Hey everyone, I'm currently paying €100/month for Claude Max and I use Claude Code as my main harness. I've built a lot of important skills and workflows on top of it (copywriting agents, automation pipelines, etc.) and I don't want to lose that. But I'm wondering: is there a way to use Mixture of Agents (MoA) with Claude Code + OpenRouter? The idea: - Instead of paying flat €100/month for Claude Max, I want to pay per-use via OpenRouter credits - Use cheaper models (Qwen, DeepSeek, etc.) for simple tasks - Reserve Claude Sonnet/Opus only for complex decisions - Keep Claude Code as my harness since all my skills run on it What I need to know: 1. Can Claude Code work with OpenRouter as a backend (not just Anthropic's API)? 2. Is there a way to route different tasks to different models (MoA-style) within the same Claude Code session? 3. Has anyone set up LiteLLM or similar proxy to make Claude Code talk to OpenRouter? 4. What's the actual cost difference? If I mix Qwen 14B (cheap) + Sonnet (expensive only when needed), can I get below €30-40/month? My current setup: - Claude Max €100/mo - Multiple custom skills (copywriting, content automation, etc.) - I use Opus 4.8 for everything right now - Clients: 2 (fitness copywriting, Instagram content) I saw what Nous Research is doing with their Portal (mixing GLM 5.2 + DeepSeek + Qwen) and it looks like exactly what I need — but I want to keep Claude Code as my harness. Any help appreciated. Thanks!

Hey guys I've been going back and forth on this for weeks and I'm stuck. I need to track Instagram data for my clients (and their competitors): views, comments, captions, and ideally reel transcripts. No Meta Graph API — I don't want to mess with the media buyer's setup and honestly it seems like a headache I don't need. Right now I'm not tracking any of this and it's killing me because I can't improve what I don't measure. I know Claude can use browser automation to click around and scrape stuff, but it's clunky and slow for this volume. I want something that just runs and feeds into a dashboard — even if it's niche-specific or just my own clients. Anyone in the same boat? What tools are you using that don't cost an arm and a leg? Would love real recommendations, not just "use the Meta API" 😅

I know claude was known for being pretty natural and I can confirm it from my experience. But I wonder if GLM or some other Fable 5 Competitors could be better at that. Also how does a model become good at copy? Is it the most trained on more quality data? Explain this to me cause I don't get it as much

The biggest mindset switch we had building this campaign wasn't about sending MORE, it was about sending BETTER. Volume done wrong is exactly what affects, and we learned that the hard way. 💧 Send like a human: Sending 500 at once from a new account is the best way to get into spam quickly. So we made the system mimic a person, warm-up (raises volume gradually), time window (office days and hours only), and one every so often, not all at once. We let it flow drop by drop, like water shaping a rock. 🛑 Automatic brake: If bounces cross a certain threshold, the campaign pauses itself and alerts us. Something funny happened here, at first it counted auto-replies like "back on Monday" as bounces and stopped for no reason. We tuned it to only count real bounces. We learned that an alarm that goes off for everything is an alarm you end up ignoring. 🔁 Follow-up depending on behavior. This part made us happy, generic follow-up, the same message for everyone, never quite convinced us, so we made the system read what each person did and react, across 4 branches: if they viewed the content, it invites them to a meeting; if they viewed it but don't reply, a close that acknowledges they've already seen it; if they opened nothing, a soft message that reopens the door; and if it's already been several touches, an "elegant close" that lets them go without pushing. One detail that cost us: the spacing. Before, someone could get two follow-ups the same day. We fixed it so each one counts from the previous one, not from the first touch. And new emails and follow-ups have separate limits, so they never compete with each other. We think looking human isn't a trick, it's respecting how email works. And follow-up isn't repeating, it's reacting to what the person did. 👉 How many follow-ups do you send before letting a contact go? We want to know how others do it, we send two, and you?!

I have, over the past 3 years, learned where to look for the best and latest ai news/tools/etc. I also have narrowed down who to learn from, for example @Nate Herk is awesome. I just built this news page like drudgereport so everyone else can learn with me :) https://aititus.com/news/