would have mentioned that at some point, but the repository does not.
Unfortunately XMPP is an absolutely terrible protocol. IRC's not much good either, but at least it has the excuse of being ancient and limited in what it wanted to achieve.
I don't know how it just happens that sending text messages to people can manage to result in specs that are painful to implement.
IRC is good in that it is open, durable, and slow moving. All of those points are intentional features that make it a stable cockroach in a nuclear world.
I briefly entertained building a modern chat experience on top of IRC when I came to this realization.
Slack was originally a side project built by a game dev company as a set of scripts on top of IRC. https://en.wikipedia.org/wiki/Slack_(software)#History
IRC really is good. It's reliable too, but it really fell behind the times and never pushed forward with additions to the spec and such. When the clients and libraries tapered off too it just felt like it froze.
IRCv3 aimed to do this, with mixed results.
>Unfortunately XMPP is an absolutely terrible protocol
Can you provide more details here? I've never seen an issue with the protocol at all, more issues with feature difference between servers depending on what they have decided to implement or not.
Most times I've seen complaints about it, it boils down to "XML being yucky" and the extension component model being difficult to write clients for, which is true but not really a fault of the protocol, it's more of an inherent tradeoff of supporting opt-in extensions at all
I don't have that many problems with XML myself, but XMPP is weird even in XML land, because it tries to do XML all the way.
For instance, it works as a constant, uninterrupted stream. You never close the document until the disconnection. Which means you absolutely have to parse it with a stream parser.
It could have been something sensible, like a document per message: "Here's 300 characters of XML document: <xml>....". But nope.
And what's up with this? https://xmpp.org/extensions/xep-0394.html
Is there a better standard available? As far as I know, XMPP is a real standard. If it's outdated, is there a better one?
OMG, that XEP is extremely and unnecessarily complex. Just do, I don't know, markdown.
That made me remember IBM JSONx... https://www.ibm.com/docs/en/datapower-gateway/10.6.x?topic=2...
What’s so complex about it? From what I gather (after skimming the document just now), you attach an extra element to a message which says how to style which parts of the message, from codepoint n to codepoint m, essentially. At least conceptually, that seems really simple and straightforward.
Are there more elegant or natural ways to do it? Probably. But when you say ‘extremely complex’, saying ‘codepoints 7 to 15 should be bold’ is not what comes to my mind.
Things I don't like:
* It's not standard. You can't plug in some normal formatting library there and be done with it.
* You'll probably end up having to write conversions back and forth.
* It's fragile -- if anything gets misaligned it'll break
* It's tooling unfriendly. Think things like Nagios, Grafana, etc sending formatted notifications. Everything can do this or <b>this</b>, but practically nothing is set up to accommodate this format.
* It leaks into other subsystems. You can't eg, just search/replace/insert/delete words if you need to for any reason without breaking this.
Is it the worst thing ever? No, but it seems to go with the pattern that nothing in XMPP is normal or comfortable. Everything has this weird particular flavor to it and more complexity than necessary.
If it were up to me, my option would be to support 2 things: Markdown for simple cases, and embedded HTML for when you really have to get fancy. Each of those can be handed out to existing code and you don't ever have to keep adding extensions because what if people want colors or something now.
None of this seems particularly different from doing <i>spans</i> inline (what if you forget to close one? or they overlap?), but it has a major benefit of being very backwards compatible, supporting future variants in parallel for gradual migration, and doesn't require any changes to search tools. And separating markup instructions from content is generally a very good idea.
And you could use markdown. Just get the rendered spanned text result and transpile it. This would be true for any editor as well, just get convert it to spanned text, edit, convert back.
And with HTML you'll still have to specify what subset you support, and many tools won't support that either. Markdown suffers from this too, by supporting embedded HTML, though at least commonmark is pretty baseline and well supported.
The following example (from the spec) seems quite fragile:
Identities. Creating an identity in XMPP is an absolute mess and requires domains along with approvals. At least in IRC this isn't complicated.
My own preference goes to NOSTR, just a set of public/private keys as identity and nothing else needed.
Or ANproto https://anproto.com/ without the nostr baggage, what little there is.
There is zero baggage on NOSTR regarding identities. An npub became the non-ambiguous way to reference anyone regardless of where they are writing from.
Thank you for the reference to that project called anproto, never had heard of it and I'm still without understanding what problem it really solves. Seems to encrypt texts but then contains zero outside metadata to route it somewhere. The website is very scarce on details for someone unfamiliar the scuttle-something they mention.
Other people provided some info:
https://news.ycombinator.com/item?id=9772968
https://news.ycombinator.com/item?id=31133082 (article and discussion)
But TL;DR:
* It's hard to even parse, XMPP uses an uninterrupted XML stream. * The contents are often baroque and complex * Standards are a mess, and stuff that should be in core isn't * Data loss is possible * Protocol wasn't made for mobile devices * Multiple devices are terribly supported * Data loss is possible
From my attempts long ago, and other testimonials, writing an XMPP client is a full time job of solving weird problems that shouldn't exist in something better designed.
Regarding the part part: https://www.process-one.net/blog/stop-telling-us-xmpp-should...
> We hear this too often: “XMPP uses XML. It should use JSON—it’s more modern.”
I can't take this write-up seriously if it starts like that. I still read the whole thing though and I couldn't find any solid argument as to why someone would prefer XML to JSON for XMPP.
> This is especially true in browser environments, where XMPP streams run over WebSockets, which naturally frames the XMPP protocol. That’s why you are never actually working with XML trees consuming large chunks of memory. Modern implementations like XMPP.js go further and use LTX—a lightweight parser built specifically for XMPP’s streaming model—rather than the browser’s DOM parser. The result: developers work with JSON-like objects anyway. The wire format becomes invisible to your application code.
That to me, is an argument for using JSON, not XML. XML is strong when you have elements referencing other elements in your document structure or DOCTYPE for grammar defs. I might be missing something but I don't get how streaming XML is to be preferred over JSON for XMPP.
If it was actually using XML as a markup language then that would be a reason to stick with it. But XML seems to be complicating things much more than it simplifies.
>> XML remains the best format for representing trees—deep hierarchies of nested data. JSON handles flatter structures well, but good messaging protocols are extensible: extensions can be embedded at different levels and composed together, like Lego bricks. That’s where XML shines.
Anything reasonable you'd be doing with messaging protocol is pretty flat.
What part part, sorry?
That there are issues with feature differences is a problem with the protocol.
Good protocols are opinionated and make concrete choices.
Extensibility typically leads to compatibility issues like the ones you are describing.
I once implemented the presence and basic messaging functionality of XMPP for a web site, using a braindead bridge (that I also wrote; so, the meaningful logic lived in the browser).
Considering I'd never hosted an XMPP daemon and didn't know anything about the protocol (I'd used XMPP clients a little bit, but had never looked at the protocol) and got a server (that part, I didn't write), an auth connector for our website's authentication system (so the daemon would authenticate against that instead), prod-ready and the features I wanted all working smoothly and reliably in maybe three weeks of very part-time work (this'd be, like, 3-4 part-time days with LLMs now, tops, from the same starting point)... seems decent to me? I mean I did direct work with the protocol, didn't just glue together libraries, and it was pretty damn good. Also (and I know browsers seem to be retreating on this front, which sucks) being XML made it very nice to work with in a Web context, since you can just ask the browser to turn ~any XML into a DOM for you, and get a bunch of functionality for free.
What's wrong with it?
These days complicated and/or complex == hard == bad.
Which is understandable when messaging shouldn’t be a hard problem to solve.
> I don't know how it just happens that sending text messages to people can manage to result in specs that are painful to implement.
This is what boggles my mind. We're talking: Text. Over the Internet. It should not be a complex, difficult problem! We've been sending text over the Internet from the moment the Internet went online. So how is it that 10 companies have managed to find 30 different ways to do it, which are all incompatible with each other? You have to TRY to fail this badly.
This is what boggles my mind. We're talking decades of arguments and discussions, dozens of different ways to do it, with dozens of tradeoffs that you could start listing with 20 seconds of thought, and still people just pop in like "why isn't it simple??" as if that's some award winning insight.
Come on, try! Start listing some of the core features of the different protocols/clients and answer your own question as to why it isn't simple! Presence. Chat history while offline. Threading. Mobile devices. Intermittent/mobile/cellular/wifi/CGNAT connectivity. Multiple clients on the same identity and presence. Identity establishing and confirming. Encryption. End-to-end encryption. Backing up and restoring chats, encrypted chats, group chats. Federation. Friends/groups/permissions/trust boundaries. Markup, markdown, formatting, images, emoji, unicode, code blocks. Audio, video, broadcast, multicast. Discovery. Extensibility. Federation. Store-and-forward.
> "We've been sending text over the Internet from the moment the Internet went online"
And IRC has RFC 1459, RFC 2810, RFC 2811, RFC 2812, RFC 2813, RFC 7194 and it does almost none of those things.
And you can still use IRC. But IRC is not good. At this point it's either wilfull ignorance of what people use computers for, or it's indistinguishable from deliberate trolling.
Threading? God, I hate threads in a live chat. Is it never not awkward to use? Slack is the example I'm thinking of, though I'm sure it's shit everywhere else too. In Slack, rather than display the one threaded reply just below (maybe indented), it makes me go to another window, but then someone else is replying non-threaded, so I have to bounce back and forth for the next 10 minutes.
As an anecdotal counter-point, I love threads in a live chat. When used correctly, it means when my idiot friends start having a heated discussion about some random topic while I'm at work, I get one 'ding' from my phone and then never again until I start interacting or they @ me. As opposed to the 427 new notifications I get on my phone about whether Golden Eye was better than Mario Kart 64.
Ultimately the tech is only as good as the humans that use it. If you're chatting with folks that can't understand the difference between a thread or not, that's a meat-bag issue and not anything an RFC can fix.
I completely agree. Threads in Slack should be a great way to make conversations easy to navigate but damn did they fsck up ever part of the UX:
- opening threads in a new window is in some hard to find and hard to remember location
- threads can’t be inlined (like you’ve described). I actually like how they’re in a pane to the right but sometimes that causes issues (as you’ve described). Sometimes it’s just that the view is too narrow.
- there’s no way to shift non-threaded comments into the thread. I’ve lost count of the number of times conversations have happened over multiple threads and/or non-threaded replies.
The way they implemented it is a complete clusterfuck.
I have half-wondered many times whether it would be possible/practical to build a chat system on top of MQTT or similar. Chat seems like mostly a pub/sub problem from where I sit.
Why is it terrible?
or maybe they know about XMPP and prefer IRC. I have always wanted to build one of these on top of IRC. I grew up on IRC and sometimes just the nostalgia of familiar tech is what drives us to build towards it not how good it is.
That’s where we’d expect a mention in the README to play a part.
https://www.bitlbee.org/
That is currently still the differentiator between people producing code and shipping something and people who care and actually have years of experience shipping. At the current state how well you steer the LLM and which questions you ask still matters, maybe in the future "build me an app" will give amazing results, but right now you still need to know what you are doing.
In this case, it's just a waste of tokens as something not very unique was generated that doesn't have a real use case, or solves anyone's problem. As with many AI generated projects, I'm willing to bet that OP themselves will not using it any more in a month.
You guys are way too critical and about the wrong things. "Waste of tokens"? Perhaps the author know of the tradeoffs and just wanted to build something.
How you went from seeing somebody's post to deciding they don't care is a pretty big jump, one that isn't warranted. This kind of post seems like the old but now more elaborated form of hating on something that you don't even know what it is.
It just isn't fair to the work that has been put in.
I agree. This is one thing that's gotten me about the new wave of poo-pooing projects made since coding agents have gotten big. Five years ago it was 'just build something!'. I understand the market is more saturated now so building something isn't as valuable as it used to be, but it still shouldn't be treated as a negative. I don't know anything about the creator's background, but even if this project is ultimately redundant and unnecessary, it's still pretty cool that they built something.
Because this isn't an example of "just build something" it's an example of "just order something from grubhub"
If telling people “build me an app” with no knowledge of technology worked, project managers would be good at their jobs already
I've done this a few times. If I had to gesture at the root cause I'd say: LLMs lack curiosity and are trained to complete assignments, not to question their validity, so both the research phase and the questioning of intent tend to get short shrift prior to building anything.
I'm not sure I agree. Seems to me to depend a lot on how one interacts with an LLM.
If you tell it to "build me an application X to do Y using programming language Z", it will comply.
If you interactively ask "I need X, can you suggest some options? use an existing product ? or build something using a library ? or build entirely myself ? can you suggest alternatives with pro's and con's ? Anything I should think of before deciding ?" then you will get an entirely different answer.
Maybe we need additional modes, aside from thinking mode, agentic mode etc. we need e.g. "sparring mode" ? or does such a thing already exist ?
Matt Pocock's skills library includes "grill-me" which is precisely this. It's excellent.
Agreed, you have to ask them to look for prior work. But then they will do a good job finding it in most cases. I think it’s partly their universal tunnel vision and partly sycophancy.
Specifically it's important to ask for more than a few examples.
> I half expect the LLM would have mentioned that at some point, but the repository does not.
The repository does not even mention this software is LLM-generated.
I wonder when/if the collective realization will land that LLM's failure to retain provenance during training means the code it generates exists in an indeterminate state of public domain / IP trap.
The AGENTS.md file in the repo with its contents is a pretty clear giveaway though.
LLMs could be asked to research extra projects.
It's valid that someone just wanted to build something like this for their own purposes.
But it's not just for their own purposes. They shared it with the public which means they will get feedback, good and bad. No one forced them to post to HN.
Agreed, and the vast majority of personal projects that are released as open source don't get attention, traction, or contributions.
In this case, someone shipped something that works for their use case, which is a great deal more than talking.
Sure but it will still be scrutinized and compared. I don't see the issue.
Maybe some people just ship and post and don't look back in terms of scrutiny / comparing.
There's lots of approaches out there that say if you don't ship something embarrassing, it's too late.
I think this is the main point I was trying to make and our exchange helped me bring it forward, thanks.
”There’s a specific downside to LLM-based development that people can get neck deep in a new project without interfacing with the existing work in the field.”
That’s because it’s a paradigm shift, not meant to be planned or organized by humans. The prior paradigm was that there were subject-matter experts who were human. The new paradigm is that the only author of all code is to be a machine. This new paradigm requires removing all prior authors (who happen to be human), and removing them simply requires overcrowding their output such that it’s not locatable or trustworthy.
Sit back, grab some popcorn.
XMPP is massively bloated protocol that is badly managed for last 2 decades
And what are you suggesting as a non-bloated and better managed alternative? :)
Maybe Matrix?
Synapse basically brought my vps to its knees with like 5 users. I would not call it light.
Is Matrix really non-bloated? My impression has been that it works more reliably, but isn’t considered lean at all.
Ya, ive stopped interacting with all new projects, the amount of slop written by people who have no business remotely putting something out there for others to use is untenable.
The system has broken under the noise.
I'm searching for a simple, self-hosted chat system that allows me to host unlimited bots and send messages (with notifications) to my iPhone easily.
I currently have an XMPP server running, which is fine, I guess. I've looked into chatMail. I have considered hosting an IRC server.
At this point, I'm considering just setting up my own NTFY server and writing a chat client for it.
Does anyone have any ideas for what the lightest weight, simplest solution might be?
Curious, why the interest in building your own ntfy client (instead of using the one already built/available)?
By the way, for years now, i have used a little python (cli) app to send messages into a private matrix room. This room has 2 people in it: me and a bot. (The bot is simply the matrix account that sends the messages.) I went this route because i did not want to install another app - such as ntfy - on my mobile simply to receive server notifications, etc. And, since i had ben on matrix already chatting with friends, it happened to be a low-effort approach. To clarify, i am no longer self-hosting my own synapse home server...so i'm simply piggybacking on the matrix.org one...but that has helped at least unify my notifications and comms to a good degree.
Now, if you were to aske me to implement my "notifications system" via a different approach, funny enough, i would gauge between XMPP and ntfy to see which is lighterweight....because i believe them to both be quite light on the need for resources. I don't know which is lighter (sorry, i know i'm not answering your question). however, i *think* the ntfy server is less hasle....Only because xmpp takes a little more work to setup...at least that was my experience a couple of years ago. I cast no shade on either xmpp nor ntfy - because to me boh are awesome...i' m merely saying that resources needs aside, i wonder if ntfy might need less time from YOU.
I'd build my own NTFY client to enable two-way chat. It's not something I'm seriously considering, because it's obviously a bad idea, but that's about where I am.
you should look at celld
>I'm searching for a simple, self-hosted chat system that allows me to host unlimited bots and send messages (with notifications) to my iPhone easily.
I do this with IRC. I used to have a blog post up about it, back when blogs weren't just training data providers for LLMs, but I'm afraid it's been deleted.
tl;dr IRC works great, but you need to wire up push notifications. For android I used pushbullet, for iOS I am still evaluating the alternatives but pushover looks okay. IRC itself is nice because the protocol is shit simple, to the point that if you can type fast enough to respond to the pingpongs you can cosplay as a bot using nothing but telnet. As a text protocol it's easy to inspect and doesn't require binary decoding of packets for debugging.
I recommend the Ergo ircd. It ships as a single binary (golang), can support tens of thousands of clients, is at the forefront of the ircv3 modernization initiative, and is under active development.
Interesting. Why not use an open link network? These will result in around the same state. Spam _will_ be an issue here, in general. And with a different backend you're going to struggle to use preexisting tooling to handle it.
Using & channels is fun, I'd be interested to see how many bots and clients fall over dead when faced with that particular bit of IRC history.
IRC linking is a mess, in part due to the spanning tree requirement. Also the server to server protocol in the RFC is spoken by zero servers so you have to pick which unofficial protocol you like best. May as well invent your own that actually fits your use case.
Except that now there's yet another standard, which nothing but this thing speaks. TS6 is very well used at the moment as everything in the Charybdis line speaks it, etc.
This one works differently, which is enough reason to make it different.
I had a pseudo-plan for something like this but for reasons related to privacy I had two types of chatrooms:
* regular chats, existing only on the server, no leaking the chat transcript except via users, but never via server to server.
* chambers: a global chatroom, located at a server, maybe with a MQTT anyone could subscribe to.
This never left the early planning stages though, but I thought the segregation between federated chatrooms and regular chatrooms was of interest to keep in sync with IRC open but closed nature of chats.
That seems to be similar on this one. &room is local, while #room is federated.
Did not see that. Thanks for pointing that out! Will re-read the page.
I quite like this, but it feels like spam could quickly become a concern.
That is always a concern, yes. You can block users, and entire instances, that spam.
Vibecoded?
clearly.
As someone very critical of mindlessly vibecoded software, I've worked with James (@prologic) years ago and very much respect his craft. AI assisted or not, I wouldn't describe him as a vibecoder. He's a great engineer.
Instead of a separate federated network for a single app, why not implement the app on top of a pre-existing federated network, so that nobody has to create yet another account? Like, on top of atproto or activitypods or...?
Or matrix
A lot of the design reads like almost but not quite ActivityPub.
This is pretty much just a reinvented Matrix protocol. Nearly identical.
To be fair, if i recall correctly, some of the folks who started up matrix protocol (and associated project) came from/and loved the IRC world (e.g. the rooms concept, etc.)...so it stands to reason that matrix and irc have at least some *conceptual* similarities, and hence other stuff that spins off from or leverages IRC may also appear similar to matrix.
Give me a day and I'll implement it in my agent-oriented IRC server https://github.com/marlinski/airc !
Cool. I did something like that so agents and people could communicate with a Jira instance using Websocket-aware IRC and a browser-based client based on superchat.
Put the issue context in a #channel and invite external parties/entities, bring the results back to the work.
I have tested it but the complexity didn't justify the results in my experiment. I couldn't get it accepted into Atlassian Marketplace, either.
Both cool and worryingly convenient for the botswarms we've been warned of...
Besides not having C runtime linked (which is great for scratch/distroless Docker images) - any other advantage of using a pure Go SQLite instead of the C binding?
> "Scrollback that follows you rather than your client"
This form of LLM speak annoys the absolute shit out of me. It's a bot trying to emulate the worst reassuring sales copy you could imagine.
This is exactly what i was thinking about for the last few weeks (but am too stupid to implement myself).
It’s like email just for IM…
You're in luck: Some people already thought of this, submitted an RFC to the IETF and implemented it. It goes by the name XMPP.
Many earth rotations ago, I was IMing through an mIRC bot because I refused to install MSN. This kind of reminds me of that too.
I used Bitlbee until not that many earth rotations ago - it's an IRC daemon that provides bridging of various other chat protocols. It worked way better than you would expect, especially back in the day when you could use XMPP to connect to the various proprietary chat services.
I didn't stop using it until I stopped caring about those chat services.
These days I run a Matrix home server with bridges to Slack and WhatsApp so I can use fewer chat clients.
I do wish the Matrix client situation was less messy.
is there a list of public instances one could join somewhere? I kind of want to dip into irc again. it was peaceful.
The idea (what's encouraged) is that's so simple you can run it yourself. If you email me (email on profile), I can create an account for you on mine.
Parlay?
IRC is popular because it’s bad. I don’t believe there will ever be a replacement for this reason.
But isn't IRC already federated?
You can link servers (Libera, for example, has many), but they are not federated, nor decentralised (services are used for registration, etc.).
Closed federations. Only trusted servers
The connection to git.mills.io was interrupted while the page was loading.
Sorry, I am sure it is the HN effect.
lol, so XMPP, but instead of XML, it is IRC. Alright, I dig it.
Instead of TCP+XML it's HTTP+JSON
The IRC is just a front end. Could use any front end
I see. So more similar to ActivityPub.
I wish them the best of luck. Federated systems struggle to achieve critical mass, and if they do, large corporations tend to snuff it out. I distinctly remember Google committing to The Microsoft EEE strategy with XMPP. We've left behind lots of great ideas in the last 20 years
Google helped make XMPP quite popular for awhile, then killed their own product and left us with a lot of useful specs they had contributed. So I'm not sure I'd call that EEE
I recall this bait and switch. It was called Slack
Slack is pretty centralized, and also closed. You can't self-host at all.
Slack is a *different* bait and switch.
Different use case no? I thought slack was enterprise focused from day one.