The ChatGPT/Codex app bundles a full copy of LibreOffice
simonwillison.netWould be nice if they donate to LibreOffice then, to improve the support of various MS Office features in files, as well as comparison/diffing features. Win-win to everyone.
It's like that scene in the Simpsons when Bill Gates comes to buy out Homer:
"You don't think I got rich writing a bunch of checks, do you?!"
As much as I would hope they would... I don't think that's in character for sam altman.
Sam "Its only Open until it does not benefit me anymore" Altman
I actually bundle LibreOffice with my app too and the reason is reading files, especially old xls files. Since I'm bundling it I'm now using it for everything docs related but the specific reason is those old files. I couldn't find anything else that I could just drop it and feel confident it'll just read anything I give it.
Does that really mean that it's bundling those apps from the start or did it just download and install them at some point to do some local work on some prompt or job you ask it to?
I don't see it making much sense to bundle it. I'm sure a LOT of LLM prompts are related with docs, excels, powerpoints etc, etc but don't really see it worth it for it to be bundled on the codex app from the get go, because otherwise, why not also install dozens of other apps?
I'm not sure. Someone who hasn't installed the ChatGPT (or Codex) apps yet could confirm this by installing the apps, seeing if that ~/.cache directory exists, then try running a prompt that needs Python or Node.js or LibreOffice and see if it downloads them when needed.
I've never installed the ChatGPT app on my Mac until today. I ran the app, signed in, and I see the same.
% ls ~/.cache/codex-runtimes/codex-primary-runtime/dependencies/native/libreoffice-headless
libreoffice manifest.json
On a Windows installation I see this:
So doesn't seem bundled with all installs at least.PS C:\Users\user> ls C:\Users\user\.cache\codex-runtimes\codex-primary-runtime\dependencies\native\ Mode LastWriteTime Length Name ---- ------------- ------ ---- d----- 8/27/2026 10:09 AM git d----- 6/16/2026 9:18 PM jxrlib d----- 8/26/2026 7:37 PM libheif d----- 6/17/2026 5:32 AM poppler d----- 8/27/2026 10:09 AM powershellIs this perhaps because the aim is to read Office documents and there are (maybe, I don't know for sure) default components within Windows that can do that?
I don't think there are.
Even when you have the MS Office suite installed, it doesn't really provide a good interface for LLMs or automation, so AI tools don't really use it.
Codex always try Libreoffice first on my computer.
Notably: this did not exist after I installed the app, it existed after I ran the app and there was the long startup screen. Presumably, and I'm not sure I'm interested enough to redo it all and find out, that long startup screen involves the default skills plugins setting up their environment and one of those pulls it as a general app requirement.
I also saw this present after first run on macOS but not Windows.
I have had the Codex/ChatGPT app installed for a few months, and can confirm that there's a 422.9 MB file named LibreOfficeDev.app in the above path.
More notably, I don't use Computer Use or any of the fancy features, just Codex for agentic coding.
> why not also install dozens of other apps?
because other apps are less likely to be useful, and there's diminishing returns
The new app is an unbelievable mess. Settings are senselessly organized, too. The whole thing has this aesthetic slickness and then underneath it's like the people who make it have never even used it.
It's not surprising that it's pulled in absolutely massive dependencies, although I'm not sure it's the wrong call on some operating systems. LibreOffice is pretty tried and true.
Personally, I like the 3-pane organization (sidebar, chat, browser). I turned ChatGPT into my bookkeeper and social post writing/kanban space, and it works well for these script+browser workflows on desktop.
However, all of this is confusing:
- the ChatGPT/Codex split on desktop
- Chat/Work split on mobile
- how you remotely access each chat from your phone (some of my most valuable chats are buried in Remote)
- which chats propagate from device to device, and which remain on that device
> - the ChatGPT/Codex split on desktop
Do tell, cause I run the "ChatGPT" program on Windows to use both the normal Chat and Codex (side menu items). I think its a more recent change.
They removed the global shortcut to show a portable chat overlay. That one entrypoint must've been the source of at least 80% of my chats, yet apparently it was considered too vestigial to keep.
Can it be reassigned under Settings / Keyboard shortcuts?
HOW DOES IT WORK FOR YOU PEOPLE?!
Istg on windows, I can't get codex app, new chatgpt or chatgpt classic to work without hanging on even a moderately sized chat. I tried all of the different tricks people recommended but that shit just crawls. It was stable at some point, classic that is, but then it started being a laggy mess.
I've uninstalled new chatgpt/codex app. I can't access phone chats on it, I can't use the desktop allocated chat messages on it. I have to use up codex limits. I am so surprised that OpenAI ships that hot garbage on windows.
One thing I've found - these things are pretty good at troubleshooting desktop problems. If the codex app is running slow, I'll note that it shouldn't be, and also you might get more traction than you'd expect asking the codex app why it's slow.
Mac Neo can be had for $600 now, and it runs Codex incredibly well!
But even on the Windows machine, if you Tailscale into a free tier VM for the workloads, you'd probably have a much better time.
Just leave Codex itself on your machine, if you set up computer use it will happily set up all of this for you while you have lunch.
Windows has never felt more archaic than now. I can't imagine trying to do work on a Windows machine in the age of LLMs.
If there is anything that LLMs are rock solid at these days, it's unix terminal usage.
They're also great at powershell, you just need to tell it to do so. sh is neat but it's 55 years old, there are newer and better shell environment concepts out there and the robots can handle them well too, try it some time!
How do you work on Windows in the age of LLMs?
That’s easy: you activate WSL and then ask the LLM how to get networking and stuff working since they changed it with WSL2. You need a config file flag to setup network mirroring now.
> The new app is an unbelievable mess.
How so? On Windows, I like it (version 26.825.51511).
Works fast (like modern Claude Code desktop, their old versions were bad), has pretty okay UI (could have an option for wider main text display, like ZCode added), allows you to change the fonts and shells and some of the layout stuff.
I mean currently I have Paseo hooked up to OpenCode that runs the actual models and often that works a bit better for me, but as a standalone desktop offering, the ChatGPT app has been okay in my experience. I can kinda see why ZCode copied them a bunch (though ZCode also lets you run 3rd party models, which is actually preferable to me).
I do like the idea that Claude Code desktop had in the earlier versions, 3 tabs - chat, work, code, except they kinda went in the direction of merging chat and work which was very ehhhh. The ChatGPT app doesn't even have a proper split between chat and work/code.
Almost like it's... vibe coded.
yeah, that seems obvious by now.
The thing that blows my mind about it is that they straight up copied Claude Code and tried to frame it as some sort of incredible accomplishment. There are no new ideas, no insightful UI/UX paradigms, no killer features. They say money can't buy taste. The ChatGPT/Codex app is proof that it cannot buy creativity either.
which is really funny/sad when you see openai staff doing victory laps whenever claude code copies one or two small codex features like /goal.
Codex, clone Claude Code, make no mistakes they also didn't make.
And when you go to their forums it says "This forum is not for discussing the ChatGPT App..."
..like excuse me? Where the fuck else would we post feedback about it?
Just tell ChatGPT what changes you want in the next version.
Is that what it is using to render and manipulate MS Office documents? That'd explain the poor rendering of some of my files. Bundling all of LibreOffice seems like a pretty huge dependency.
What's ironic about this is, afaik there's no programmatic way to get a screenshot of a PowerPoint on windows other than opening up PowerPoint and screenshotting it. I only discovered this when I first tried out Anthropics official PowerPoint skill when it came out this winter.
If anything happens from this, I hope it forces Microsoft to address some of its long standing hostile to power users issues so they can stay relevant in the world of AI tooling.
This can absolutely be done programmatically. I just added the ability to do that using PowerPoint for Windows yesterday in my uv/python tool `markdown-pptx`. Codex added the functionality in less than 5 minutes.
I have a skill that instructs ChatGPT/Codex to verify the rendering using the Mac versions of Microsoft Office. Even without registering, you still can open files read-only, which is enough. Apparently Office refuses to open some Open Office XML documents that LibreOffice is fine with, so that's another reason to use this. After a few times, Codex ended up storing some memories to stop producing broken files in the first place.
I generated the skill originally for ChatGPT 5.5, so some of these may not strictly be necessary anymore with later models. But posting here in case others find it useful: https://gist.github.com/telotortium/844386f762c4b3bab49999ba...
For a docker image of mine I was bundling the full version of LibreOffice as well but I managed to get some space savings by bundling just the no-gui version.
Seems to match the screenshot in the blogpost but libreoffice-headless still shows as 429.7 MB.
> Bundling all of LibreOffice seems like a pretty huge dependency
What's the alternative library you see?
pandoc
Apache POI
I don't think that's anywhere near equivalent. That's for reading and writing Microsoft Office docs. It has no office functionality. LibreOffice has a full API, for example to run calculations on spreadsheets [1].
The goal of bringing in dependencies is so you don't have to duplicate the functionality, in this case, crafted over decades of effort.
[1] https://wiki.documentfoundation.org/Documentation/SDKGuide/C...
POI can run calculations. [1] That doesn't make it a full office suite but it does a lot.
[1] https://poi.apache.org/components/spreadsheet/formula.html
Why not computer use or using the tools the user prefers instead? The primary office suites have APIs/CLIs now
If you've ever actually gone down this route you will try exactly this realize how much of a cluster f these products are and just use libre.
I did the exact same thing in my AI system. Spent an ENOURMOUS amount of time trying everything I could and in the end libre was the only thing even somewhat functional.
Easily 2 months of my life lost to these god forsaken systems. APIs and CLIs sound reasonable until you actually attempt anything in this space.
Computer use is exactly the same thing. Go try this: go to a google doc and dev tools check out the dom and now do the same for a word doc. Eye gouging pain of trying to get anything to work.
Computer use is slow and can take control of the application focus. I’ll be doing things while Claude runs in the background. Suddenly it decides to open the app and try to do something + screenshot it.
I don't know how many times I've had to yell at codex or claude to use/implement headless versions of whatever it's doing at the time, interrupts me trying to multitask and play Battlefield or Factorio while they work :P
don't try Dyson sphere project (/s, its acrually really great), but maybe try x4 foundations, I'm contemplating how I might have my agent help manage my empire
That's where the CLIs come in for me, I'm not going to waste tokens on screenshots and input device manipulation
https://devblogs.microsoft.com/microsoft365dev/getting-start...
Yes. If you have word you can instruct to use word via com or AppleScript.
Just the headless version looks to be bundled. The title as posted might be incorrect.
There is a serious threat to Microsoft here, if everyone starts using AI for generating and updating Office docs. At some point Office will become just a viewer.
im stoked for that outcome, standard document tooling should not live behind subscription models anymore. i dont care that microsoft spent zillions bankrolling education systems and companies to fall into this cultural lock in over the decades
whether they like it or not spreadsheets word docs etc are just ubiquitous parts of existing in society at this point you're gonna need to reach for these tools eventually just to apply for jobs or do homework
You assume tokens will be free after these get their monopoly positions? Actual libre office is actual open source running on weak hardware. Truly free.
And I suspect we will look back at MS with nostalgia once Altman and Musk get their ai imperiums.
I have not created or read any Word document for a very long time. I only create PowerPoint presentations a few times a year for work, which could be easily replaced with PowerPoint online or Google Slides or even just some markdown files.
But I have a license for the entire office 365 including Copilot premium.
Guess why?
For Excel?
I think they meant to view documents.
Don't worry, they are ahead of this.
Satya said some time ago that they expect soon to have more agents using Microsoft software than humans, and are working on transitioning their focus from serving humans to serving agents.
https://www.news18.com/business/satya-nadella-announces-micr...
So they will switch to per-agent pricing instead of per-user. Fine, but that doesn't change the fact that Microsoft Office will become a just bunch of viewers as far as human usage is concerned. As far as generating Office documents, you can do that without using any Microsoft code.
This is one that's already true and "everyone" hasn't figured it out yet. But you don't really even need Office as a viewer, PDF is fine (for printable/portable fixed-layout) or just HTML. "Office" is a product that doesn't need to exist anymore, at least for the core apps. For day-to-day office work, you probably shouldn't be using those legacy apps, you should be reviewing portable document formats and asking an agent to make the revisions that you want.
> For Microsoft, this shift means products like M365 will act as the foundational layer where AI agents operate. Users will still need places to store documents, manage projects, search files, and track workflows
They bet on the online office suite (M365).
... and without the license, crucially!
This sentiment is also why Microsoft crushed earnings this quarter. Their AI play is to create corpo-slop AI middleware, which sits between AI models and enterprise applications. Sure, others can go and build harnesses which can generate MS office documents, but MS is already doing it, and conveniently such harnesses will come with everyone's existing Microsoft 365 subscription.
The analogy is IBM. Just like IBM helped every old-money corporation on the planet figure out the whole "computers" thing 50 years ago, MS will help everyone solve the "AI" thing. The software will suck and will not really innovate, but it will make a ton of money
> The software will suck and will not really innovate, but it will make a ton of money
Microsoft’s goal isn’t to provide high quality software that outperforms other software, their goal is to provide an obvious and reasonably proximate answer to clueless execs trying to check things off of a procurement list so that existing contracts can snowball and competition is excluded.
We, collectively as a global species, really lost out when MS wasn’t divvied up in their anti-trust trial. Office as its own corporate entity would be in a radically different place right now, as would their enterprise customers.
But why would openai and antrhopic pay them if they already libreoffice for free and why would users pay if they already have an Ai subscription
Haven't used MicrosoftOffice in the past 7 years.
Curiously, I didn't find any reference in the Open source licences section of the codex app.
Is this a MPL 2.0 violation?
Despite the title, I'm not sure it's actually bundled. Some of the skills plugins will download it on the first run. So it depends where they are grabbing it from rather than the presence of the binary on the system, but I haven't checked that deep into where they actually grab it.
The license itself may be present in the cache folder.
Alternatively, ChatGPT might happily disclose authorship and licensing if asked.
I'm not convinced of the argument that the fact that it is not bundled at download or even at install-time, would free them of the GPL responsibilities. And at this point.
I'm also not sure HOW the devs bundled libreoffice, whether they just mentioned in a prompt (use libreoffice), or whether they hardcoded a command (like apt install libreoffice), more likely the former, considering the recursive vibecoding nature of this software.
Whole lotta words to say that I don't know whether this is a GPL violation, but interested in the answer. If it is a violation, the likely fix will probably be yet another prompt that says something like "comply with licensing", which probably already exists, so a more specific "provide attribution (in a way that doesn't bother the user)"
> Whole lotta words to say that I don't know whether this is a GPL violation,
I don't need a lot of words to say it can't be a GPL violation since Libreoffice is distributed under the MPL v2.0 license.
I finally did find the license in
~/.cache/codex-runtimes/codex-primary-runtime/dependencies/native/libreoffice-headless/libreoffice/LibreOfficeDev.app/Contents/Resources/LICENSE
Copyright violation is the last thing one could accuse ChatGPT of.
Well, this one is particularly obvious, other one are more fuzzy.
I think that depends on if they distribute it with the install package, or if they download it once installed.
Reminds me of pulling in an entire JS framework just for a single button. The bundle size must be something else.
Now someone please use ChatGPT/Codex or Claude Code to rewrite the entire LibreOffice source into a single Rust binary already! Just like Jarred Sumner did for Bun [1] or Carmine Paolino did for Fastpotify [2] - granted there are some rough edges for sure, but the slim diet trimmed down and performance gains were wrothy.
[1] https://bun.com/blog/bun-in-rust [2] https://fastpotify.rocks/
> rewrite the entire LibreOffice source into a single Rust binary already!
Why?
Someone may need to explain to me what is wrong with this sorry. Is it because it bloats the app with functionality that some users will never use? I feel like that is fairly usual for productivity/business apps generally.
I don't see anything wrong with it, I just think it's interesting.
Oh right, yeah I agree. It's the kind of thing I can imagine the harness downloading in the background anyway (or instructing the user to do) at some point so I can see the sense in just skipping that bit. Kind of comforting they just do it the same way the rest of us do - if it ain't broke and so on.
It does, however, introduce a potentially enormous attack surface into ChatGPT, along with creative opportunities for less security conscious individuals and LLMs.
I think it's overbloating my computer
Actually, people do use and require the functionality, which is the point. The title as posted might be incorrect.
Yes, that's what I mean - maybe that didn't come across cos I said "bloats". I work for a business that builds web apps for corporates - LibreOffice is bundled in most of the repos of ours that I've seen, a bit like Imagick is in the broader sector.
First Claude installed 10GB VM [0] without asking, but OpenAI ships LibreOffice. What's next? Somebody ship their monorepo verbatim?
Honestly, not sure it's worth having those app on your localhost at all. I ended up uninstalling all frontier labs bloated apps and start using Google AI Mode + Antigravity CLI / Claude Code sometimes.
Imagine the enormous complexity of office file formats if your business is literally building coding models/agents and still you choose to bundle LibreOffice.
I suspect that it’s used in headless clip mode to reliably convert file formats back and forth.
I wonder why they didn't try to discover if the user already has those binaries and use them before downloading.
Why would they? Storage [they don't have to pay for] is cheap! And just bundling it saves them time.
Because they vibe coded it.
Also in the newest app I think the transcription feature is unusable. Ok, I can have a shortcut to start and stop recording. Wow, cool! Ok, but where did the transcription go? In my clipboard? To my cursor? Hey? :(
I've noticed that when I have codex read/modify word files the live reasoning snippets often allude to LibreOffice, so that makes sense. I think it was a quick and dirty way for them to interface with docx and xlsx files?
Bundling LibreOffice with a chatbot app feels like peak software bloat. My hard drive just let out an audible sigh.
having had to manually save some xls files by editing their xml, yeah it makes sense
Are they at least respecting the licenses (.i.e.: providing a verbatim copy of each license?)
It might be for their new Work product which can produce docx, pdf and pox files.
(This is more of a tweet than a blog post, it really wasn't written with Hacker News in mind.)
Indeed, but that likely wouldn't have stopped someone from submitting it with a misleading title. HN is known to get a lot of misleading titles submitted and to have to change them. The <title> tag is better, it doesn't say "a full copy". In the screenshot it shows "headless" in the filename. I would simply add the tiny detail of headless to the <title> and use this title for HN: Codex Bundles Headless LibreOffice
frankly how dare you
How is simonw to blame for somebody else submitting it to HN?
True, I feel like the entire HN community is to blame for this, particularly pg and dang.
Thanks Obama
I have a feeling you should learn about Poe's law: https://en.wikipedia.org/wiki/Poe's_law
Anyone quoting Poe's Law is clearly just a Nazi.
...
For those that can't see the sarcasm: https://en.wikipedia.org/wiki/Godwin%27s_law
spinning up an entire copy of libreoffice sounds super heavy?, if someone here is actually using chatgpt to edit docx documents then they can feel free to use us. vespper.com
You want something heavier? They were trashing people's SSDs a few months ago. IDK if it's been resolved, yet.
DNS_PROBE_FINISHED_NXDOMAIN
My agent did it? I didn’t do it.
I don't see the problem?
Who said it was a problem? This looks like an observation to me, not an accusation.
Last time I used the Windows version it pegged all my cores at 100% usage for some reason. The machine became unusable. Instead I used Codex with JetBrains Rider (doing a C# project) and it worked flawlessly.
(I made a similar comment in a different post by Simon Willison yesterday about GPT Work)
I am unsure the direction that AI assisted document creation work will go in. I am opinionated and building something in the space. In this reply I discuss my view + what I have been building.
My opinion is: We now have AI, this new amazing flexible tool. We can throw it at any problem. But in terms of document creation we have a choice between old fashioned siloed document formats (.pptx, .pdf, .xlsx, .docx, etc.) and a freshly rendered - slightly randomly styled - html site with some basic javascript within it. I think these combinations are odd and do not maximise the powers of AI.
I've been building something called SmallDocs [0][1] which is a new take on work documents. It's all Markdown, but it always renders (100% privately) on the SmallDocs site. This means the SmallDocs renderer can convert recognised Markdown combinations into powerful traditional document formats [2]. For example, we convert a ```cells block into an in-line spreadsheet you can export to Excel.
This gives an agent the ability to express itself with a great deal of depth and flexibility. For example, your agent can produce a single data analysis "SmallDoc" with text, charts, spreadsheets, etc. [3] I find that combining formats quickens the speed I can understand a topic.
SmallDocs also have standardised and pretty boring styling. This means you spend less time understanding the layout of a flashy HTML page and more time on the content.
I also believe a pre-determined rendering engine has a lot of advantages over a freslhy built HTML artifact. It allows you to build in lot's of useful defaults that the agent doesn't have to code from scratch. These can be small things, for example, every time your agent renders a Markdown table in SmallDocs you can instantly copy it as a CSV or a PNG. And these can be bigger things, such as a rich slides DSL with solid styling defaults, which saves your agents a lot of tokens and results in consistently attractive HTML slides [4]. (Slides can also be automatically exported to .pdf and .pptx.)
I've been at it for a few months, and am still working on the project. I am looking for feedback and open source collaborators. Thanks for reading!
[1] https://github.com/espressoplease/smalldocs
[2] https://smalldocs.org/blogs/what-is-a-smalldoc
[3] https://smalldocs.org/s/46fWiAyF77T9c7tqYwAQqp#k=j78ZEOOrgo8...
[4] https://smalldocs.org/s/QrmeWMsULGS871nXQYOacB#k=ubOHH9QtTWp...
(And something totally different, your agent can also use SmallDocs to walk you through code files: https://smalldocs.org/s/JsI_-tWSH0UpSd_YXxVLPm#k=G5jLCYA4JUX...)
Presumably this is for ChatGPT Work? https://simonwillison.net/2026/Aug/30/understanding-chatgpt-...
There's no security benefit to doing this versus demand-downloading hashed-locked components on need.
Ok, it's probably not a security decision. It's a common pattern with both storage and memory management to push most reservations to initialization, in order to cluster resource insufficiency errors to a specific install-time when the user or admin can do something about it. An excessive just-in-time approach in both storage and memory would cause OOM and Disk exhaustion to occur at many different times.
And wait till you find out about @oai/walnut
do tell!
Walnut is the .NET and OpenXML backend. I think the point parent was trying to make it that it shouldn't be surprising Codex bundles runtimes for damn near everything.
probably uses it extensively to convert between LLM working on a doc and exporting it out to excel, word etc
I still don’t understand why with all their cash these companies dont redo their apps in Rust or native. They are all so heavy and dog slow. There are already some very fast cross platform (wasm too!) libraries that can do what poppler does as well as parse and produce various office file formats. Because 2gb is crazy town, these things could be around 100mb, just throw some tokens at the problem.
You should go actually try to make any basic automation work around these systems.
They purposely obscured, google doc is another example they completly hide the dom!
I spent months fighting word processing systems and ended up shipping libre. There's almost zero alternatives without that becoming your entire company.
Not that easy. Also, they have "priorities." Although their priorities are different than the users'.
Just throw some tokens at the problem. I’ve vibed up a Rust Crux hybrid app that runs well on macOS, iPad and wasm and does some pretty complex xml manipulation and other document work. Absolutely smokes electron and you get native UI.
On top of everyone else’s great points, if you make your own “fork” of something, then you have to consistently maintain it. Which can be done with tokens, but an extra cost with something that’s not as user tested.
I’m not talking about forking, but using standard libraries and native frameworks instead pretending electron is a desktop app.
Ah good point, I thought this was coming across from the "recreate" your own approach. I would fully agree with you on using native frameworks, though the eletron desktop is a great "entry" onto desktops ensuring companies only have to update 1 source, as opposed to natives everywhere.
one does not just "throw tokens" at a problem, Ai is not that capable, and we are talking about MS Office file format, likely a NP hard problem
But in case they succeed the return on image will be astonishing.
Can you expand upon this?
A full implementation/rewrite of something like word, which is pretty complex and the quantity of edge cases is insane, in a relatively short amount of time, months, is very hard.
If they succeed with AI it will be huge: it's basically not possible to do it today with humans, no matter how many people you put on the job.
I would agree a full rewrite of word which is very complex and take quite a bit of time.
It would be impressive of AI to make a 80% rewrite, though I fully doubt it would cover the 100% rewrite (disregarding long term support as well).
Why do you think it’s impossible for humans to rewrite word, especially even humans made it in the first place?
I think it's not possible with humans in a relatively SHORT amount of time (let's say 1 year).
I mean, aside from all the NP Hard sub problems (I might not be reading your satire though), the real hard problem is that a "product" is not just software, but an ecosystem and a promise of maintenance of the abstract user experience & quality of it.
AI is especially good at ports like this where there’s a known and well tested surface area.
*that only microsoft has access to.
These guys puposely obscure controls and understanding of these products 100% to prevent you doing a port.
Just go open a microsoft word doc in the browser and look at the dom.
Despair! Horror!
Especially when there are three different versions, and each has a different team, and they all can work together at the same time. Crazy amount of tech supporting it that breaks a humans context let alone a models context.
"ports like this" skips the part about Microsoft/enterprise software having a known and well tested surface area, the few examples we've seen are exceptional cases I suspect
What % of time, for a an average session, do you think is app overhead vs waiting for tokens? And there's your answer for why it's not a priority.
Jon Blow's response to this take was, "yes, which is why you have to work even harder to hide latency", instead of adding more on top.
From OpenAIs perspective, resources on your computer are free and wasting them is inconsequential
Kind of. Eventually it gets too slow even for the OpenAI engineers using it, and then they need to fix it.
What makes you think Rust is automatically better?
(Also, "Rust or native" is a strange dichotomy. What's not native about Rust?)
I think “or native” is a gesture at swift/objc for mac and C# or whatever people write windows apps in.
Rust or native is objectively more performant and have smaller binaries than these web apps in Halloween costumes…