For those that use LLMs in a similar manner to a search engine the Anthropic team (unlike Meta AI and Google Gemini) have made it easy to use Claude from right within your browser. In Firefox add a bookmark with the following values:
Also in Firefox Dev Edition settings (in the Firefox Labs section) you can turn on their AI chatbot feature which gives you the option to send the selected text to Claude or other chatbots and opens their response in a sidebar.
Also, if you prefer using AI assistants as desktop apps that feel separate to the browser and you can open directly, you can download Progressive Web Apps for Firefox [1] and create a separate shortcut for a sandboxed profile of Firefox with less visual clutter that open the assistant directly.
Yes, I'm aware of @gemini on Chrome. If you look at how it's implemented they are purposely stashing the query into a HTTP header so it's not possible to use a simple bookmark to launch it.
Query parameters are so tired. All my homies embed their queries into the password portion of a basic auth header. Keeps your queries super secure from spies and competing browsers.
This doesn't seem to work for me. I'm getting "%s" as the initial prompt, even though the text is properly displaying in the URL. Almost seems like an issue on the Claude side.
I only have two minor complaints with AnythingLLM:
1. No light mode (why does everyone want dark mode, it sucks for anyone older than 30)
2. It's electron
- Customization: You can add any OpenAI compatible API (including X.ai) as a model, and just edit the url/model id
- Chat with files: Not as complete as complete as a RAG solution for now, but we feature simplicity. Basically you can drag and drop PDF files onto a session or add files/folders to a model (like custom GPT) to start chatting.
- And, yes we support light mode! And several light mode code themes as well
No, it's related to Astigmatism. The eye lens gets distorted. Bright light shrinks the pupil, meaning the light goes through less of your lens and hence is not as distorted.
You said: “ 1. No light mode (why does everyone want dark mode, it sucks for anyone older than 30)”
I’ve had astigmatism literally my entire life and, once again, have loved dark mode anywhere it’s been offered (except my Kindle, which just looks weird).
Every programmer over the age of 30 that I know uses dark mode exclusively, and there’s a reason companies offer it: it is widely used and demanded.
Acting everybody in the demographic of people over 30 feels the way you feel about it on account of their age is just laughable.
That's what the "Workspaces" are. Create a workspace, go into the settings then "Chat Settings" and configure the Prompt. There are more configurations available.
I have an astigmatism and I prefer light mode because of it. Dark mode causes the letters to jump more and the optometrist agreed light mode is better for people with astigmatism and its a common issue.
Some haven't been updated for a while. Try the "software-evaluation.md" prompt, it's one of my best. I use it often. Use the bin/gp script to get them into your clipboard.
For me the window is laggy and takes a few seconds to resize or move. It's also not reliable when prompting sometimes it just sits there with no output and I have to quit and restart. I've since uninstalled the desktop app and I'm happier with the "Add to Dock" web version.
That’s not entirely true. GDPR covers all forms of storing and processing PII. Wether that is a website or a desktop application, you will need to get opt in from your user to share PII with third parties.
This is not true - both GDPR & ePrivacy rules apply to any software, none of it is specific to websites (although obviously that's where it's most easily abused, and where most of the attention is).
Oh I agree, but this is definitely not it, I don't understand why Anthropic released this, other than squatting the same keyboard shortcut the ChatGPT app uses, with no ability to change it :P.
Chat UIs are really the perfect candidate for a localfirst PWA. I have no idea why they are being built as either a full desktop app or UI where nothing is stored on device.
You’d think with all the funding these companies have they’d make better technical decisions
I do see value, but the Claude app is just a webview with a keyboard shortcut, I don't even think it has voice dictation, I uninstalled it immediately.
Hmm yes, I don't see any issues giving a corporate controlled model with the operational precision of a coin toss full access to my entire system. There is absolutely nothing that could ever go wrong.
Nah but seriously, can we start a counter of how many times a chatbot agent has deleted someone's system32 because it was trained on data of the average tech forum?
I don't think they could release computer use for $20/month plan. It seems super expensive and you could easily involve Claude for hours of work for manual tasks like data entry or data cleaning.
yeah, that's what it seems like. As of today, there's no additional features I notice compared to having a Safari-based desktop version of this (which I uninstalled to install this one).
It's disappointing that these companies are perfectly happy to use it for their own wealth creation strategy, but have no interest in giving back and furthering the ecosystem.
The whole electron app uses less RAM than the page opened as a PWA? That’s surprising to me, how come? I somehow thought Safari was more memory efficient than Chromium
I suppose I use the PWA differently than most people? I just leave it open all the time, so I don't care how long it takes to start or how much data it takes to download to boot.
It’s such a poorly designed PWA. They really failed to take advantage of any of the browsers storage mechanisms. Like why not put conversation history in indexedDB?
This caused me a problem because I signed up with my personal email address, and then wanted a separate account for work but I only have one phone number and they won't let me use the same one on two accounts. I don't really care about giving them my phone number (it's out there everywhere anyway) but this has really reduced my ability to use their product.
Is this for the paid version? I'm using the free version, and it hasn't asked me for a phone number, but if the paid version requires a phone number, my reaction would be the same as yours: sorry, no.
In the age of eSIM, just order a free eSIM with zero credit, never charge it, never have any cost, and only enable it on your phone when a service like Anthropocene wants to send you some (completely unsecured) SMS code.
They add the same Option+Space (on macOS) shortcut that ChatGPT has, but the app feels a bit like a quick build overall, as it's just a web container. Hopefully they'll add some nice features that wouldn't work in a browser such as asking about what's on your screen or a shortcut to be able to speak questions and get a voice response.
Indeed not everyone is able to write Web sites with good performance across multiple browsers, that is 3l1t3 skill, better ship Chrome alongside the application.
It's more like 400MB due to how they're packaged. And when you could have it run with 10x to 20x less RAM, it just scream "we don't care about you". Not everyone have 64GB RAM and 22 cores of CPU in their laptop.
It's like using a semi-truck (with the trailer attached) to go buy groceries 100 meters away.
I'm unsure where I'm supposed to look. In the Task Manager, the app I created (Video Hub App), when running, is routinely between 40MB and 120MB under the "Memory" column. Are you saying there is RAM use tucked away under different row than the app name?
It's not for the storage or ram space. It's for being inefficient which always making me doubt about the company culture. If you want native capabilities so much, just build a proper software for the platform. If VLC and Calibre can do cross platform, you can do it too.
How is 400MB trivial? I run more than 2 programs at once, and this adds up. Imagine if every trivial process in your OS used 400MB on startup. And what pisses me off, is when I have to close resource heavy electron applications (signal, vs code) running in the background to free up RAM.
More like 400MB, and the problem isn't that I can't spare it. The problem is that I can only spare 400 MB a few times at most before it starts becoming an issue.
I have stuff to do on my computer, stuff that often involves multiple programs (one of them almost always being a browser!). Then there's bunch of support stuff that I want, or have to, keep around. A music player, an IM, a mail client, etc. When everything gets packaged as Electron apps, even those tiny utilities add up quickly, and everything slows. down. Or worse, gets starved for RAM and starts stuttering (Windows) or just gets OOM-killed (Linux).
And I say that as a dev/techie, who has plenty of RAM. Non-tech people tend to have computers that are memory-starved by default. Building trivial stuff as Electron apps, where native would be 100x less resource-intensive, is just peak developer laziness/selfishness.
Kids these days insist that they need the "app" of something, even when it's literally the exact same thing as the webpage.
Between this phenomenon and the refusal of many to close tabs in their browsers, I can't help but wonder if a huge percentage of the populace have never heard of or used Bookmarks on their web browsers before.
I'm not a kid and I know about bookmarks, but I prefer using apps to apps in my browser. Electron apps I have less interest in, but I really don't like working in something like Figma (which is in the top of it's class) in the browser.
In-browser apps do all sorts of clunky things like going back a page because I'm trying to scroll right or left or hotkeys not working. Native apps feel so much better to use and I can close my browser and stay focused on the task.
I think it’s less about not knowing what a bookmark is and more about just having a more direct way to access it, where it’s more high-level than just in a tab. There’s just something mentally different about opening a program from your dock as opposed to from a tab, even if they’re both basically a website.
This line of reasoning was why I was hopeful that PWAs on the desktop would become first-class citizens, as far back as Mozilla Prism[1] in the late 2000s. They never really did seem to take off.
It used to be fairly common to have web-page shortcuts back in the mid-00's. Lots of apps would add spam shortcuts to your desktop that would just open your web browser to a web page where you were encouraged to buy something/donate/start a free trial.
Anyways I think that capability still exists on all modern OSes.
I swear if all the ai chat apps started as fat apps and then Claude just released the first port to the web, I would be reading this exact same comment on hacker news except complaining that there was nothing wrong with the fat app.
I like to ALT-TAB instead of CTRL-TAB because I always have fewer applications open than I do sites. I don't have dozens and dozens of tabs always open either, but always more than my programs.
I was thinking more about people who use Tree Style Tabs and have nested Tab folders and such when I wrote that part. It's simple enough to save an entire session as a bookmark folder, to open an entire folder of bookmarks as new tabs (rather than needing to hibernate tabs to save ram while still keeping them "open"), and to search through bookmarks ad-hoc. Bookmark folders can be infinitely nested, categorized, ordered by when you were looking (though searching through your browser history also works for that), combined, copied, pasted, moved, put into or removed from the bookmark toolbar.
It's an incredibly flexible system that allows for a massive variety of workflows and it feels like people just keep finding ways to recreate it, but worse: less platform-independent, eating more system resources, dependent on third-party plugins, or (in the case of using an app for everything) eating up orders of magnitude more storage space at rest.
The web went to shit a long time ago; you can't rely on being able to bookmark a site and then go back to where you were. Half the sites are infinite scroll, or dynamically generated pages, or SPAs[1], or some other ephemeral invention du jour. Keeping a tab open gives you some chance to return to where you were for some time; bookmarks are just giving up.
--
[0] - At least until the browser decides to screw with you and unloads the tabs you needed. Firefox on Android is particularly aggressive at that, which incidentally makes PWAs unusable, too.
[1] - You're lucky when those let you make a bookmark that won't drop you back to index page when loaded.
Now that's an interesting perspective I hadn't considered.
I suppose I actively avoid most infinite scroll webpages because to me they feel like they're meant for "consumption" and visual appeal rather than actual usefulness and I tend to think of webpages as tools, either for information or socializing. As a dev I'm certainly biased towards sites like Github or news sites, which are both meticulous at sending you to the page you were linked to specifically rather than redirecting you elsewhere.
I do tend to keep a few tabs open at any given time, but only when I'm in the middle of reading something like an article or blog post. I try to close down my tabs to something like 5-8 max weekly, if not daily. Something about closing out the mental context and making a conscious decision about whether I really want to finish reading something or if it was boring me feels very freeing once I've decided to close the tab is deeply satisfying to me.
I have ChatGPT app and Claude and Gemini PWAs on my dock. I gravitate towards using ChatGPT much more. It's just faster vs launching Chromium for each small query.
Please, tell me where I said there's only one way to do things.
My question was not meant as a critique of people who know what they're doing. I'm not talking about engineers or sys-admins like the typical HN crowd, I'm talking about the population at large. The average person understands their devices just enough to get them to work the way they want. It stands to reason that a fairly large percentage of the modern population, used to using apps for every individual service, may never have had a use for browser bookmarks and may therefore never have even attempted to use them.
I have a few friends who work as professors and tell me stories about how their students don't know how to navigate a filesystem in Windows. Why would a person like that, for instance, have bothered to learn what a bookmark is vs just keeping a tab open?
There isn't one, it's just an Electron app that opens slower than the webpage and flashes white before it appears.
You'd think, a company that has such powerful AI and no restrictions on requests to it would be able to bash out truly native apps no rather than something that looks quite shoddily made, like making your electron app not flash white when in dark mode is amateur level stuff.
Tangentially, what are people using to have a unified interface to query multiple LLMs? Juggling multiple apps or websites for each provider gets annoying. I was using openrouter but the chat history is only stored in browser, so I'm looking for something else now. I tried open-webui[1] and that looks pretty good but I wonder if there are better alternatives.
Give https://recurse.chat/ a try - I'm the developer. One particular advantage over alternative apps is importing ChatGPT history and speed of the app, including full-text search. You can import your thousands of conversations and every chat loads instantly.
Regarding chat history, I've been thinking that people should have ownership over chat history. We are migrating the chats towards SQLite, so your data is going to be a timeless format like files* - SQLite has long term support through year 2050. https://www.sqlite.org/lts.html
I'm a very, very happy user of TypingMind. It's not free, but it's worth the (one-time) license cost in my opinion. It has support for multiple providers, plugins, "agents" (pre-set templates), chat grouping, import/export, automatic syncing across devices, etc.
There are a ton of alternatives if you search around. I created my own because I wanted something keyboard centric, and use OpenRouter with it: https://github.com/iansinnott/prompta. They all accomplish roughly the same thing though.
I agree, I'm also firmly convinced it's the best. Until a month ago it has some bugs I disagreed with some design choices, but since 1-2 weeks ago it's the absolute best I've ever used and have easily convinced others to use it.
Try MultiGPT which is a Chrome extension that consolidates several AI chatbots, including ChatGPT, Bing Chat, Bard, Claude, and more, into a single interfacee
I was surprised that simple things like sharing a chat in Claude doesn't seem to be there. I read somewhere that it was a feature that was disabled later.
It's annoying, I'd really love to export my chats. Any Anthropic PMs reading this, as a paying subscriber I command you to prioritise this feature, on pain of failed training runs
I wish I knew of a good way to voice chat with Claude, and unfortunately, the new mobile app isn’t it. The voice recognition is terrible. Does it just use iOS’s built-in text-to-speech? It needs to be as good as Whisper to be useful.
(I hope this thread doesn't go anywhere. I just feel like complaining.)
There's been some inflection point where the desktop version is now often worse than the web app. And you can't blame web tech because the web app in the browser is usually quite good.
WhatsApp on macOS has catastrophic bugs for years. I regularly can't focus the input field and have to relaunch it. Until recently, its image modal would regularly bug out and leave behind a global dimmed window that eats all input so you can't even close it.
Poe's desktop app is less polished than poe.com. Off the top of my head, its desktop app scrollbar becomes so tiny that you can't even click and drag it.
ChatGPT's desktop app is simply worse than its web app or was when I stopped using it months ago for that reason.
Youtube Music is another pointless desktop app since it's just the web app inside a web view, though at least it embraces that.
It's like the only thing going for these desktop apps is that you can alt-tab to them individually. It's time to come up with a trivial one-click first-class OS-level way to turn websites into apps and put all these sad desktop apps out of their misery.
Oh well, it's come at a good time because I've mostly soured to the idea of having to install software at all. 90% can be run in a sandboxed browser away from my file system.
Lots of people are correctly suggesting this could function just as equally well as a PWA, but I will point out that (all things being equal) there are some valid use-cases when strictly choosing between Electron vs PWA that a LOT OF PEOPLE seem to be oblivious of:
- Electron apps don’t rely on a web server to function, so as long as you archive the app’s installer, you can install it on future computers even if the original website goes down. This assumes that the PWA/Electron App in question doesn't need server features, for example Photopea, Obsidian, etc.
- A bit easier to block network traffic to/from a dedicated application from a privacy perspective. Still can be done on a PWA, but it's trickier to set up.
It would be nice to see a de-coupling of the backend (as a service, deamon, etc.) so that we could choose the desired frontend of our choice (net win from a UI/UX, accessibility perspective, etc.), but it's been a long time since I've encountered a consumer level application that did this.
I mean, re: point 1, we're talking about an app that calls an API for every interaction already, so that doesn't seem super relevant. If Claude is down, you can't use it whether it's through a PWA or an Electron app.
> "Lots of people are correctly suggesting this could function just as equally well as a PWA, but I will point out that (all things being equal) there are SOME VALID USE-CASES"
YES - I am speaking GENERALLY. That's why I literally called that out in my comment that this particular use-case (Claude Desktop) could easily be done with a PWA.
Gotcha. Sorry, I guess that was unclear to me based on your phrasing. When you said "some valid use-cases", I thought perhaps you meant (for this app) and not (for Electron apps in general) as the implied next clause.
No worries, yeah I just meant that if a dev has no intention of building a native app and is only deciding between either a PWA or an Electron/Tauri App, then there are some advantages to the latter.
I'd say for the case of Tauri it's doubly so. You can have natively compiled Rust code for especially compute-intensive functions, rather than having it running via WASM.
I like LibreChat a great deal, but its pretty heavy-weight (over a more bundled solution like Jan) in terms of spinning up a bunch of docker containers (meilisearch, mongo, vector db, etc.)
This LLM hype is out of control, they literally bullshit on any novel input (admittedly there's not a lot of this) and are wrong 15-20% of the time for all inputs
This isn't a "we'll just add more context", "we'll just add more instructions and params", "our multi-head transformers will auto-attention the tits off your query" etc - this is search based on probability maps. It is structurally unable to be intelligent, no matter how many models you chain together and how much compute you throw at it. But it can sure mimic that shit, which is why LPs are going to lose their shirts and GPs are going to struggle to raise follow on funds.
For those that use LLMs in a similar manner to a search engine the Anthropic team (unlike Meta AI and Google Gemini) have made it easy to use Claude from right within your browser. In Firefox add a bookmark with the following values:
- Name: Claude AI
- URL: https://claude.ai/new?q=%s
- Keyword: ai (or whatever you'd like to use as a shortcut)
Now from the URL bar you can type: ai <my question>
For ChatGPT use https://chatgpt.com/?q=%s as the URL and for Microsoft Copilot use https://www.bing.com/search?showconv=1&sendquery=1&q=%s.
Also in Firefox Dev Edition settings (in the Firefox Labs section) you can turn on their AI chatbot feature which gives you the option to send the selected text to Claude or other chatbots and opens their response in a sidebar.
https://blog.nightly.mozilla.org/2024/06/24/experimenting-wi...
Also, if you prefer using AI assistants as desktop apps that feel separate to the browser and you can open directly, you can download Progressive Web Apps for Firefox [1] and create a separate shortcut for a sandboxed profile of Firefox with less visual clutter that open the assistant directly.
[1] https://addons.mozilla.org/en-GB/firefox/addon/pwas-for-fire...
This is not unique to FireFox. This is also possible in Chrome (Settings -> Search Engine) and looks like this:
https://imgur.com/a/qxap6iI
Setup the way I have in the image you can simply type @claude and your search
Worth noting that Gemini also offers a URL shortcut for searching Gemini via @gemini within the Chrome browser.
Yes, I'm aware of @gemini on Chrome. If you look at how it's implemented they are purposely stashing the query into a HTTP header so it's not possible to use a simple bookmark to launch it.
Why would they intentionally limit that?
Query parameters are so tired. All my homies embed their queries into the password portion of a basic auth header. Keeps your queries super secure from spies and competing browsers.
Are you on Linux by chance? Your font doesn’t look like it has anti aliasing which seems like a common Linux thing
It's not a common Linux thing, all major distros ship with antialiasing configs ootb. The person just disabled it.
BTW you can use s[1]
you can combine it with your shell to call claude, chatgpt
sc expands to s --provider claude sgpt expands to s --provider chatgpt
[1] https://github.com/zquestz/s
This doesn't seem to work for me. I'm getting "%s" as the initial prompt, even though the text is properly displaying in the URL. Almost seems like an issue on the Claude side.
worked fine for me on google chrome by setting it up as a search engine/site search.
This is exactly how Vimium`s custom search works
Entries like these should be submitted to the MyCroft search database.
I'm using AnythingLLM for this now. https://anythingllm.com/
MIT License on GitHub
This week it got the X.ai API added which is what I'm using. Grok is not as good as Claude or ChatGPT yet, but don't bet against Elon.
https://x.com/ServeTheHome/status/1850917031421399543
https://x.com/elonmusk/status/1830650370336473253
It's the desktop solution I've been looking for. If you have a heap of prompts, you can create a Workspace for each. Super easy.
Here's my prompts if anyone is interested:
https://github.com/grantcarthew/notes/tree/main/Prompts
I only have two minor complaints with AnythingLLM: 1. No light mode (why does everyone want dark mode, it sucks for anyone older than 30) 2. It's electron
Edit: replaced cluster link with source (I think)
Shameless plug: If you are on a Mac, check out RecurseChat: https://recurse.chat/
A few outstanding features:
- Fast: Import ChatGPT history, loads thousands of conversations at once.
- Floating Chat: spotlight / ChatGPT desktop app like floating window.
- Customization: You can add any OpenAI compatible API (including X.ai) as a model, and just edit the url/model id
- Chat with files: Not as complete as complete as a RAG solution for now, but we feature simplicity. Basically you can drag and drop PDF files onto a session or add files/folders to a model (like custom GPT) to start chatting.
- And, yes we support light mode! And several light mode code themes as well
I’m nearly 40 and I can’t use light mode anymore.
Judging from how many older coders use it I think it’s just you, mate.
No, it's related to Astigmatism. The eye lens gets distorted. Bright light shrinks the pupil, meaning the light goes through less of your lens and hence is not as distorted.
And?
You said: “ 1. No light mode (why does everyone want dark mode, it sucks for anyone older than 30)”
I’ve had astigmatism literally my entire life and, once again, have loved dark mode anywhere it’s been offered (except my Kindle, which just looks weird).
Every programmer over the age of 30 that I know uses dark mode exclusively, and there’s a reason companies offer it: it is widely used and demanded.
Acting everybody in the demographic of people over 30 feels the way you feel about it on account of their age is just laughable.
OK, you win.
Does it allow you to create custom agents like OpenAIs GPTs? I can't find it in the documentation.
That's what the "Workspaces" are. Create a workspace, go into the settings then "Chat Settings" and configure the Prompt. There are more configurations available.
I will have a look. Thanks!
You have to be psychotic to not use dark mode and prefer light.
I have an astigmatism and I prefer light mode because of it. Dark mode causes the letters to jump more and the optometrist agreed light mode is better for people with astigmatism and its a common issue.
You've gotta be nuts not to be psychotic in a world like this one.
haha, you guys.
These prompts are terrific, thank you!
Some haven't been updated for a while. Try the "software-evaluation.md" prompt, it's one of my best. I use it often. Use the bin/gp script to get them into your clipboard.
Is there something like AnythingLLM that is a Mac native application?
https://msty.app/
(Note: it has first class support for Obsidian.md vaults.)
The subject of the thread is "Claude for Desktop".
AnythingLLM supports Claude out of the box.
but it doesn't say "only Claude for desktop", which makes GP's comment relevant.
if your "desktop app" gives me a cookie consent banner, i'm going to assume it's a pretty lazy effort to have a desktop app.
i can click the "install page as app" button in chrome myself, thanks.
On the flip side, the ChatGPT desktop app is native on Mac and it's buggy and feature-incomplete compared to the web version.
100% - amazing this crap get released when "Add to Dock" is far better - zoom, copy/paste, search, font size - all works just like in normal browser.
Amazingly both implementations burn 100% CPU when rendering text, which is just bonkers.
Claude's web app crashes after few days and uses 30% on idle.
When rendering text, what would you expect CPU to be? For however long it takes to render, you’d expect it to max out.
What are some of the bugs? I’ve not noticed any. The feature lag is real, still no canvas.
For me the window is laggy and takes a few seconds to resize or move. It's also not reliable when prompting sometimes it just sits there with no output and I have to quit and restart. I've since uninstalled the desktop app and I'm happier with the "Add to Dock" web version.
Also search is unavailable, regarding missing features.
Have you updated? Mine has search, in 1.2024.289.
And doesn't work on Intel Macs. :-/
A desktop app that renders a website can still be useful if it exposes native APIs to the website.
does this do that?
Uuuh.. so take all my data, as long as it's native?
Desktop apps can store what they like. There is no EU law preventing desktop apps from storing anything, unlike websites.
That’s not entirely true. GDPR covers all forms of storing and processing PII. Wether that is a website or a desktop application, you will need to get opt in from your user to share PII with third parties.
This is not true - both GDPR & ePrivacy rules apply to any software, none of it is specific to websites (although obviously that's where it's most easily abused, and where most of the attention is).
Which part of General Data Protection Regulation in GDPR don't you understand?
I only use the chatgpt desktop app for quick throwaway interactions.
For anything more involved, I want a tab and an URL. Perfect use case for a webapp, native is pointless for this, especially if it's just a webview.
Native desktop apps give these companies access to more system-level functions which will be super useful for...
- Voice dictation / interactions
- Agents accessing other applications and controlling your desktop
- Agents performing background tasks like continuous monitoring, periodic data processing, or ongoing analysis
I would expect to see these kind of features start to take off next year
Oh I agree, but this is definitely not it, I don't understand why Anthropic released this, other than squatting the same keyboard shortcut the ChatGPT app uses, with no ability to change it :P.
Chat UIs are really the perfect candidate for a localfirst PWA. I have no idea why they are being built as either a full desktop app or UI where nothing is stored on device.
You’d think with all the funding these companies have they’d make better technical decisions
The shortcut can be changed. Open the app settings from the menu bar, or the usual Command-, shortcut, and it's the only setting they expose.
By not it, do you mean those features aren’t available yet, or you don’t see the value in them?
I do see value, but the Claude app is just a webview with a keyboard shortcut, I don't even think it has voice dictation, I uninstalled it immediately.
You're not wrong, but I'm willing to guess the biggest reason is: Analytics
You can also do a filesystem scan pretty easily. Both for useful features, as well as spying.
That requires user to accept permissions (which lately macos giving so many popups made me fatigued and accept everything)
Also data mining! :)
> - Agents performing background tasks like continuous monitoring, periodic data processing, or ongoing analysis
Who said that this was to help the end user!
Hmm yes, I don't see any issues giving a corporate controlled model with the operational precision of a coin toss full access to my entire system. There is absolutely nothing that could ever go wrong.
Nah but seriously, can we start a counter of how many times a chatbot agent has deleted someone's system32 because it was trained on data of the average tech forum?
> For anything more involved, I want a tab and an URL.
Why, exactly?
Probably for bookmarking or working iteratively over different sessions on the interaction and allowing more than one session at a time.
Thanks. After using the Claude app, I noticed several things lacking. The ChatGPT Mac app is significantly better, IMHO.
For some reason, I expected the same from Anthropic, but the Claude app, in its current state, is quite inferior.
Also works really nicely with your shared-cloud-tab provider of choice. If I have a tab open {anywhere} I know it's accessible {everywhere}.
Or sharing.
Can share links from the native ChatGPT app.
And where does it link to?
https://chatgpt.com/share/<chat GUID>
This will open in the native app if it's installed, and on the Web if not. Basically, it works like any modern app link.
Perhaps a gateway to run computer use locally?
https://www.anthropic.com/news/3-5-models-and-computer-use
I don't think they could release computer use for $20/month plan. It seems super expensive and you could easily involve Claude for hours of work for manual tasks like data entry or data cleaning.
My experiments with computer use clocked in at $15/hour
I was hoping that's what this was, sadly it's not, but defn opening the door for it in the future
yeah, that's what it seems like. As of today, there's no additional features I notice compared to having a Safari-based desktop version of this (which I uninstalled to install this one).
No Linux, I should have expected that. Web works fine, but damn no Linux.
There's an old saying: "Judge a tree by its fruits". Be sure to remember this the next time Anthropic starts talking a big game about "our values".
It's disappointing that these companies are perfectly happy to use it for their own wealth creation strategy, but have no interest in giving back and furthering the ecosystem.
On macOS, why have an electron app when you can add the webpage to the dock as a PWA?
Here is the comparison:
Claude Electron app vs Claude Safari PWA
Pros:
* Can be opened with a customizable shortcut
* Takes less RAM (400 MiB < 600 MiB)
Cons:
* Needs to be downloaded
* Weight 150 MiB
* Slower to start
* Loose language localisation
* Loose Safari's capabilities:
The whole electron app uses less RAM than the page opened as a PWA? That’s surprising to me, how come? I somehow thought Safari was more memory efficient than Chromium
I was surprised, too, so I just checked. Sure enough, https://claude.ai takes 663MB.
I wonder if there’s an architectural difference that allows the Electron app to be more lean?
(Note I didn’t download the new app, just installed the PWA, so I didn’t verify those numbers.)
that's wild that essentially a website that's a few textboxes should take 600MBs+
The browser takes 663MB, or specifically the tab?
The Claude tab, itself. From Activity Monitor:
640.6 MB https://claude.ai
51.5 MB Claude
28.8 MB Claude Web Content
20.9 MB Open and Save Panel Service (Claude)
17.5 MB Claude Graphics and Media
15.5 MB Claude Networking
5.6 MB com.apple.Safari.SandboxBroker (Claude)
5.1 MB QuickLookUIService (Open and Save Panel Service (Claude))
---
If I open Claude in Safari, Safari itself is 204MB, and the Claude tab is 640MB.
Those numbers don’t add up to 640mb. If you have 10 Claude tabs open, you’re not using 6gigs of resident memory.
Insane, what could possibly take so much space?
I suppose I use the PWA differently than most people? I just leave it open all the time, so I don't care how long it takes to start or how much data it takes to download to boot.
It’s such a poorly designed PWA. They really failed to take advantage of any of the browsers storage mechanisms. Like why not put conversation history in indexedDB?
as a chrome PWA it's 259MB for me, vs ~700MB for the electron app.
Claude requires a phone nu.ber to sign up and the won't accept a Google number so they are completely un-usable.
I would love to try their services, but they refuse to take my money.
This caused me a problem because I signed up with my personal email address, and then wanted a separate account for work but I only have one phone number and they won't let me use the same one on two accounts. I don't really care about giving them my phone number (it's out there everywhere anyway) but this has really reduced my ability to use their product.
Same. my work even provides a SIP number, but Claude won't do voicemail codes... So no help there either.
Is this for the paid version? I'm using the free version, and it hasn't asked me for a phone number, but if the paid version requires a phone number, my reaction would be the same as yours: sorry, no.
I was unable to use the free version without a phone number.
In the age of eSIM, just order a free eSIM with zero credit, never charge it, never have any cost, and only enable it on your phone when a service like Anthropocene wants to send you some (completely unsecured) SMS code.
Don't see linux, also no screenshots?
They add the same Option+Space (on macOS) shortcut that ChatGPT has, but the app feels a bit like a quick build overall, as it's just a web container. Hopefully they'll add some nice features that wouldn't work in a browser such as asking about what's on your screen or a shortcut to be able to speak questions and get a voice response.
An Electron app is anything but fast.
It can be faster than a normal web page depending on how they configure it though.
Indeed not everyone is able to write Web sites with good performance across multiple browsers, that is 3l1t3 skill, better ship Chrome alongside the application.
What a simplistic (and I think erroneous) view of Electron!
You can't spare 40mb of RAM?
It's more like 400MB due to how they're packaged. And when you could have it run with 10x to 20x less RAM, it just scream "we don't care about you". Not everyone have 64GB RAM and 22 cores of CPU in their laptop.
It's like using a semi-truck (with the trailer attached) to go buy groceries 100 meters away.
I'm unsure where I'm supposed to look. In the Task Manager, the app I created (Video Hub App), when running, is routinely between 40MB and 120MB under the "Memory" column. Are you saying there is RAM use tucked away under different row than the app name?
https://github.com/whyboris/Video-Hub-App
Your app uses 150MB of ram upon launch.
It peaks to around 3GB during the hub creation (on a folder with ~100 videos but I see your demo is limited to 50).
It finally settles at 350MB.
Here's the breakdown: https://ibb.co/N9MYvNx
My guess is that what you view in your task manager is a single process, not all the processes started by your application.
The fact people prefer webapps over native locally run software because of '400MB' honestly pisses me off it's so stupidly trivial.
It's not for the storage or ram space. It's for being inefficient which always making me doubt about the company culture. If you want native capabilities so much, just build a proper software for the platform. If VLC and Calibre can do cross platform, you can do it too.
It's not native though. It's just a webapp in a second browser without any of the usual browser features, and reduced sandboxing.
It's all the negatives of a webapp and all the negatives of a native app with none of the positives.
A website packaged with the Web browser isn't a native app.
How is 400MB trivial? I run more than 2 programs at once, and this adds up. Imagine if every trivial process in your OS used 400MB on startup. And what pisses me off, is when I have to close resource heavy electron applications (signal, vs code) running in the background to free up RAM.
I already have a browser installed.
Apparently the developers couldn't get rid of turning the Web into ChromeOS.
More like 400MB, and the problem isn't that I can't spare it. The problem is that I can only spare 400 MB a few times at most before it starts becoming an issue.
I have stuff to do on my computer, stuff that often involves multiple programs (one of them almost always being a browser!). Then there's bunch of support stuff that I want, or have to, keep around. A music player, an IM, a mail client, etc. When everything gets packaged as Electron apps, even those tiny utilities add up quickly, and everything slows. down. Or worse, gets starved for RAM and starts stuttering (Windows) or just gets OOM-killed (Linux).
And I say that as a dev/techie, who has plenty of RAM. Non-tech people tend to have computers that are memory-starved by default. Building trivial stuff as Electron apps, where native would be 100x less resource-intensive, is just peak developer laziness/selfishness.
Probably a mac user. Those devices are low in ram.
Windows/Linux over here and anti lazy developers that help Google spread their browser monoculture.
I use macOS and don’t really have complaints about the few Electron apps that I use daily.
“Electron bad!” is as much of a cargo culted snobby HN myth as “Macs have no RAM!”
Electron apps are terrible at following platform conventions. Scrolling is wrong, menus are wrong, etc. There’s plenty to dislike.
Electron bad ===> Couldn't be bothered to learn how to do proper Web development and Google is my overlord.
They're called desktop applications for a reason. Browser apps should run in a browser.
can someone tell me what the difference is to the webclient? Seems like a webview to me. No local caching/store.
Kids these days insist that they need the "app" of something, even when it's literally the exact same thing as the webpage.
Between this phenomenon and the refusal of many to close tabs in their browsers, I can't help but wonder if a huge percentage of the populace have never heard of or used Bookmarks on their web browsers before.
I'm not a kid and I know about bookmarks, but I prefer using apps to apps in my browser. Electron apps I have less interest in, but I really don't like working in something like Figma (which is in the top of it's class) in the browser.
In-browser apps do all sorts of clunky things like going back a page because I'm trying to scroll right or left or hotkeys not working. Native apps feel so much better to use and I can close my browser and stay focused on the task.
I think it’s less about not knowing what a bookmark is and more about just having a more direct way to access it, where it’s more high-level than just in a tab. There’s just something mentally different about opening a program from your dock as opposed to from a tab, even if they’re both basically a website.
This line of reasoning was why I was hopeful that PWAs on the desktop would become first-class citizens, as far back as Mozilla Prism[1] in the late 2000s. They never really did seem to take off.
1. https://wiki.mozilla.org/Prism
It used to be fairly common to have web-page shortcuts back in the mid-00's. Lots of apps would add spam shortcuts to your desktop that would just open your web browser to a web page where you were encouraged to buy something/donate/start a free trial.
Anyways I think that capability still exists on all modern OSes.
You aren’t as wise and smart as you think you are. Perhaps people just have different preferences to you?
I have no issue with others having differing preferences, just making an observation about the continual reinvention of this particular wheel
I swear if all the ai chat apps started as fat apps and then Claude just released the first port to the web, I would be reading this exact same comment on hacker news except complaining that there was nothing wrong with the fat app.
I like to ALT-TAB instead of CTRL-TAB because I always have fewer applications open than I do sites. I don't have dozens and dozens of tabs always open either, but always more than my programs.
> the refusal of many to close tabs in their browsers
I have too many bookmarks, I'd never find anything in there. Keeping it open is the only way.
I was thinking more about people who use Tree Style Tabs and have nested Tab folders and such when I wrote that part. It's simple enough to save an entire session as a bookmark folder, to open an entire folder of bookmarks as new tabs (rather than needing to hibernate tabs to save ram while still keeping them "open"), and to search through bookmarks ad-hoc. Bookmark folders can be infinitely nested, categorized, ordered by when you were looking (though searching through your browser history also works for that), combined, copied, pasted, moved, put into or removed from the bookmark toolbar.
It's an incredibly flexible system that allows for a massive variety of workflows and it feels like people just keep finding ways to recreate it, but worse: less platform-independent, eating more system resources, dependent on third-party plugins, or (in the case of using an app for everything) eating up orders of magnitude more storage space at rest.
It's simple. Tabs keep state[0], bookmarks don't.
The web went to shit a long time ago; you can't rely on being able to bookmark a site and then go back to where you were. Half the sites are infinite scroll, or dynamically generated pages, or SPAs[1], or some other ephemeral invention du jour. Keeping a tab open gives you some chance to return to where you were for some time; bookmarks are just giving up.
--
[0] - At least until the browser decides to screw with you and unloads the tabs you needed. Firefox on Android is particularly aggressive at that, which incidentally makes PWAs unusable, too.
[1] - You're lucky when those let you make a bookmark that won't drop you back to index page when loaded.
Now that's an interesting perspective I hadn't considered.
I suppose I actively avoid most infinite scroll webpages because to me they feel like they're meant for "consumption" and visual appeal rather than actual usefulness and I tend to think of webpages as tools, either for information or socializing. As a dev I'm certainly biased towards sites like Github or news sites, which are both meticulous at sending you to the page you were linked to specifically rather than redirecting you elsewhere.
I do tend to keep a few tabs open at any given time, but only when I'm in the middle of reading something like an article or blog post. I try to close down my tabs to something like 5-8 max weekly, if not daily. Something about closing out the mental context and making a conscious decision about whether I really want to finish reading something or if it was boring me feels very freeing once I've decided to close the tab is deeply satisfying to me.
I have ChatGPT app and Claude and Gemini PWAs on my dock. I gravitate towards using ChatGPT much more. It's just faster vs launching Chromium for each small query.
Having a hotkey to access the app removes a lot of friction
Dinosaurs these days insist that there's only one way to do things, even when most apps are literally designed to accommodate different workflows.
Please, tell me where I said there's only one way to do things.
My question was not meant as a critique of people who know what they're doing. I'm not talking about engineers or sys-admins like the typical HN crowd, I'm talking about the population at large. The average person understands their devices just enough to get them to work the way they want. It stands to reason that a fairly large percentage of the modern population, used to using apps for every individual service, may never have had a use for browser bookmarks and may therefore never have even attempted to use them.
I have a few friends who work as professors and tell me stories about how their students don't know how to navigate a filesystem in Windows. Why would a person like that, for instance, have bothered to learn what a bookmark is vs just keeping a tab open?
From what I can tell you're right, it's essentially a web front end made to compete with ChatGPT's desktop app.
The only real benefit is the hotkey (ctrl + alt + space) for starting a new prompt when the window is closed, but still running in the background.
chatgpt has that shortcut too.
option + space for mac, and alt + space for windows.
There isn't one, it's just an Electron app that opens slower than the webpage and flashes white before it appears.
You'd think, a company that has such powerful AI and no restrictions on requests to it would be able to bash out truly native apps no rather than something that looks quite shoddily made, like making your electron app not flash white when in dark mode is amateur level stuff.
but this can start with Windows and sit in your system tray, sending you popup reminders!
We're 2-3 sprints away from a claude popup saying "It looks like you're writing a resume, would you like help with that?"
A Copilot version is probably slated for 2025H1.
Pssst, they are probably here reading
You can do exactly the same with a PWA without Electron garbage.
I would consider using this if I could just shove my API key in and have it run.
Tangentially, what are people using to have a unified interface to query multiple LLMs? Juggling multiple apps or websites for each provider gets annoying. I was using openrouter but the chat history is only stored in browser, so I'm looking for something else now. I tried open-webui[1] and that looks pretty good but I wonder if there are better alternatives.
[1] https://github.com/open-webui/open-webui
Give https://recurse.chat/ a try - I'm the developer. One particular advantage over alternative apps is importing ChatGPT history and speed of the app, including full-text search. You can import your thousands of conversations and every chat loads instantly.
We also recently added floating chat feature. Check out the demo: https://x.com/recursechat/status/1846309980091330815
Regarding chat history, I've been thinking that people should have ownership over chat history. We are migrating the chats towards SQLite, so your data is going to be a timeless format like files* - SQLite has long term support through year 2050. https://www.sqlite.org/lts.html
* See [File over app](https://stephango.com/file-over-app).
I'm a very, very happy user of TypingMind. It's not free, but it's worth the (one-time) license cost in my opinion. It has support for multiple providers, plugins, "agents" (pre-set templates), chat grouping, import/export, automatic syncing across devices, etc.
https://github.com/TypingMind/typingmind https://www.typingmind.com/
There are a ton of alternatives if you search around. I created my own because I wanted something keyboard centric, and use OpenRouter with it: https://github.com/iansinnott/prompta. They all accomplish roughly the same thing though.
this is the best app i’ve found so far, better than anything else. Unfortunately it’s not open source and is only free for personal use.
It has web search and RAG that actually works properly out of the box. No fiddling with 100 settings.
https://msty.app/
I agree, I'm also firmly convinced it's the best. Until a month ago it has some bugs I disagreed with some design choices, but since 1-2 weeks ago it's the absolute best I've ever used and have easily convinced others to use it.
I use Chaz to just talk to any LLM from Matrix, my normal chat client.
https://github.com/arcuru/chaz
Try MultiGPT which is a Chrome extension that consolidates several AI chatbots, including ChatGPT, Bing Chat, Bard, Claude, and more, into a single interfacee
This is so wasteful it boggles the mind.
I was surprised that simple things like sharing a chat in Claude doesn't seem to be there. I read somewhere that it was a feature that was disabled later.
It's annoying, I'd really love to export my chats. Any Anthropic PMs reading this, as a paying subscriber I command you to prioritise this feature, on pain of failed training runs
I’ve been trying to sign up for Claude for a couple of weeks now (I was even planning to pay), but alas, signups remain closed for now.
Somewhat unrelated question for developers: when is an app generally better suited for the desktop vs a web app? Asking for myself.
Electron slop, a literal insult to the people who worked hard on what powers all of that bloated crap
I wish I knew of a good way to voice chat with Claude, and unfortunately, the new mobile app isn’t it. The voice recognition is terrible. Does it just use iOS’s built-in text-to-speech? It needs to be as good as Whisper to be useful.
I use superwhisper for this.
+1, that fixes the voice recognition part, but it doesn't automatically read Claude's response aloud.
The main questions: Can this be used with no limit on file "uploads"?
Is there an actual good use case for these desktop apps vs the web?
(I hope this thread doesn't go anywhere. I just feel like complaining.)
There's been some inflection point where the desktop version is now often worse than the web app. And you can't blame web tech because the web app in the browser is usually quite good.
WhatsApp on macOS has catastrophic bugs for years. I regularly can't focus the input field and have to relaunch it. Until recently, its image modal would regularly bug out and leave behind a global dimmed window that eats all input so you can't even close it.
Poe's desktop app is less polished than poe.com. Off the top of my head, its desktop app scrollbar becomes so tiny that you can't even click and drag it.
ChatGPT's desktop app is simply worse than its web app or was when I stopped using it months ago for that reason.
Youtube Music is another pointless desktop app since it's just the web app inside a web view, though at least it embraces that.
It's like the only thing going for these desktop apps is that you can alt-tab to them individually. It's time to come up with a trivial one-click first-class OS-level way to turn websites into apps and put all these sad desktop apps out of their misery.
Oh well, it's come at a good time because I've mostly soured to the idea of having to install software at all. 90% can be run in a sandboxed browser away from my file system.
Yeah agreed on all of this. Most desktop apps that come after their web-based versions feel like an afterthought.
Lots of people are correctly suggesting this could function just as equally well as a PWA, but I will point out that (all things being equal) there are some valid use-cases when strictly choosing between Electron vs PWA that a LOT OF PEOPLE seem to be oblivious of:
- Electron apps don’t rely on a web server to function, so as long as you archive the app’s installer, you can install it on future computers even if the original website goes down. This assumes that the PWA/Electron App in question doesn't need server features, for example Photopea, Obsidian, etc.
- A bit easier to block network traffic to/from a dedicated application from a privacy perspective. Still can be done on a PWA, but it's trickier to set up.
Apparently running a daemon with the system browser has gone out of fashion, or a skill lost in modern times.
Not to mention the whole calling into a server anyway.
It would be nice to see a de-coupling of the backend (as a service, deamon, etc.) so that we could choose the desired frontend of our choice (net win from a UI/UX, accessibility perspective, etc.), but it's been a long time since I've encountered a consumer level application that did this.
I mean, re: point 1, we're talking about an app that calls an API for every interaction already, so that doesn't seem super relevant. If Claude is down, you can't use it whether it's through a PWA or an Electron app.
If you could access chats offline, that'd be useful.
But you should be able to do that with PWA.
> "Lots of people are correctly suggesting this could function just as equally well as a PWA, but I will point out that (all things being equal) there are SOME VALID USE-CASES"
YES - I am speaking GENERALLY. That's why I literally called that out in my comment that this particular use-case (Claude Desktop) could easily be done with a PWA.
Gotcha. Sorry, I guess that was unclear to me based on your phrasing. When you said "some valid use-cases", I thought perhaps you meant (for this app) and not (for Electron apps in general) as the implied next clause.
No worries, yeah I just meant that if a dev has no intention of building a native app and is only deciding between either a PWA or an Electron/Tauri App, then there are some advantages to the latter.
I'd say for the case of Tauri it's doubly so. You can have natively compiled Rust code for especially compute-intensive functions, rather than having it running via WASM.
I'll keep using mine https://github.com/Merkoba/Meltdown
Any site that hijacks the back button has lost my trust.
I hope this will be available from Homebrew soon.
Other than ease of set up, why would anybody use this over LibreChat, which allows you to plug in api keys for all services?
Other than ease of set up, why would any one use this over training their own large language model from scratch using magnets?
I like LibreChat a great deal, but its pretty heavy-weight (over a more bundled solution like Jan) in terms of spinning up a bunch of docker containers (meilisearch, mongo, vector db, etc.)
LibreChat is still a webapp, not a native app (ie, the topic of this post).
This LLM hype is out of control, they literally bullshit on any novel input (admittedly there's not a lot of this) and are wrong 15-20% of the time for all inputs
This isn't a "we'll just add more context", "we'll just add more instructions and params", "our multi-head transformers will auto-attention the tits off your query" etc - this is search based on probability maps. It is structurally unable to be intelligent, no matter how many models you chain together and how much compute you throw at it. But it can sure mimic that shit, which is why LPs are going to lose their shirts and GPs are going to struggle to raise follow on funds.
Let's move on.
Software engineering tools without Linux versions are baffling to me.
Why? Linux users are a minority even among developers.
https://survey.stackoverflow.co/2024/technology#1-operating-...
You sure about that? To my eyes those stats say Linux is more popular than MacOS.
I'm still waiting for the ability to change my email address :/
I'm still waiting for the ability to change my email address...
...and also to log on without visiting an AlphaGoogle domain.
It speaks poorly of Anthropic that their 1FA login requires a third party.
Then again, Google isn't exactly a 3PP in relation to Anthropic, more of a 2½PP.
Thanks guys for the linux release !!!
i've lost my faith to claude because of gpt4o and the advanced voice features.
It’s a rat race. They keep pacing each other, and it feels like each of them takes over the other one for a few months.
I do not want to give my phone number to a company.
Phone number is the new cookie. Just about every company now requires users to provide phone number. It’s insane.
No demo?
Yet another Electron app... In comparison, ChatGPT's awesome client is native.