Has anyone here tried to make everything run locally similarly to how they do with Claude? While I enjoy Claude having a way to not be reliant on it (like with Ollama and Qwen/Gemma) might not be a bad idea. If you have done this what does your workflow look like? What was your process?

There are two skills in Claude that I use over everything else that have been really successful for me. I love the folder system here and I use something similar just to keep things organized. As I'm sure most people have heard of Superpowers and I think it's good to a point. Superpowers is good at brainstorming and asking me the questions I didn't consider but it's bad on execution. That's where GSD or Get Shit Done comes in. It can take the plan from Superpowers, break it down into Phases and Waves and then you verify at every checkpoint to make sure a certain feature works. This effectively allows me to "one-shot" apps and make sure it works along the way. The part of the stack that I don't have nailed down yet (leaning into my cyber security background a bit) is security and just trying to make sure we cover stuff like prompt/code injection for example. Does anyone else have a similar stack or different process?

can we turn claude code into an agent similar to Hermes? has anyone tried? similar memory management? cron scheduling, and persistent long running agents?

Curious how you all gather all the sources of information coming out on new agentic frameworks, concepts, AI news, youtube and podcast videos, blog posts etc? Are you building a scraper that ingests all this into your own knowledge wiki? Are you using the ICM approach to synthesize the raw sources?

The thesis was simple. Treat AI like a DP, not a generator. Today I shipped that thesis as a feature. What I built Project Guide. A user-facing wizard that walks you through the same brief a director of photography would hand a director. Subject, framing, lighting, blocking, motion, transitions. No vibes. No "make it cinematic". Structured fields in, structured output back. How the pipeline works 1. Schema lives at the data layer 2. The wizard promotes every field into a question the user answers 3. Gemini structured output enforces the schema on the way back 4. The AI never gets to freelance the shape. Only the content. That last point is the whole game. The system around the AI is the intelligence. The model just fills in slots the system already decided matter. The session - Started at "let's brainstorm a feature" - Ended at merged-in-main with the wizard live - Under 8 hours of focused work - One Opus session in the advisor seat. Hands in the shadows did the file work. The thesis isn't a pitch slide anymore. It's in the schema, in the wizard, in the structured-output call. You can click it. Stay tuned. Results from it dropping later today. //A<3