I built Luna HQ in Claude Code. She lives on my Hostinger VPS, runs inside Telegram, and operates three of my businesses every single day. 7 command channels. 9 integrations. 2 AI brains. Persistent memory. 5 AM morning briefs. All mine. I started this build BEFORE Managed Agents dropped. By the time that option was available I was already too deep and too invested to stop. Honestly, glad I kept going. The level of customization I have on my own infrastructure is something a managed platform would have had a hard time giving me. Claude Code made this possible. CoWork helped me stay organized through the build. If you are sitting on an idea in this community — build it. Not someday. Now. Juan died so Luna could live. Honor his memory. Drop your questions below. Happy to share what the build looked like.

Well I finally got to install and setup my first OpenClaw instance today! Nothing fancy, just got it running in a local Docker container, was able to get the proxy setup and it connected to Telegram and responding back! Super excited to finally start learning how to use OpenClaw!!! More to come on this front, just only so many hours in the day!

Hey everyone, I’ve been working on something called "ASES", and I think it’s finally at a point where it needs real-world testing instead of just me building in a vacuum. This is not just another “AI tool” or prompt setup. It’s basically my attempt at: "Turning LLMs into a structured, end-to-end software development system" --- ## What ASES actually is ASES is a "schema-driven Scrum workflow for AI-assisted development". Instead of: * random prompts * messy context * inconsistent outputs It gives you: * a "full project lifecycle" * "structured artifacts (PRD, HLD, LLD, tasks, tests, etc.)" * and a system where models operate inside that structure --- ## What makes it different ### 1. Everything is structured (schemas + templates) Almost everything in ASES is backed by schemas: * PRD → requirements * HLD / LLD → architecture * Tasks → execution * Decisions → tracked explicitly * Test suites + reports → validation * Sprint summaries + audits → closure So instead of “ask the model and hope for the best” you get: "deterministic, repeatable outputs" --- ### 2. Full Scrum lifecycle (not just tasks) This isn’t just a task runner. It actually maps a full flow: * Planning → PRD / roadmap * Design → HLD / LLD * Execution → sprint-based tasks + snapshots * Testing → structured validation * Closure → audit + summaries Everything lives in a "project structure", not chat history. --- ### 3. Runtime context control (this is the core) Instead of dumping context into every prompt, ASES: * Injects context "just-in-time (per action)" * Uses "layered + scoped context" (global / sprint / execution) * Adjusts how much context to include based on state So the model only sees: > what it actually needs *right now* This is where a lot of the "token efficiency + consistency" comes from. --- ### 4. Model orchestration (multi-model workflows) ASES is designed for using multiple models with "clear roles", not one model doing everything. For example:

I want to do violence to Claude code right now. Driving me up the F'n wall. I can't wait for my call with charles.... tuesday can't come fast enough. I am spending nonstop time arguging with CC..... fighting with it. I have rules. it doesn't follow them. I ask why not? oh, sorry, I should have. you put a QC check in the steps..... it just passes stuff. I'm looking at the copy, I'm like WTF this is garbage. how can this pass? oh, I was just scoring it to pass the next step. It was outputting the stuff before so well....IDK WTF happened in the last 2 weeks or what I did.... it's gotten so shit. I feel like i'm going backwards right now honestly. INstead of outputting quality stuff it should be doing, I'm spending 90% of my day it feels like fixing stuff, back and forth, debugging.... arguing. Am I the only one here? WTF do I need to do to fix this shit?

Every time you open a new Claude Code session, it rereads all your files from scratch. Hundreds of tokens burned just to "remember" what it already knew. That's the problem nobody talks about. Graphify fixes that. It turns your entire codebase, your docs, your strategy notes, even your images, into one persistent knowledge graph. Claude doesn't have to go searching through files like it's 2019. It just knows. 70x fewer tokens per search. One-time build. Done. Here's what I'm breaking down: • What Graphify actually is and why it exists (shoutout to Andrej Karpathy for the inspiration on this one) • Full install and setup from zero • How to query the graph straight from Claude Code • How to visualize the whole thing in Obsidian • The real token savings, with receipts I appreciate you all being patient while this one came together. I didn't want to drop another surface level walkthrough that wastes your time. This one's the real deal. Try it out and drop a comment. I want to see how your graphs look.