How To Institute — Agentic AI Knowledge Base

The How To Institute is a free, open resource dedicated to the practical side of Agentic AI. No hype, no theory papers — just clear, recipe-style guides that walk you through building, deploying, and living with autonomous AI agents.

Whether you are setting up your first agent on a five-euro VPS or wiring one into your company’s workflow, the articles below will get you there step by step.

Articles

How to Self-Host a Company RAG with Open WebUI? A private, multi-knowledge-base RAG on your own VM — staff sign in with their Microsoft account, upload Office, PDF and Markdown, and query corporate knowledge bases alongside their own personal ones.
How to Self-Host RAGFlow for Deep-Document RAG? When the knowledge lives in gnarly PDFs, scanned contracts and dense spreadsheets — layout-aware parsing, Microsoft Entra login, and multiple per-team and personal knowledge bases on a VM you control.
How to Self-Host a Company RAG with LibreChat? The strongest Microsoft integration of the bunch — Entra login with group sync, a per-agent permission model, and Office/PDF/Markdown knowledge for shared corporate and personal assistants, all on a VM you control.
How to Claw? A recipe-style guide to getting your own AI agent up and running — from renting a VPS to chatting with it on Telegram.
How to Run Hermes in a macOS Sandbox? Run Hermes inside a self-contained Linux box on macOS Sonoma, with its own Python, its own Node, and exactly one folder shared with the host — OrbStack recommended, four alternatives documented.
How to Build a Mac-Within-a-Mac Toolbox? A persistent Linux box on your Mac for installing things you do not entirely trust — one shared folder, services that start at login via systemd and launchd, burn it down and rebuild in a minute.
How to Host Qwen 3.6 with vLLM on Two RTX 3090s? Self-host Alibaba’s brand-new Qwen3.6-27B or 35B-A3B MoE on a dual RTX 3090 desktop — AWQ 4-bit, FP8 KV cache, full 256k context.
How to Run Qwen 3.6 on an M5 Pro MacBook (64 GB)? Run Qwen3.6-27B or 35B-A3B locally on the new M5 Pro MacBook — MLX and llama.cpp side-by-side, MTP speculative decoding, quant accuracy table, OpenAI-compatible HTTP server.
How to Submit to IndexNow? Tell Bing, Yandex, and friends about your fresh content the moment you publish it — no more waiting days for crawlers.
How to Get 60,000 Requests per Second out of SQLite on a $5 VPS? Six PRAGMAs, two connection pools, and one rule about transactions — turn the world’s most-deployed database into a capable production backend with no managed database bill.
How to Build a Karpathy-Style LLM Wiki? Let Claude Code compile your reading list into a personal knowledge base that maintains itself — three folders, one CLAUDE.md, Obsidian on the side, and the two failure modes the quickstarts skip over.