Context (and a little more about me): I've been using Claude to vet and analyse the "shiny objects" that appear on my socials. If your algorithm is working as hard as mine, the deeper you go into the rabbit hole of Claude.ai, Claude Code, OpenClaw etc. and all it's wonderful capabilities the more you will naturally see all the cool tools and skills people post about on X, Instagram etc. (The main way I came across Jakes content!) Well... as someone with ADHD, I sometimes (most of the time lol) fail to let go of things and that question of "Is this something that is useful to me now or could be useful for my later goals and what I plan to build/create/do" constantly rings in my head. To combat the possibility of focusing on glittery sh*t, I've been sending reels, X posts and the like through transcript converters and then asking Claude to vet it thoroughly and confirm if it's something that is of use now (learn/study/having fun phase) or something that could help me in the future (end game/gigachad final boss/bigstonks) achieve what I'd like it to achieve. Claude has been great at doing that, and has actually pushed back and has been super helpful in outlining VERY clearly if I am simply adding jargon to the mix and saving the really useful ones that do fit in context. The problem: The issue is that on Free/Pro/Max (not sure about other plans) you're limited to 30 memories per account. When you hit this limit, Claude can no longer save anything new. Naturally, being the lazy, rather cheap and very broke person I am (I make bad decisions sometimes), I started wondering if there was a workaround that could include a file system. After about an hour of brainstorming, I managed to conclude a workaround that offers little friction. Here's what I found. You can get Claude to list all existing memories and output each one as text. From there, copy each one into a separate note or doc, 1 per memory and save them into a "Memory" folder in your Google Drive. This part is manual but takes minutes, not hours. Once saved, you can then head to Settings > Memory in Claude.ai and delete all existing memories (scary). Once that part is confirmed, you can create a new chat and ask it to save the first memory with something like "Every time I open a new chat, can you please remind me to connect you to my memory bank in Google Drive".

Hey everyone, Curious what the community is running in production or for daily development. Are you mainly using OpenAI, Anthropic, open-source models, or something else? And more importantly why did you choose it? (performance, cost, context window, tooling, etc.) Also curious about your workflow: what IDE are you using with it? VS Code, Cursor, Antigravity, or something else? Would love to hear: • Your main LLM provider • Your IDE / dev setup • One reason you stick with it Always interesting to see what people actually use vs what’s trending.

Hey guys, I feel like you can drastically improve your agent by adding skills, but I find it hard to choose the right ones. I'd also like to understand their impact on token usage better: If a skill.md file has ~4k tokens, will the whole 4k be added to my context as soon as that skill is invoked? Can anyone provide clear numbers as to how much their token usage has increased by using skills? But first and foremost I'm interested in your recommendations for specific skills for certain problems. Which do you like and why? Do you have a great way of finding new ones, when you're coming across new problems?

No App, No typing, No training needed... Just speak --> get a professional PDF Invoice in second. Here the workflow : 1. Contractor sends a voice message on Telegram 2. Whisper AI transcribes it 3. GPT-4 extracts items, quantities, prices 4. A professional PDF invoice or Angebot is generated automatically 5. Bot sends the PDF back instantly on Telegram Full stack running in production: → n8n on Hetzner VPS 🇩🇪 → HTTPS via domain kinvoice → Telegram Bot as the UI → html2pdf for document generation → Multi-document support (Rechnung + Angebot) The contractor doesn't need to learn any software. They just talk. The bot does the rest. This is what I'm building for German SMEs — simple AI automation that saves hours every week. Onboarding first customers now.

Hey guys, has anyone worked with both Cursor and Claude Code and would tell us about their experience? I've been using Cursor for.. maybe a year now, and I'm quite happy, but recently I've been hearing almost exclusively about Claude Code, and I'm wondering about the differences and if I'm limiting myself by using Cursor.