I recently upgraded to a desk of the same size, it's great! I have a habit of sitting cross legged and my legs were always bumping into my old desk whose legs were spaced too close together.
I had to the wooden top cut at a home center and then I sanded it and applied a finish myself because I couldn't find anywhere local that sold a desk top that size. Combined with a dual motor standing desk set of legs, it turned out great!
I also 3D printed brackets for all the power bricks of various devices I have and mounted them on the underside.
I was actually planning to write a small blog post about it because some parts are specific to where I live, but I'm sure there are plenty of articles already around. Just look up "custom standing desk" or "countertop/butcherblock desk".
The woodworking was fun but you're basically screwing a big piece of wood to some pre-made metal legs that have motors to raise it up and down.
In proxy mode, here's what forge applies on each request (handler.py builds these):
Response validation: ResponseValidator(tool_names) checks each tool call against the declared tools array. If the model emits a call to a name not in tools[], or a malformed call shape, it's caught before the response goes back.
Rescue parsing: When the model emits tool calls in the wrong format — JSON in a code fence, [TOOL_CALLS]name{args} (Mistral), <tool_call>...</tool_call> (Qwen XML) — rescue parsers extract the structured call and re-emit it in the canonical OpenAI tool_calls schema. This is the biggest practical lift, especially on Mistral-family models that ignore native FC and emit their own bracket syntax.
Retry loop with error tracking: ErrorTracker(max_retries=N) — if validation fails, forge retries inference up to N times with a corrective tool-result message on the canonical channel, rather than returning a malformed response to your caller. From your perspective the proxy looks like a single request that just took a few extra ms.
What proxy mode does NOT do (because it's single-shot, not multi-turn): prerequisite/step enforcement (those need a workflow definition spanning turns), context compaction, session memory. For that surface you wrap the WorkflowRunner class in Python — proxy mode trades that depth for "use forge with your existing setup, no Python rewrite."
So yes — the proxy is fortifying the response shape and retry behavior of /v1/chat/completions. The full agentic guardrails are at the Python class level above it.
For greenfield projects, I've been building on forge native using WorkflowRunner so I get all guardrails. But obviously as a drop-in replacement in existing systems then proxy is the way to go.
the funniest thing I see in opencode with tool calling is the model calls 10.0 and opencode says it's an error because the spec is an integer, even though it's obvious to anyone that if a float can be coerced properly to a integer, then that should be a success.
Yeah it's a delicate balance between precise and silly, and too permissive.
I'm definitely still iterating on forge, but so far sending the model a friendly and gracefully handled error message works wonders (instead of barfing a stack trace or something).
> They've essentially gotten roped into maintaining a huge chunk of internet infrastructure, for free. If they ever shut it down the whole world would end up rioting because it's so widely used.
Not even remotely true. They regularly shut down products and services with impunity. If Gmail cost more than the data they directly or indirectly mine and sell from their users, Gmail wouldn't exist either.
Shutting down GMail would practically amount to shutting down email. It's by far the largest email provider in the US (and probably in the world but I don't have that data). There's no other provider who could take up the slack; if it were to abruptly shut down, a lot of users would simply lose access to email altogether.
They'd generate a huge amount of ill will by shutting it down, and that in turn would likely lead to a nontrivial share of people moving away from Google core products (like search) out of pure spite.
To what? Google Search sucks thanks to the idiot who ran Yahoo into the ground, but everyone else sucks more. Every time I try to use non-google search the results are virtually useless.
Google has firmly been in the "we're so big we can suck at everything, but you'll still use our stuff because you have no other choice" phase that Microsoft was (is?) in.
They've dominated email so much that their spam filter makes it a very risky proposition to run your own domain; chances are very good it'll just start dropping your messages. Even if chances aren't great, can you take the risk of an important email getting zapped?
To this day I still routinely have to fish out my gmail spam folder dozens of emails from various open source mailing lists that have been around for a decade or two, some hosted on kernel.org, because the spam filter is convinced they're spam. Google is too fucking stupid or lazy to whitelist sites like kernel.org.
FFS even google groups I'm in that are technical get obviously-not-spam messages tagged as spam!
No, that wouldn't happen. Lots of people don't have email through Google, for one. Those people will still use email just fine. Moreover, the people who do use Gmail will simply sign up with another provider. It won't be a big deal.
> No, that wouldn't happen. Lots of people don't have email through Google, for one.
Based on some data I collected around five years ago, roughly 80% of US customers used GMail for personal email. It was overwhelmingly the most common choice. I suspect that number has only drifted upwards since.
(What about the rest? 15% were using Yahoo; the rest were spread thinly across AOL, Microsoft, ISPs, and colleges.)
At one point AOL was the largest ISP and email provider on Earth too. If gmail died off people would just move to something else. It'd be annoying, but it wouldn't be the end of email
Google could actually do everyone a solid by killing gmail. They have enough influence in the industry that they could create a standard for email address portability, and then slowly force everybody to move off. By the end, one of the biggest problems with email would be solved and people would be able to switch email providers like how we can switch phone providers without needing to change our phone numbers. And Google would get to save a lot of money by no longer needing to provide everyone's emails
In a better time I would expect the government to step in a acquire this fundamental service and fund it with tax money. Right now? The only intervention I would expect is a massive subsidy to pay Google to keep providing it, while also letting them continue to spy on everyone's mail (which is a crime, but not if the mail is on a computer, I guess).
Why is this inconceivable? I don't know where you live, but the Post Office is extremely cheap and reliable around here. What drives you to pretend that states can't provide services to their people?
An excellent example of how not to do a government program!
> On October 1, 2013, HealthCare.gov was rolled out as planned, despite the concurrent partial government shutdown. The launch was marred by serious technological problems, making it difficult for the public to sign up for health insurance.[4] The deadline to sign up for coverage that would begin January 1, 2014, was December 23, 2013, by which time the problems had largely been fixed. The open enrollment period for 2016 coverage ran from November 1, 2015, to January 31, 2016.[5] State exchanges also have had the same deadlines; their performance has been varied.[6][7][8]
> The design of the website was overseen by the Centers for Medicare and Medicaid Services and built by a number of federal contractors, most prominently CGI Inc. of Canada. The original budget for CGI was $93.7 million, but this grew to $292 million prior to launch of the website. While estimates that the overall cost for building the website had reached over $500 million prior to launch[1][9][10][11][12] and in early 2014 HHS Secretary Sylvia Mathews Burwell said there would be "approximately $834 million on Marketplace-related IT contracts and interagency agreements,"[13] the Office of Inspector General released a report in August 2014 finding that the total cost of the HealthCare.gov website had reached $1.7 billion[14] and a month later, including costs beyond "computer systems," Bloomberg News estimated it at $2.1 billion.[15]
Got it. So if you're fighting an obstinate faction that would rather the government not exist than provide services then that can cause issues. Further, contractors will fleece you for everything you're worth. Compare to a successful project like the Post Office that gets pushed through with overwhelming political will and is run directly by a government agency (oddly structured as a government-owned corporation) and then even despite attempts to destroy it it continues to provide good service.
It's not easy; you need someone competent heading it up and setting it up for success. If the Democrats were to propose it in 2028 under president Gavin I would expect it to be a boondoggle. That doesn't change the fact that I want it to be done and done well.
Depends on the health of our institutions. In the US at least they're legally obligated not to by the highest law in the land. It gets ignored now, but it's a more promising path to privacy-preserving digital infrastructure than letting the private market handle it.
Unfortunately, the Constitution has been flagrantly ignored by the federal government for close to 100 years now, if not longer. Everything that FDR did was blatantly unconstitutional, but nobody stopped him, nor did they roll it back when he was gone. The Constitution has no real practical power to restrain the government if the people don't exercise their rights as voters to hold it accountable, and it is abundantly clear that the unconstitutional stuff the government gets up to is (largely) actually pretty popular.
bullshit, email exists outside of gmail, and email would continue to exist without it. many would have to get a new account somewhere, but that would be not a problem. there are shitloads of providers that would be quite happy
it's insane to frame anything a company like Google does as some kind of goodwill. rather than an amoral profit optimization. contrary to OP, what people often overlook about GMail is not their "plight". but the powerful brand awareness it creates
Yes it is remotely true. Name one thing they have shut off that a large number of people actually used and it was important. We all joke about Google dropping things and yes they have, but saying they can just drop Gmail is.. well, insane.
The reason people mention importance is because corps like Google don't just care about per-product profitability, they assess how one product affects the rest of their business.
A few days ago I switched again from Qwen3.6 to Gemma 4 - for personal use I've experienced better average performance with the 26B version of the latter than the 27B of the former.
For someone who's been running local models for a long while, these are very very exciting times.
Oh that's fascinating. 3.6 27B is pretty damned good, but slow in wall-clock times on my DGX Spark-alike. It generates huge reams of thinking before it gets the (usually correct!) answer, so wall-clock time is rough for tasks even at ~20tk/s
I'm surprised the 26B-A4B is better? It should be faster too, interesting. I'm excited to try 31B with MTP, because MTP-2 is what makes 27B bearable on the GB10.
What are you using it for? Agent-based coding, or something else?
General purpose, mostly internet research in the form of slow-crawling. (Emphasis on slow - I've ultimately landed on Scrapling's API for seamless content rendering, and I use image support so as not to exclude informative images or weirdly rendered text.)
For coding I don't need image support so I stuff the entire GPU with text-only mode. I don't have a workflow where I send LLMs off to generate thousands of lines of code but what little coding I did I did with Qwen3.6 and it was spectacular, as you likely suggest.
I've been thinking about doing more of this too. What spec machine are you running? And are you using long-running autonomous agents or more of the IDE/co-pilot style of collaboration?
Gemma certainly was trained for tool calling, but the implementation in llama.cpp has been troubled because Gemma uses a different chat template format. The processor from the transformers library works fine though.
I'm using llama.cpp with Gemma and tool calling is mission critical. It's perfectly fine on my end.
There are definitely differences in the eagerness to tool-call that you'll need to manage. And for all local models I've ever used, I've had to micromanage the tools provided by servers to eliminate any possibility that they reach for something wonky or confusing.
> However I find qwen unbeatable for toolcallling. I think gemma wasnt trained on that at all.
Gemma4 chat template seems to had multiple issues, at least with llama.cpp, not sure they're all fixed yet. It assumed simple types for parameters for example.
Also, embedded servers are now much much much more popular. Stuff an HTTP server directly into your application and do whatever you gotta do without gateways.
That is way! Unfortunately, sometimes you have to do path-based routing to different backends, and now you're back to needing a proxy between your clients and your applications.
This is the way only if you're operating in a trusted environment (eg. homelab, intranet) or you're sticking CloudFlare or some other "reverse proxy as a service" in front of it. If you expose an embedded HTTP app server directly to the Internet you're almost guaranteed to get pwned, as the Internet has now become an extremely hostile place.
These are often not enough ‘battle-tested” and come with a warning to never expose to public internet. So then you put a WAF in front of it, and you are back to HTTP reverse proxy setup.
Just wondering, once an 'AI Model of Some Form', is in a Physical Body a 'robot', and is provided with some rules about survival so it doesn't fall into a hole. After a series of these events, does it matter? Does mimicry become reality, or no longer differentiable.
Kind of the philosophical zombie argument. If a robot can perfectly mimic a human, can you really know the internal state of the 'real' one is different from the 'mimicked' one.
The paper isn't concerned specifically with survival. It's saying that you cannot achieve "abstraction" (presumably the structure that underlies critical thinking, creativity, etc.) through shear mimicry.
Again, just echoing the paper here. I don't know that I'm doing it justice.
reply