A year and still no implementation for such a basic need as offloading MoE layers onto the CPU selectively. On llama.cpp I can get models like Qwen 35BA3B running partially on gpu/cpu with 40t/s on a laptop thanks to --n-cpu-moe but on this VC funded joke it would be simply unusable.
I can't quite understand how you make a wrapper so much worse than the code you're ripping out.
>This funding is fuel for what’s ahead. Ollama sits front and center in the open model ecosystem
No.
fnlsnd 1 days ago [-]
Will Ollama finally properly credit llama.cpp ? Or will they continue to rebrand new llama.cpp features as their own?
mrcwinn 17 hours ago [-]
I’m sure you know.
mottosso 13 hours ago [-]
A lot of hate for Ollama in this thread. Am I understanding correctly that it is all based on not crediting llama.cpp correctly? I can't find any other argument against it, other than some minor performance differences. Where Ollama stands out is user interface. It's easy to overlook how difficult and important that is.
Towaway69 7 minutes ago [-]
I dumped ollama yesterday after reading a similar thread. I definitely don’t condone what ollama has done but yes, definitely their UX is far better - as an end user.
Their docker-like extending and defining new models is really simple (coming from a docker background).
Installing new models with llama.cpp seems nontrivial. Perhaps it was using the mac dmg instead of curl-pipe-sh.
Unfortunately then llama.cpp crashed my laptop - no error, no feedback, just frozen desktop and reboot.
Went back to a much slower experience with ollama but it just works out of the box… sorry to say.
entropyie 1 days ago [-]
To any ollama.com dudes and dudettes reading this: why won't you offer any embedding models on your subscription when they are readily supported by the software and fairly efficient to run?
woadwarrior01 1 days ago [-]
I think the "8.9 million developers use ollama" figure they throw around is contrived. I found this 2025 article which posits that there were 47.2 million software developers in the world. Even the number of developers has doubled in a year (plausible if you account for the rise in vibe coding), I think it's very unlikely that 1 in 10 developers use ollama.
No no no…. You see, with AI, everyone and their dog is a developer now. 8.9 million looks pretty pathetic when considering the global computer-using population.
khurs 1 days ago [-]
>Today, we’re announcing that Ollama has raised money from VCs...
Note to self, Enshittification ahead. Don't use any ollama services unless it's calling an industry standard api.
Cool to learn that the founders also had a hand in developing Docker Desktop in the past!
allanrbo 16 hours ago [-]
Does anyone actually like Docker Desktop? I much preferred Docker before it had weird UI parts to too. Have since switched to Podman.
Dansvidania 15 hours ago [-]
on macOS it was horrible. I have been using orbstack for a while now and it has successfully stayed out of my way.
trilogic 1 days ago [-]
How did they raise that amount of money if they are so hated, I see only bad comments everywhere about ollama. Not a fan of them myself, but they played a good part in local ai since the beginning. The investors screw everything: Ollama announced an $88 million financing on July 9, 2026. The named participants were:
Investment firms and organizations
Benchmark — represented by Peter Fenton
Theory Ventures — Tomasz Tunguz
8VC — Alex Kolicich
Y Combinator
Garage Capital
Pace Capital
49 Palms
GTMFund
Individual investors
Solomon Hykes — Docker founder
Aaron Katz — ClickHouse CEO
Spencer Kimball — Cockroach Labs co-founder and GIMP co-creator
Quinn Slack — Amp CEO
Marianna Tessel — Cisco board member
Michael Montano — former Twitter head of engineering
Other unnamed angel investors
>like lmstudio and google/alphabet maybe!
Investors are not free
gkapur 1 days ago [-]
Peter Fenton invested in the company when it was called Infra.App originally (I think it's still online: https://infra.app/). It was an access management product, then it became desktop Kubernetes product, both they pivoted out of. I imagine he invested based on the technical acumen of the founders and then worked with them to pivot when Ollama started working out. Almost all of the other investors invested in the founders based on that original idea space (which wasn't a great one but was quite adjacent.) I think that includes 8VC, YC, Pace Capital, GTMFund, who all followed on here.
Since it pivoted, I imagine the story they are telling is the usage. But you can add to that Peter Fenton has had incredible success in commercial open source (Docker-ish, Elastic, JBoss, CockroachDB, TimeScaleDB, and if you include Benchmark there is Elastic, Confluent, etc.) so a bunch of these people are probably trying to ride the Peter Fenton/Benchmark train. My guess is that's what Theory is doing.
Investors are not free but they are very hype driven. There are < 10 investors who are truly experts in open source in my opinion in the market (and maybe < 5.) It's a really wacky cadence and I would argue there are different types of OSS businesses, which makes it even more difficult to be an expert. It's something I struggle with personally investing in OSS -- I think I understand the likely motion for an OSS startup and then > 50% of the time I am wrong.
tredre3 1 days ago [-]
Why blame the investors? You went out of your way to even name them so we can bash them, why? Ollama founders sought them. Ollama founders wanted the money. Ollama founders took the money. Ollama founders took all the decisions that made people now dislike Ollama. Nobody forced their hands, it was their plan all along...
trilogic 1 days ago [-]
That´s the whole point, once you take the money, not your decision anymore. We also received good offers at Hugston (over 3 million dollars) to buy our decision making but we didn´t take it. It is not easy to deal with the team and the perhaps but we kept our principles.
Nothing against investors, but once they "invest" they are all over your neck and board of administrators weekly. There finishes the joy/innovation and starts the unpleasant route.
We will go out with the new version of HugstonOne (in the coming week) which is way superior and very powerful to everything worldwide right now for Local AI and Privacy and features, but users are 100% ín control, NO TRICKS.
ekianjo 1 days ago [-]
dont use Ollama. llamacpp is better in every way.
h2aichat 1 days ago [-]
Would anybody here recommend LM Studio?
twsted 1 days ago [-]
Another chance to tip their hat to llama.cpp, missed. I really don't get it.
wazoox 1 days ago [-]
Annoyingly many useful tools like the "Page Assist" browser extension or "Murmure" are plug-and-play only with Ollama...
lioeters 1 days ago [-]
> The personal computer moment for AI
Georgi Gerganov, the author of llama.cpp, is the real hero here.
I really wish they would figure out a better way to ship it for Mac. Last time I tried it a few months ago, it spewed a bunch of stuff into the system's pip archive and broke other software.
18 hours ago [-]
fwipsy 1 days ago [-]
Unsloth has better quants, but when I tried their studio a couple months back it seemed like a buggy mess. Ollama seemed very basic but low-frustration.
I wasn't aware of the politics behind Ollama but I have noticed the lack of variety in quant versions, the lag between availability on Ollama, and their push to get you to use their cloud versions.
I'm going to uninstall Ollama.
Grombobulous 17 hours ago [-]
This article is really helpful for finding alternatives. I was already using LMStudio but I love that I now know about a couple of true open source alternatives I can try out.
aceazzameen 1 days ago [-]
I don't use local models often enough, but this convinced me to uninstall ollama. Now I need to figure out which of the options at the end of that post is the best alternative for me.
el_io 1 days ago [-]
llama.cpp?
Ollama is/was based on that.
sillystuff 21 hours ago [-]
Just adding to your suggestion.
For those new to llama.cpp, the llama-server component running in "router mode" is fantastic; you place your common configuration in a defaults section of an ini file, and then model specific configuration in separate sections of the ini file. You can switch between models, on the fly, using the API, and the built-in web front-end for llama-server supports model switching as a drop down menu. Llama-server can just always be running, in the background (you can set an idle time for it to automatically unload weights to free vram/ram, and a maximum simultaneously running model count which will evict models on an LRU basis before loading new ones, if the count is exceeded).
llama-bench is another great tool included in llama.cpp, where you can provide ranges of options to test, and just let it run through all the variations that you requested, then take the best performing and add those settings to the ini file above.
There are a lot of optional tunables that can really improve performance with llama-server. E.g., typical average tok/sec generation speed improved over 7X for Gemma4 31B, on my laptop, with tuned batch-size / ubatch-size and enabling both spec-type draft-mtp and ngram-mod (with some custom settings on both). I.e., even though you don't need to do this for llama-server to be useful, it can be worth your while to spend a little time optimizing settings for your favorite models especially if your hardware is marginal.
thot_experiment 1 days ago [-]
Came here to post an anti-ollama rant, real glad this is already the top comment. Nature is healing.
It's not my favorite but we do have to be aggressive in fighting these gross attempts for vc-backed silicon valley non-contributors to exploit open source labor.
electroglyph 15 hours ago [-]
i handed out so many upvotes here, good work everybody!
arikrahman 1 days ago [-]
Immediately thought of this article.
fragmede 1 days ago [-]
If that's what I'd have to do to get $88 million, I don't want $88 million.
verdverm 1 days ago [-]
I've been very happy with vLLM as an alternative on the serving side, and OpenCode Go on the subscription side
senectus1 20 hours ago [-]
I had no idea about any of this. thanks. Might give some of the alts a go.
moralestapia 1 days ago [-]
Yeah, they're trash.
Looks like it's their second time, as Kitematic was a very thin wrapper over the Docker CLI.
kypro 1 days ago [-]
I see a lot of these "don't use [insert product]" posts and it's normally always BS about how some exec at the company has the wrong political views or upset some community who was the product by taking it in a different direction from what they wanted.
I think this is probably first time ever I read one of these and was convinced immediately that I should probably look into using something else.
inigyou 1 days ago [-]
This isn't different in kind from any of the other ones you disagreed with, so maybe ponder why you agree with this one.
halJordan 1 days ago [-]
Or, you know. Don't tell me what to do. One dirty hand washes another and many of llama.cpp's new features are straight copies of ollama functionality. You do you of course. But stop denying them their contribution to this ecosystem and stop denying me my agency to salve your politics.
abtinf 18 hours ago [-]
I have no opinion or knowledge of the claims in that link, nor will I research them. The entire thing is obviously claudeslop.
zishha1b0njq 1 days ago [-]
This post is empty of any relevant content.
I wish HN would stop sharing drama-seeking posts, especially by anonymous authors, attacking the hard work of people.
Are ollama's authors perfect? No, they are just humans. But many people find their work valuable and we should support Open Source.
I can assure you that you won't like the world where nobody release Open Source projects anymore because of the constant negativity.
pyaamb 1 days ago [-]
Do you see any negativity towards llama.cpp? This isnt drama-seeking. The author makes very valid points.
pierotofy 1 days ago [-]
Taking credit for other people's work while raising millions? Yeah, pretty bad and relevant.
inigyou 1 days ago [-]
See also: why friends don't let friends use permissive software licenses.
The llama.cpp authorized ollama to do this... Unintentionally.
msm_ 23 hours ago [-]
Literally the main point of this blog post:
>The project’s binary distributions didn’t include the required MIT license notice for the llama.cpp code they were shipping. This isn’t a matter of open-source etiquette, the MIT license has exactly one major requirement: include the copyright notice. Ollama didn’t.
inigyou 9 hours ago [-]
How sure are we that it didn't include the notice? There isn't any prescribed form or location for the notice, so you're trying to prove a negative.
Krasnol 1 days ago [-]
This anonymous author made some good points, supported by good sources.
It looks more like you're the anonymous author attacking the hard work of the author of this blog post.
The blog post alone has more relevance then the ad-post.
ojinai 16 hours ago [-]
[flagged]
JSR_FDED 1 days ago [-]
The VCs listed in the article aren’t the dumbest ones around. Why on earth would they invest in a shell around the actual technology?
Were they so enraptured with the “docker for LLMs” line?
mrcwinn 17 hours ago [-]
Do not use Ollama.
rvz 1 days ago [-]
> The local LLM ecosystem doesn’t need Ollama. It needs llama.cpp. The rest is packaging, and better packaging already exists.
Then put those LLMs to work and build an open source alternative to everything that Ollama is doing.
After all, open source is a pricing weapon to race everything to $0.
m_kos 1 days ago [-]
I get that people here have reasons to hate on Ollama but it has two genuinely strong points:
- it makes it very easy to use open models with a lot of harnesses and assistants via `ollama launch`;
Their $20 subscription is very generous, and they claim not to log or train on your data.
I will be curious to see if their subscription ever supports the ~3T open weights models announced this week.
eastbound 1 days ago [-]
> $20 subscription is very generous, and they claim not to log or train on your data.
The principle of running locally is that you’re not in the cloud.
A year and still no implementation for such a basic need as offloading MoE layers onto the CPU selectively. On llama.cpp I can get models like Qwen 35BA3B running partially on gpu/cpu with 40t/s on a laptop thanks to --n-cpu-moe but on this VC funded joke it would be simply unusable. I can't quite understand how you make a wrapper so much worse than the code you're ripping out.
>This funding is fuel for what’s ahead. Ollama sits front and center in the open model ecosystem
No.
Their docker-like extending and defining new models is really simple (coming from a docker background).
Installing new models with llama.cpp seems nontrivial. Perhaps it was using the mac dmg instead of curl-pipe-sh.
Unfortunately then llama.cpp crashed my laptop - no error, no feedback, just frozen desktop and reboot.
Went back to a much slower experience with ollama but it just works out of the box… sorry to say.
https://www.slashdata.co/post/global-developer-population-tr...
Note to self, Enshittification ahead. Don't use any ollama services unless it's calling an industry standard api.
Investment firms and organizations
Benchmark — represented by Peter Fenton Theory Ventures — Tomasz Tunguz 8VC — Alex Kolicich Y Combinator Garage Capital Pace Capital 49 Palms GTMFund
Individual investors
Solomon Hykes — Docker founder Aaron Katz — ClickHouse CEO Spencer Kimball — Cockroach Labs co-founder and GIMP co-creator Quinn Slack — Amp CEO Marianna Tessel — Cisco board member Michael Montano — former Twitter head of engineering Other unnamed angel investors >like lmstudio and google/alphabet maybe!
Investors are not free
Since it pivoted, I imagine the story they are telling is the usage. But you can add to that Peter Fenton has had incredible success in commercial open source (Docker-ish, Elastic, JBoss, CockroachDB, TimeScaleDB, and if you include Benchmark there is Elastic, Confluent, etc.) so a bunch of these people are probably trying to ride the Peter Fenton/Benchmark train. My guess is that's what Theory is doing.
Investors are not free but they are very hype driven. There are < 10 investors who are truly experts in open source in my opinion in the market (and maybe < 5.) It's a really wacky cadence and I would argue there are different types of OSS businesses, which makes it even more difficult to be an expert. It's something I struggle with personally investing in OSS -- I think I understand the likely motion for an OSS startup and then > 50% of the time I am wrong.
Georgi Gerganov, the author of llama.cpp, is the real hero here.
https://llama.app/
But please don't use ollama, or their quants. Not only is the app itself slower than pure llamacpp. But their quants are often no where near the best.
I really hope people start with something like unsloth, as their software and quants are really much better all around.
- https://sleepingrobots.com/dreams/stop-using-ollama/
I'm going to uninstall Ollama.
Ollama is/was based on that.
For those new to llama.cpp, the llama-server component running in "router mode" is fantastic; you place your common configuration in a defaults section of an ini file, and then model specific configuration in separate sections of the ini file. You can switch between models, on the fly, using the API, and the built-in web front-end for llama-server supports model switching as a drop down menu. Llama-server can just always be running, in the background (you can set an idle time for it to automatically unload weights to free vram/ram, and a maximum simultaneously running model count which will evict models on an LRU basis before loading new ones, if the count is exceeded).
llama-bench is another great tool included in llama.cpp, where you can provide ranges of options to test, and just let it run through all the variations that you requested, then take the best performing and add those settings to the ini file above.
There are a lot of optional tunables that can really improve performance with llama-server. E.g., typical average tok/sec generation speed improved over 7X for Gemma4 31B, on my laptop, with tuned batch-size / ubatch-size and enabling both spec-type draft-mtp and ngram-mod (with some custom settings on both). I.e., even though you don't need to do this for llama-server to be useful, it can be worth your while to spend a little time optimizing settings for your favorite models especially if your hardware is marginal.
It's not my favorite but we do have to be aggressive in fighting these gross attempts for vc-backed silicon valley non-contributors to exploit open source labor.
Looks like it's their second time, as Kitematic was a very thin wrapper over the Docker CLI.
I think this is probably first time ever I read one of these and was convinced immediately that I should probably look into using something else.
I wish HN would stop sharing drama-seeking posts, especially by anonymous authors, attacking the hard work of people.
Are ollama's authors perfect? No, they are just humans. But many people find their work valuable and we should support Open Source.
I can assure you that you won't like the world where nobody release Open Source projects anymore because of the constant negativity.
The llama.cpp authorized ollama to do this... Unintentionally.
>The project’s binary distributions didn’t include the required MIT license notice for the llama.cpp code they were shipping. This isn’t a matter of open-source etiquette, the MIT license has exactly one major requirement: include the copyright notice. Ollama didn’t.
It looks more like you're the anonymous author attacking the hard work of the author of this blog post.
The blog post alone has more relevance then the ad-post.
Were they so enraptured with the “docker for LLMs” line?
Then put those LLMs to work and build an open source alternative to everything that Ollama is doing.
After all, open source is a pricing weapon to race everything to $0.
I will be curious to see if their subscription ever supports the ~3T open weights models announced this week.
The principle of running locally is that you’re not in the cloud.