Right now we are "OpenRouter for Images", with video following this week.
Our north star is creating a broader developer platform for AI media generation that includes observability, with fine-tuned vision models as a judge to monitor production traffic.
We also have a model arena and showdown page that ranks models by task, so you can find the best model for e.g. photorealism: https://lumenfall.ai/leaderboard
Our stack is Rails for the dashboard and Cloudflare Workers (Typescript / Hono) for the engine.
A developer platform for AI image generation that includes observability, with fine-tuned vision models as a judge to monitor production traffic. (Still working on the last part.)
We also have a model arena and showdown page that ranks models by task, so you can find the best model for e.g. making infographics: https://lumenfall.ai/leaderboard
We just launched the MVP. Tech stack is Rails for the dashboard and Cloudflare Workers (Typescript / Hono) for the gateway.
How does this relate to the recent changes of not being able to use a Claude Code Max subscription outside of Claude Code anymore? I assume technically this would work with Gigacode but would still be against their ToS?
# Enable on-device summaries: Users on macOS 26 have the option to have web content summarized on their device, so that web contents are not sent to our servers.
Another Google Maps request: Places I‘ve labelled used to at least sometimes (if inconsistently) show up in search. Now they never shop up, even when I type the exact name of the label.
A combination of good pricing, sane behavior and an offering with many TLDs right now is https://www.netim.com, based in France. Their UI is quite outdated, but it works ok. I've switched here after leaving Gandi.
If you're fine with a US-based provider, https://porkbun.com/ also has good pricing and a tech oriented mindset. They don't support many ccTLDs though.
In general, https://tld-list.com/ is the best place to research domain registrars in my opinion.
Where does the point of being disagreeable come from and what purpose does it serve in the business world, in your opinion or according to these books?
You need at least one or two somewhat disagreeable folks in a team. Because without this, groupthink emerges, teams have too much inertia, they follow the assumed norm instead of challenging it for something better, they don’t debate the options enough. That disagreeable energy, in the right dose, leads to better decisions. If you don’t have it naturally, you can encourage someone to “play devil’s advocate” in decision discussions (or do it yourself) and you’ll find sometimes the devil’s advocate is actually right.
This has also been my biggest gripe with Gemini 2.5 Pro. While it is fantastic at one-shotting major new features, when wanting to make smaller iterative changes, it always does big refactors at the same time. I haven't found a way to change that behavior through changes in my prompts.
Claude 3.7 Sonnet is much more restrained and does smaller changes.
This exact problem is something I’m hoping to fix with a tool that parses the source to AST and then has the LLM write code to modify the AST (which you then run to get your changes) rather than output code directly.
I’ve started in a narrow niche of python/flask webapps and constrained to that stack for now, but if you’re interested I’ve just opened it for signups: https://codeplusequalsai.com
Would love feedback! Especially if you see promising results in not getting huge refactors out of small change requests!
Interesting idea. But LLMs are trained on vast amount of "code as text" and tiny fraction of "code as AST"; wouldn't that significantly hurt the result quality?
Thanks and yeah that is a concern; however I have been getting quite good results from this AST approach, at least for building medium-complexity webapps. On the other hand though, this wasn't always true...the only OpenAI model that really works well is o3 series. Older models do write AST code but fail to do a good job because of the exact issue you mention, I suspect!
Having the LLM modify the AST seems like a great idea. Constraining an LLM to only generate valid code would be super interesting too. Hope this works out!
Asking it explicitly once (not necessarily every new prompt in context) to keep output minimal and strive to do nothing more than it is told works for me.
Really? I haven't tried Gemini 2.5 yet, but my main complaint with Claude 3.7 is this exact behavior - creating 200+ line diffs when I asked it to fix one function.
One idea I‘ve been toying with is to set up a non-profit registrar funded by donations. It would just pass through the registration fees from the registries without markup.
Similar to Let‘s Encrypt but for domains. Of course it would be more complicated because the entity would handle money, but nothing that couldn‘t be solved from what I can see.
Does anyone with experience in the field have any insights on what roadblocks would be encountered?
What would your pitch to donors be? Let's Encrypt was transformative by combining free and automated. Would your registrar be doing anything besides subsidizing domains for people who are willing/able to pay $9.15 (.com price without markup) but not $10.37 (Porkbun .com price)?
Also, registrants will not understand that you are just an intermediary who is making no profit. Since they paid you, they will hold you responsible for any problems they have, and won't shy away from disputing credit card transactions if you don't provide the level of support they expect. That's a lot harder to deal with than a free service, which can get away with providing no support.
> Buy and renew domains through Cloudflare Registrar at cost, without markup fees. You only pay what is charged by registries and ICANN
The problem is, running a registrar is complicated. It costs money. People get angry with you when things break - especially if they've paid you money.
You also have to deal with abuse reports, copyright complaints, and a legal demands. Good luck finding people who want to volunteer for that particular job!
Cloudflare does sell domains at cost but the catch is that domains registered through them must use CFs nameservers, so they can try to upsell you to their paid services. In that regard they are like most other registrars which treat domains as a loss leader for other products, except they take it to the limit of making nothing on the domains rather than a few cents.
I don't have a lot of deep experience here, but I registered by first domain the old fashioned way, send an email, get an invoice from NetworkSolutions, and mail it back with a check. And I setup two low volume OpenSRS reseller accounts for two companies in 2000, IIRC.
Registrar markup is not really that much. As I understand it, PIR, the operator of the .org registry charges registrars $9.05 / domain year [1], and low cost registrars typically charge end users $10 / domain year. I know there was a dust up over PIRs management recently, but I don't remember the details and couldn't find them in a quick look; if these aren't the actual numbers, they're pretty close. After payment processing, costs of included registrar provided services, customer service and operations, there's not really huge profits being made by registrars; providing service as a pass through at direct cost wouldn't be that compelling.
You'd need some other reason to encourage people to use your services, but I'm not sure what that would be. I've used specialty registrars at work, and they've got features like presence services where they have real people in the jurisdiction that satisfy the requirements of TLDs that require someone in the country to register a domain and corporate registrars that will work with the registry to enable registry locking that makes it incredibly difficult to change domain settings [2]. These are compelling features for the right kind of customer, but I don't think it makes sense for a non-profit to provide them.
[2] We moved to one of those after the current flavor of Network Solutions was phished and an unauthorized person used a customer service account to change our domain's glue records as well as some others; with registry lock, no changes can be made by the registrar unless the registry unlocks the domain after doing a song and dance routine with the end customer --- not very convenient, especially when the authorized person ignores the call to dance, but better than when a registry employee can get phished and change our domain without our consent
Exactly. I run a registrar price comparison service, and a lot of domain extensions are basically sold ‘at cost’. The catch is that a) a registrar will sell some tlds with a tiny/no margin, but charge a healthy extra for other tlds (like a ‘loss leader) and/or b) get you on the renewals (or c) keep increasing fees once their number of customers is high enough and hope most of them won’t notice for some time, double or triple of regular fees is not uncommon!).
So even though margins are low, as a customer it still makes sense to shop around.
Ps https://dot.bs is the service I run to compare tld registration and renewal prices
That's a good, but separate question. My point is an at cost registrar probably saves people 10% on a .org, which doesn't meaningfully increase access... If you can't afford $10/year, $9/year isn't really affordable either.
To your question though, I think PIR actually contracts out the operation of the registry to Affilias. I don't know what the current rate is, but before they renegotiated, they were paying about $3/domain to Affilias [1] based on a reported payment of $33M on just under 11 million domains.
I don't really know where the rest of the money goes. There were a lot of questions when PIR tried to sell .ORG to private equity in 2020, but I don't know if there was much follow up after the deal got quashed.
PIR is owned by the Internet Society, a 501(c)(3) nonprofit which does a bunch of other stuff, like running the IETF. Most of the Internet Society's revenue comes from selling .org domains.
It doesn't. It could hypothetically cost PIR $0. 501(c) organizations are not prohibited from collecting and spending profits, they are prohibited from unreasonably distributing such earnings to private shareholders and individuals (see https://www.irs.gov/pub/irs-tege/eotopicc90.pdf).
Right now we are "OpenRouter for Images", with video following this week.
Our north star is creating a broader developer platform for AI media generation that includes observability, with fine-tuned vision models as a judge to monitor production traffic.
We also have a model arena and showdown page that ranks models by task, so you can find the best model for e.g. photorealism: https://lumenfall.ai/leaderboard
Our stack is Rails for the dashboard and Cloudflare Workers (Typescript / Hono) for the engine.
reply