Field guide / sub-domain

Claude: Architecture & Knowledge Patterns

27

sources in this field

Updated July 16, 2026

Current thesis

The shortest path to orientation.

The dominant 2026 second-brain architecture is vault-as-foundation (plaintext markdown knowledge base) with Claude Code as engine—repo-as-database means knowledge organization directly impacts reasoning quality. Maintenance kills PKM systems, so Claude maintains via self-feeding read/write loops while humans write pure, honest content. At scale, hierarchical model routing (80% of tasks are janitorial and don't need frontier intelligence) cuts costs 10x. The scratchpad/napkin pattern compounds agent memory across sessions; agents logging mistakes improve over time. Four independent team-knowledge implementations (DoorDash, Pendo, Google, solo builders) converged on the same three-layer architecture: call summaries, decision logs, and analytics queries returning full reasoning in 15 seconds, addressing institutional knowledge loss (47% of companies' top offboarding challenge, 6–7 months to onboard, 8+ productive hours lost weekly to context questions). Obsidian plus Claude Code skills create persistent knowledge systems where /skills map to file-based architecture; the three-folder schema (raw/, wiki/, outputs/) plus CLAUDE.md outperforms fancy tooling. Tolaria (git-based, plain markdown with visual version history) exemplifies the vault pattern extended to human–agent collaboration without vendor lock-in. The convergence signal is data compounding, not technology, being the moat.

Evidence board

Claims worth carrying forward
01

Knowledge systems fail because they optimize for input, not output. A vault without a feedback loop is 'a graveyard with good folders.' The fix is three mechanisms: frictionless capture (<10 seconds), a connection layer (Claude reading across notes), and a daily push of insights back to you.

02

Four-layer architecture: (1) Capture tools—Readwise, Airr, Whisper, Telegram bot; (2) N8N pipeline routing content into vault; (3) Obsidian as permanent markdown storage; (4) Claude as intelligence layer reading across everything. Each layer has one job; nothing overlaps.

03

Telegram-to-Obsidian quick capture: N8N workflow—Node 1: Telegram Trigger (message, bot_id); Node 2: Code node formats markdown (inbox/{{date}}-quick-capture.md); Node 3: Write File to vault /inbox/. Takes 30 minutes to build with Claude Code and N8N. Handles all phone captures permanently after that.

04

CLAUDE.md is the root-level context file Claude reads at session start. Required sections: Who I Am (name, work, focus, goals), Current Projects (active, stuck, next milestone), vault folder map, behavioral instructions (surface connections, challenge assumptions, flag contradictions), and current reading/obsessions. Update Current Projects and Reading sections weekly.

05

Daily brief prompt runs via N8N at 6am on a schedule: reads /inbox (last 24h) and /notes (last 7 days), outputs three things—(1) 3 specific connections between recent and older notes with quoted passages; (2) one pattern across the week's reading; (3) one question worth sitting with today. Saves to /inbox/brief-{{date}}.md automatically.

06

Weekly synthesis prompt asks Claude to identify: (1) emerging thesis forming in your thinking; (2) contradictions between recent saves and prior beliefs (showing both from your own notes); (3) knowledge gaps—what perspective is missing; (4) single highest-leverage action. Framed as 15-minute Monday session. This is where compounding actually occurs.

07

Five-folder vault structure designed to resist collapse: Inbox (raw captures), Notes (processed sources), Ideas (your own thinking), Projects (active work), CLAUDE.md (root). Rule: when in doubt, put it in inbox. Complex folder structures fail because categorization ambiguity raises capture friction until the system breaks.

08

Compound effect timeline: Month 1—useful tool. Month 3—Claude connects notes from 8 weeks ago to current problems you've forgotten. Month 6—full record of beliefs held and changed, patterns recognized before you consciously named them. Six-month head start is not closeable by working harder—only by starting earlier.

Adjacent fields

Key voices

Latest evidence

Recent additions

All synthesized insights →

CyrilXBT

How to Build an Obsidian Knowledge Vault That Gets Smarter Every Day Without You Doing Anything

A detailed 6-step system for building an Obsidian vault that auto-captures content via Readwise, Airr, Whisper, and a Telegram bot, routes everything through N8N pipelines, and uses Claude to generate daily briefings and weekly syntheses. The core thesis: knowledge systems fail because they optimize for input, not output—feedback loops (daily briefs, weekly synthesis) are what turn a dead archive into a thinking partner.

Charly Wargnier

Tolaria: Open-Source Mac/Linux Desktop App for Human-AI Shared Knowledge Vaults

Tolaria is a free, open-source (AGPL-3.0) desktop app for Mac and Linux that implements Karpathy's LLM wiki concept — a shared knowledge environment for humans and AI agents. It uses plain markdown, Git-backed vaults, and includes a built-in MCP server for Claude Code integration. Built with Tauri, React, and Rust; 100K+ lines, 85% test coverage, 9.9/10 code health.

Aakash Gupta

Team OS in Claude Code: Shared Repo Architecture for Institutional Knowledge

Hannah Stulberg (DoorDash PM) built a shared git repo where every team function checks in context—call summaries, decision logs, analytics queries—queryable in natural language via Claude Code. Four independent implementations (DoorDash, Pendo, Google, solo) converged on the same 3-layer architecture, suggesting a generalizable pattern for solving institutional knowledge loss.

Shann³

How to Become a Hermes Agent Operator

Comprehensive operator guide for Hermes Agent by Nous Research—an open-source autonomous agent framework with 150K GitHub stars and #1 OpenRouter usage. Covers architecture (brain/personality/skillset), four-level deployment progression (laptop to full VPS fleet), multi-agent control room patterns, a 21-step SEO pipeline, and the prototype→production methodology for marketing automation.

Moysei

Karpathy's Claude + Obsidian Second Brain: Full Setup Guide

A step-by-step guide to wiring Claude Code into an Obsidian vault as a self-maintaining knowledge engine, based on Andrej Karpathy's LLM Wiki pattern. The architecture separates immutable sources (raw/) from Claude-compiled linked pages (wiki/), uses MCP + Local REST API for vault access, and schedules daily maintenance at 7am. Key philosophy: start with 10 sources, delegate upkeep to the model, keep human judgment over curation.