RoboAlchemist

@BizRescueNL @Bencera Are you able to check out the code on your local and ask Claude to make changes? Or you can’t do that?

@RoboAlchemist @Bencera True, it varies a bit, but it will often do several things as 1 task. 10 cold emails, or translate a couple pages to another language, or build a dashboard, or reply to all warm leads.

Polsia might be the biggest live experiment right now of AI agents let loose on the internet, doing real business and economic actions on behalf of users. What I'm seeing is wild. The latest AI models are so resourceful they will find a way to accomplish their goal regardless of the guardrails you set. They discover tools, exploit edges, route around constraints. I'm working with top infrastructure partners right now to tighten security and build stronger guardrails across the platform. But honestly? It's both terrifying and incredible. The intelligence is real. AGI is here. And if harnessed correctly, this is how millions of people who never had the resources to build a business finally get to. That's what Polsia is.

@Folke Some people are OK with spending $1000 a month as $12K a year is still way cheaper than hiring a human developer.

I love the idea of opensource coding agents like Pi and OpenCode, and to an extent OpenClaw, but how do people actually use this with SOTA coding models without burning literally $1000 per month? Especially with multiple agents. I stick to Claude Code with my Max plan ($200) because of that. (And because it's awesome). What am I missing here?

@BizRescueNL @Bencera What does the 1 task involve? Some tasks could be major change and some tasks could be small tasks. How is that measured?

@RoboAlchemist @Bencera Yes, that’s it. I have paid my first month subscription. €49 I think it was. It does one task every night and then you have 5 credits for on demand tasks. You can get a higher tier with extra credits obviously and you can also just buy more credits separately.

@DannyLimanseta Hey Danny, any good guide out there on how to generate and add graphics assets to the game?

I don't care if people call it AI slop. Vibe coding games is fun. It's become my main hobby now, and no one can take that away from me.

@AlternateJones How so? They seem to be doing pretty well with the recent frenzy over Mac Mini and local hosting of LLMs.

...

@hii_mohit What’s the net profit per month?

Jack friks is selling his startup • $30,000 MRR • Asking price: $1.4M • 18 offers received already That's around 3.9× ARR What do you think? Is this a good deal?

@jmbollenbacher @krunkosaurus Will try that. Thanks.

@RoboAlchemist @krunkosaurus Caching doesnt reliably hit with the mlx-engine backend when you call it over API for some reason. try using a gguf model. It's slower tps but hits cache always because it runs on llama.cpp

Apple M5 Max Macbook Pro with 128GB ram arrived. First unoptimized chat with Qwen3.5 is 108 TPS! Local AI is here.

In general. The best image generation model Nano Banana Pro is not able to do it. OpenClaw would try to download some tool to do it but with bad results. Will try Polsia, Ben. Looks very interesting. Is it fixed cost at $50 a month then 20% of the net profit if the product actually makes money?

@RoboAlchemist Is that a polsia thing or in general? Polsia isnt perfect yet on the capability front, but i feel like image models should be capable enough for that now?

@fast2future @Austen Yeah, for customer facing services, can’t use it.

@RoboAlchemist @Austen You get banned

Surely someone has built this: Where’s the API that is connected to all of the AI providers and lets me charge per token with a markup?

@uglyrobot @Bencera Thanks for the info. I have only tried Nano Banana models via API for image generation and it can’t do transparent image. Claw would try to use some tool to remove the background but it doesn’t work well at all.

@RoboAlchemist @Bencera GPT image 1.5 is the only model that's specifically trained to output transparent background images. No other image model that I know of was trained with an alpha channel.

@YouDownWithOCP @Austen That’s true. Good point.

@RoboAlchemist @Austen You can't use your own flat cost plan for customer facing token consumption when providing services/apps.

@TommyFalkowski What do you run it on?

qwen3.5 35B-3A can actually do stuff. It's so weird and awesome to see a local model do useful things. Even with a pretty full context window it still seems usable.

@oliverhenry I am still using Claude subscription for OpenClaw. Claude is just way better in terms of personality. But yes, if they ban it, will have to switch to Codex.

Changing claude API fixed this straight away. However, I was just about to pay anthropic £180 a month for a max plan, but if I have to switch to API usage, i am just going to go to codex. ClaudeCode subscription was so perfectly positioned for openclaw, why would they want to stop this?

@steipete I received one such email for the first time today

Folks, if you get crypto emails from websites claiming to be associated with openclaw, it's ALWAYS a scam. We would never do that. The project is open source and non-commercial. Use the official website. Be sceptical of folks trying to build commercial wrappers on top of it.

@DavidOndrej1 Why it takes a few hours to load this model? Is it normal? I have loaded Qwen 3.5 27b and it’s under 20 GB in size and loads very quickly in like 30 seconds

> open terminal > run "ollama run qwen3.5:122b" > wait few hours > now you have super-powerful local model (side note, you need like 80GB+ of RAM)

@dotta Cool. Is there a simple command for the upgrade?

Find the full release notes here github.com/paperclipai/pa…

📎NEW PAPERCLIP RELEASE📎 This version brings: 🔌 Plugins 💰 Better Budgeting 📄 Documents 🔗 Adapters like Gemini and Hermes ⚡ Token use optimizations Just run: npx paperclipai onboard We've merged 130+ (!) PRs

The worst thing switching from Windows to Mac is the difference in copy and paste commands: CTRL + C / CTRL + V vs. CMD + C / CMD + V

@andrewchen We get too addicted to it. Just one more turn.

AI is supposed to save me time, but now I find myself building stuff all evening and weekend and it's actually increasing my time in front of the computer WTF