how are you structuring your AI coding setup? (sharing mine, want to compare notes) been deep in the weeds optimizing my coding stack lately and want to see how you guys are actually running yours. here's where I'm at: what I pay for - Ollama - ChatGPT / Codex - Claude Code ($100 plan) how I split it - manual coding sessions → Claude Code - my orchestration layer (I call it Hermes) → runs on Codex with an Ollama fallback the issue: I cap out on the $100 Claude plan fast. so lately I've been running fcc-claude (free-claude-code) — it keeps the Claude Code terminal workflow but lets you point it at whatever API you want. feels like working in Opus, but the model underneath is swappable. I rotate the backend between: - DeepSeek - MiniMax M3 - and lately Qwen3:30b fully local right now I'm routing local Qwen through Ollama into fcc-claude, so I'm basically running the Claude Code experience on my own hardware for free. for the workflow layer I lean on Superpowers + a few MCP connections, and that's pretty much it. so what I'm trying to figure out: - how are you structuring sessions — one main agent, or orchestration with subagents? - what's your model routing look like? mixing local + API like this, or all-in on one provider? - what are you running for the "framework" layer — superpowers, custom skills, raw prompts? - and real talk: what am I missing? where am I leaving gains on the table? drop your setup below, even a rough sketch. trying to make sure I'm actually maxing this out and not just stacking tools for the sake of it.

Has anyone built something like this already? The idea is to have two windows side by side: one with the original document and one anonymized version that gets sent to an AI. After the AI completes its work, the changes are then mapped back to the original document. I only found a German video explaining a similar concept:https://www.youtube.com/watch?v=MP1ycNNZChw Would be interested to hear if anyone has implemented this or knows existing tools.

Hi everyone! First of all, sorry my english level, i'm spanish and a lot of times i make mistakes. I just want to ask you how to make AI Agents that doesnt look idiots, i try to give them less control and a good finetunning but it seems like it doesnt perform better thana single ChatGPT in robust tasks like data analysis in depth, so i don't know how to improve them, do you have some tips? Thanks in advance to everyone!

Hello everybody is there somebody already has experience with scraping your own email boxes and then those persons automatically invite on LinkedIn to be connected? Has anybody consistently already has a workflow for that developed? Thank you very much for your help and support, it's for a colleague of mine. 🙏

Hi everyone! Does anyone have a working JSON example for the Google Gemini / Veo 3 API that successfully uses an input image in an image-to-video request? I’m specifically looking for the correct JSON structure required by the **official Veo 3 API**, not any third-party tools, wrappers, or integrations. I’ve tried multiple formats and examples I found online, but none of them actually used the image I sent in the request — the generated video always ignored my input image. If you have a working request example or an official reference, I’d really appreciate it!