Consider this a thank you @Jake Van Clief for the inspiration and giving us the mindset and tools to grow our businesses. ***Disclaimer: this is an advanced setup. DONOT use until you're fully comfortable with Jake's method*** =============================== Many of us eventually hit the limit with basic chat interfaces. The agent forgets context between sessions, data security starts to feel risky, and the setup creates more friction than value. This stack addresses those issues directly It combines Hermes Agent with Cognee as the memory engine, hosted on a simple DigitalOcean Droplet and secured through Cloudflare Tunnel. The entire deployment follows Jake Van Clief’s Interpretable Context Methodology for clean, repeatable orchestration. The result is a private, self-improving AI agent that grows more capable over time while keeping your data and server fully under your control. The components stay minimal and transparent: Hermes Agent as the autonomous gateway, Cognee for structured relational memory, and Cloudflare Tunnel for secure outbound-only access. You deploy once using the ICM workflow, then the system handles the repetitive memory management and self-improvement loops. The agent becomes a genuine thinking partner instead of a one-off responder. That frees up your attention for the judgment and creative work only you can do. If you are running a self-hosted agent setup or exploring similar private stacks, I would like to hear what you are using and what friction you have solved. Drop your thoughts below.

Quick version in case you missed it: instead of using RAG to re-chunk your sources every time you ask a question, you compile each source once into a persistent markdown wiki. The LLM extracts concepts, writes entity and concept pages, updates cross-references, flags contradictions, and maintains the whole thing. Future queries read the pre-synthesized wiki. The part that clicked for me: the reason most of us abandon our second brains is that backlink and cross-reference upkeep is boring. The LLM doesn't care. It's happy to touch fifteen pages in one pass. I spent a couple of weeks turning Karpathy's pattern into a Claude Code plugin that actually scales (atomic pages, sharded indexes, BM25 fallback past ~300 pages). It also runs in Codex, Cursor, Gemini CLI, Pi, and OpenClaw through the skills CLI. Install in Claude Code: /plugin marketplace add praneybehl/llm-wiki-plugin /plugin install llm-wiki@llm-wiki Or in any other supported agent: npx skills add praneybehl/llm-wiki-plugin -a <your-agent> Five slash commands (init, ingest, query, lint, stats), stdlib-only Python, no dependencies. Plays well with Obsidian if you want the graph view. Repo: https://github.com/praneybehl/llm-wiki-plugin Karpathy's gist: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f Curious if anyone here has tried the pattern themselves. What did you ingest first, and what broke before it worked?

Okay don't judge me on the title. It's actually what'll help you save up to 75% tokens while your using any Cloud LLM model. So as you can see in the screenshot below, there's this dev that made claude speak in "caveman" terms. Turns out that saves tons of credits for you. Tbh the more AI evolves, the more dumb and easier ways to save tokens arise, maybe soon enough the em dashes and 50 line answers would be eradicated by default? Anyways, here's the link to the post if you'd like to snoop around there: https://www.reddit.com/r/ClaudeAI/comments/1sble09/taught_claude_to_talk_like_a_caveman_to_use_75/

Hey community!! yesterday I Launched MinAITaur, my Human Tech consulting agency!! We dedicate to mitigate your bureaucracy and uplift your job!! I decided to start on a clean slate, everything from 0. I am so happy, everything has been possible thanks to Jake's teachings and ways. The Agency is a mix of organizational psychology consulting and IT automations using Claude! We help you redesign your Northstar giving your human capital the proper space and the repetitive tasks an architectural system which helps you mitigate it to stay more human than ever before!!! Here is my brands branding proposition. Everything is still in Launch state, happy to help anyone who resonates with this message and if you guys would like to follow I would be thankful to have you there as well. Socials: ig & fb: @minaitaur, www.minaitaur.io. (PD: webpage is not launched yet)

For those of you who've been following me on social media for a while, you've heard me say this over and over: building your own agent frameworks is a waste of time. Stop doing it. Stop paying people to do it. Stop learning how to do it. (Unless you like local projects, that's fun) Anthropic just published this today: https://www.anthropic.com/engineering/managed-agents Here's what it is in plain terms. Managed Agents is a hosted service where Anthropic runs the entire agent infrastructure for you. The loop that calls Claude, the sandbox where code runs, the context management, crash recovery, security, scaling. All of it. You define what the agent should do (system prompt, tools, connections to your systems) and they handle the rest. The key line from their own engineering team: "Harnesses encode assumptions about what Claude can't do on its own. Those assumptions need to be frequently questioned because they go stale as models improve." They gave a specific example. They built a workaround into their harness because Sonnet 4.5 would quit tasks early when it sensed its context limit approaching. When they ran the same harness on Opus 4.5, the problem was gone. The fix became dead weight. One model release made their own engineering work obsolete. Now think about what that means for every startup and every freelancer building custom agent harnesses and selling them to clients. Every assumption they baked into their code is a bet against the next model release. And the frontier labs are shipping new models faster than anyone can maintain a harness. This is the thesis. Your value lives above whatever just got commoditized. The infrastructure layer of agents just got commoditized. The thinking about what to build, why to build it, and how to structure the work around it did not. That's what we teach here. That's why we focus on the 60/30/10 framework, on understanding which layer a problem belongs on, on prompt architecture and workflow design.