Setting it up was easy enough, but just as I was about to start linking it to some test accounts, I noticed I already had blown through about $5 of Claude tokens in half an hour, and deleted the VPS immediately.
If you have an old M1 Macbook lying around, you use that to run a local model. Then it only costs whatever the electricity costs. May not be a frontier model, but local models are insanely good now compared to before. Some people are buying Mac Minis for this, but there's many kinds of old/cheap hardware that works. An old 1U/2U server some company's throwing out with a tech refresh, lots of old RAM, an old GPU off eBay, is pretty perfect. MacBook M1 Max or Mac Mini w/64GB RAM is much quieter, power efficient, compact. But even my ThinkPad T14s runs local models. Then you can start optimizing inference settings and get it to run nearly 2x faster.
(keep in mind with the cost savings: do an initial calculation of your cloud cost first with a low-cost cloud model, not the default ones, and then multiply times 1-2 years, compare that cost to the cost of a local machine + power bill. don't just buy hardware because you think it's cheaper; cloud models are generally cost effective)
Yeah, I looked at Clawdbot / OpenClaw at the beginning of the week (Monday), but the token use scared me off.
But I was inspired to use Claude Code to create my own personal assistant. It was shocking to see CC bang out an MVP in one Plan execution. I've been iterating it all week, but I've had it be careful with token usage. It defaults to Haiku (more than enough for things like email categorization), properly uses prompt caching, and has a focused set of tools to avoid bloating the context window. The cost is under $1 per check-in, which I'm okay with.
Now I get a morning and afternoon check-in about outstanding items, and my Inbox is clear. I can see this changing my relationship to email completely.
A lot of the system prompt, skills and tools center around my specific needs (I manage separate IMAP and Gmail inboxes, use Granola, and have iCloud calendars). And there are some hard assumptions baked in (I want to have a morning & afternoon check-in). It probably wouldn't be useful as-is, but maybe as inspiration?
I'd love to see even a filtered version of it. I've been doing very similar things with an "everything" database. That's been my own personal northstar.
BTW, OpenCode has free Kimi (I haven't hit a quota yet) right now and it's done pretty great things for me in the last 24 hours.
They're neck and neck for me, in terms of PRDs, coding, and web searching. CC built the bulk of my current project, I did a lot of analysis of it with Antigravity (the interface is esp good for reviewing/commenting on long .md output files) and then, after building a simple roadmap of v2 features, OpenCode + Kimi was the most aggressive about running in a fairly autonomous manner and finishing the items on said roadmap. OC was also pretty hardcore about misinterpreting a limit I expressed earlier in one context as a limitation in another context -- which was fine, I'd rather say "no, really, you can go do that, I'm giving you permission and here's what I meant before" than find out it was too brazen.
It's a lot like managing two experienced mid- to sr- engineers each of whom have slightly different personalities and intro/extro verted personalities. CC has more personality but OC wants to race. They can both code, but for disparate tasks you might pick the personality and posture of one person over the other.
I find myself picking daily tasks based on which of the tools I'm in the mood to sit with. But across a few days I sit with all three.
I wouldn't say it was oneshotted, but it did produce a working MVP in one Plan execution. Meaning, I went back & forth a few times about requirements, it built a plan, and then CC spent just under 15 minutes writing the code. Once I got the credentials plugged in, the core integrations (Slack, gmail, IMAP, iCloud calendar) and agent loop did work. I can share the initial message if you're curious.
I think one thing these things could benefit from is an optimization algorithm that creates prompts based on various costs. $$, and what prompts actually gives good results.
But it's not an optimization algorithm in the sense gradient descent is, but more like Bandits and RL.
I won't claim I understand its implementation very well but it seems like the only approach to have a GOFAI style thing where the agent can ask for human help if it blows through a budget
That's the sad thing. There are so many millions of talented under-employed people in the world that would gladly run errands or set up automations for you for $200-$1000 per month or whatever people are spending on this bot.
Developers trust lobsters more than humans.
The other wild thing is that many of these expensive automations that are being celebrated on X can already be done by voice using Siri, Google, or any MCP client.
part of me sympathizes, but part of me also rolls my eyes. Am i the only one that’s configuring limits on spend and also alerts? Takes 2 seconds to configure a “project” in OpenAI or Claude and to scope an api key appropriately.
Can you get meaningful work done with CC at $20 at a time? I load $20 at a time onto the API for general chatting purposes and it lasts a few months at a time. I've always avoided trying CC because I got the impression people were burning $100+/mo, which is beyond my personal hobby budget.
/Not a software engineer perspective working on side projects
I guess if you're letting it vibe code huge chunks. I'm doing mostly handwritten code for my current project with a little bit of "I don't want to deal with this, Claude can handle it" and I've spent $1.26 this month for my 446 lines of code.
But yes I suppose at that rate, if Gastown or Beads or whatever is 300,000 lines of code (just to use a project known to be fully vibe coded with rough LOC reported), that would be over $800.
Don't let it vibe code hundreds of thousands of lines of code I guess.
I was doing that initially, but I think the subscriptions are generally worth it for personal projects. $20/mo is good if you're like me and you can do this stuff maybe a couple nights a week, I haven't run into the limitations on that yet. The $100+ subscriptions are needed if you're doing it every day. YMMV
I'm successful with personal projects (reverse engineering USB devices, sledding spot finder, silly stuff) on the $20/mo Claude plan. I rarely use Opus except for planning larger things.
I keep a master llm.md file and rotate between Claude Code (Pro), Antigravity Opus, Antigravity Flash, and OpenCode Kimi. I don't actually mind hitting limits.. though I'm least happy when Opus goes away.
My entire process is to build a generic llm.md file that all the tools can use and record to. I don't want to be tied completely to any one solution. You can get pretty far without spending a lot on tokens. I can run almost continually, and presently I'm the bottleneck anyway.
For Claude Code, I now pay the $20/mo subscription for pro because I was spending more using it via API credits.
Even if I had to reload manually very often, I still would not enable auto reload. These APIs are crazy expensive and I'm not looking for a surprise bill.
not only that, but clawdbot/moltbot/openclaw/whatever they call themselves tomorrow/etc also tells you your token usage and how much you have left on your plan while you're using it (in the terminal/console). So this is pretty easily tracked...
Isn't that explictly against the TOS? I feel like Anthropic brought out the ban hammer a few days ago for things like opencode because it wasn't using the apis but the max subscriptions that are pretty much only allowed through things like claude code.
This was first reported by Apache.be, not exactly what I would call a zionist publication.
The speech in question was a speech given at the opening of the academic year at the second largest university in Belgium. De Sutter became rector last year. As part of her election platform, she warned about the dangers of AI in terms of plagiarism, bias and hallucinations - so in this context this is quite relevant.
First of all, you are conflating Hamas and Hezbollah.
Second of all, the stories about beheading of babies and mass rape on October 7, 2023 have been thoroughly debunked.
Third: the pager operation caused indiscriminate explosions at places where non-combattant citizens were present. Not very gentlemen-ly (to use your words), and indeed a war crime.
Fourth: What they did in Gaza is arguably worse than carpet bombing.
But the hundreds of concert goers who Hamas killed is very true. Remember how they paraded the broken body of that young German woman around like a disgusting hunting trophy?
How did you come up with that tally? Israel has refused to comply with requests from international investigators into the matter, likely because a lot of the casualties were due to IDF actions.
I remember the footage of "that young German woman" but it is to me extremely peripheral and did a lot less of an impression than the thousands of images of destroyed baby bodies I've seen that were caused by the IDF. The criminal actions perpetrated by palestinians on October 7th 2023 were pathetic compared to what the israelis have done for decades.
Claiming the IDF killed all of these people is a truly despicable lie that destroys your credibility.
On 7 October 2023, the al-Qassam Brigades, the military wing of the Palestinian nationalist Islamist political organization Hamas, initiated a sudden attack on Israel from the Gaza Strip. As part of the attack, 378 people (344 civilians and 34 security personnel) were killed and many more wounded at the Supernova Sukkot Gathering, an open-air music festival during the Jewish holiday of Shemini Atzeret near kibbutz Re'im. Hamas also took 44 people hostage, and men and women were reportedly subject to sexual and gender-based violence. Some 20 of the attackers were also killed by Israeli security forces in the area of the festival.
I did not write "all of these people". I pointed out that the state of Israel has refused to provide conditions for an investigation of what happened, and instead it has mainly been the press and leaks to the press that have shed light on the issue.
What we have known for sure since then is that the IDF brought helicopters to the area where the festival was held and Hellfire:d generously, hence the large amount of burnt cars and the typical markings on asphalt roads and so on that are clearly visible in the early photos.
This and the use of tank artillery against inhabitants of the kibbutzim has caused several scandals in israeli politics and the opposition has been requesting thorough investigation for a long time by now. The IDF calls this policy of killing your own soldiers and civilians the Hannibal directive.
I've been following this genocidal colony for decades, every time they've been "mowing the lawn" as they call it there is a massive amount of imagery of murdered kids coming out of the Gaza strip. The reason you think I'm lying is that you haven't been paying attention, and this is probably also why you react so strongly to a single recording of palestinians parading Shani Louk. It might also just be that you're racist and deem israelis or zionists generally more human than the people they are exterminating.
This has been national news in Belgium over the past few days.
It seems the company operating check-in services for many airports, Collins Aerospace (no news on their website so far - https://www.collinsaerospace.com/news?#newssearchresults_but...) was hit by a ransomware attack on Friday night, disrupting operations at Brussels Airport, London Heathrow, Berlin and Dublin airports.
Media are reporting that apparently Loki Locker ransomware was used. Check in desks are reverting to pen and paper, and many flights have been cancelled. More disruptions and cancellations are expected tomorrow (Monday September 22nd).
When accessing from Belgium the link is blocked by Cloudflare:
Error HTTP 451
Unavailable For Legal Reasons
In response to a legal order, Cloudflare has taken steps to limit access to this website through Cloudflare's pass-through security and CDN services within Belgium
CF is in a position such that if they aren't cooperating with national laws, then they are actively hindering them. National governments don't like that, and will have ISPs block CF wholesale if that's what accomplishes their goals.
To operate in Belgium, they have to follow local laws and comply with legal orders. They either make the site unavailable to local IPs or leave that market.
I'm unable to resolve the domain on EE UK - looks like it's DNS blocked.
By comparison, on my work network (TalkTalk) I can resolve the domain but I get a connection reset from the site.
I think this might be the first time I've hit a DNS block. It feels rather eerie seeing people talking about a site that, from my point of view, doesn't even exist...
There's an inconsistent censoring of numerous websites across the UK. In short, the biggest ISPs (a list which changes over time), will block various sites (TPB, libgen, AA, and others), based on court orders taken out at different timesIn general, it's a good idea to use Private Relay if you're using Apple devices and have access to it, no matter what network you're on, and if you're doing anything you don't want your ISP to traffic capture you should be using VPNs and/or Tor.
There are a lot of legitimate reasons to want to use scraping sites that UK copyright law is not nuanced enough to protect, and so blanket bans just end up emerging at the demands of copyright owners (which more often than not, means Disney or Springer).
Yes, Ofcom really needs to sort this out properly. I shouldn't be able to access this site from a UK ISP. Makes no sense that it's blocked on some and not others.
Idk, I went there a couple of times, I just love the people, the country. It’s a trip back in time. So it was my “random pick” for an exit node. And now I can read rt.com, sail the high seas, open any libgen or Anna's Archive. They're not part of the EU, seem far away from it (no euro, guarded borders, ditched their communist dictator who completely isolated the country ~40 years ago). Perhaps they are less easily coerced into censoring as practiced by countries primarily governed based on GDP and what the big corps want (although everybody seems to smoke everywhere so they could use some of that EU influence).
There's "431 Request Header Fields Too Large" which you will see occasionally. But after that 451 is the only other 400-level error code above 429. It was chosen as a reference to the book Fahrenheit 451.
Setting it up was easy enough, but just as I was about to start linking it to some test accounts, I noticed I already had blown through about $5 of Claude tokens in half an hour, and deleted the VPS immediately.
Then today I saw this follow up: https://mastodon.macstories.net/@viticci/115968901926545907 - the author blew through $560 of tokens in a weekend of playing with it.
If you want to run this full time to organise your mailbox and your agenda, it's probably cheaper to hire a real human personal assistant.
reply