Website Brain Vault Build Method

Last updated: 6 June 2026

Transform any website into a fully interlinked, AI-ready Obsidian vault containing all text, images, screenshots, design DNA, and internal link structure — buildable in under an hour — so any future AI agent can edit, audit, or extend the site with full context.

// TL;DR

The Website Brain Vault Build Method transforms any website into a fully interlinked Obsidian vault containing all text, images, screenshots, Design DNA, and internal link structure — typically in under an hour using Claude Code, Firecrawl, and multi-agent parallel execution. Use it at the start of any AI-driven website project — SEO audits, redesigns, content creation, competitor analysis, or legal reviews — so every downstream AI agent has deep, persistent, structured context about the entire site before performing a single task.

Framework

// When should you use the Website Brain Vault Build Method?

Use this skill whenever you need to give an AI agent deep, persistent context about a website — your own, a client's, or a competitor's — before performing tasks like content creation, SEO audits, redesign work, legal reviews, or social media generation. Trigger it at the start of any website-related AI project, before you run any other skill like claude-seo or claude-blog.

// What inputs do you need to build a Website Brain vault?

target_website_urlrequired
The full URL of the website to be scraped and vaulted.
firecrawl_api_keyrequired
API key from firecrawl.dev (free tier gives 1,000 credits). Stored in a .env file inside the project folder.
vault_folder_namerequired
A descriptive name for the new Obsidian vault folder, e.g. 'MinneapolisMade-WebsiteBrain'.
obsidian_skillrequired
The Claude Obsidian skill loaded into the Claude session — handles vault file creation and markdown formatting.
brainstein_skillrequired
The Brainstein skill loaded into the Claude session — handles domain-expert multi-agent orchestration and interlinking.
ide_pathrequired
Path to the vault folder opened inside your IDE (VS Code or equivalent) for Claude Code session.
existing_marketing_brain
Optional path to an existing marketing brain vault if you want the website brain linked into it.

// What are the core principles behind building a Website Brain?

The Brain

A Brain (or Vault) is not a bookmark dump or a folder of files — it is a fully interlinked, AI-readable knowledge structure in Obsidian markdown format. The quality of a Brain is measured by its interlinking: a well-linked Brain operates 'on all cylinders'; a poorly linked Brain is an ADHD Brain.

Design DNA

Every website has a Design DNA: its color palette, typography, logo set, component library, CTAs, and hero imagery. The Website Brain must capture Design DNA explicitly so that future AI agents generating images, ads, or new pages always match the site's visual language without being told.

Multi-Agent Wizard / Parallel Execution

Never run the scrape and vault-build as a single sequential task. Instruct the orchestrating agent to spawn relevant sub-agents and execute in parallel using the multi-agent wizard approach — this is what compresses a potentially hours-long job into 45 minutes and ensures all page types are handled simultaneously.

Plan Mode First

Before writing a single file, run the orchestrating model in Plan Mode (using the Max model tier, not ultracode) to let it ask clarifying questions, confirm the scraping engine, and define the output structure. Only after the plan is locked do you escalate to ultracode / workflow execution mode.

Secure API Key Practice

Never paste API keys directly into prompts. Store all keys in a .env file inside the project folder and share only the file path with the agent. This is the non-destructive, reusable, and secure way to connect tools like Firecrawl, DataForSEO, and others to your AI agents.

Brains Compound

A Website Brain is not a one-time artifact — it is a living foundation. Every subsequent skill run (claude-seo, claude-blog, social media brain, image generation) should reference and update the Brain. The bigger the Brain, the more relevant every future AI output becomes.

// How do you build a Website Brain step by step?

1
Create the vault folder
Create a new, empty local folder with a descriptive name (e.g. 'ClientName-WebsiteBrain'). This becomes the Obsidian vault root AND the Claude Code working directory. Open it in your IDE.
2
Create a .env file with your Firecrawl API key
Inside the vault folder, create a file named '.env'. Add the line: FIRECRAWL_API_KEY=your_key_here. Get a free key at firecrawl.dev (1,000 free credits). Never paste the key into the chat — always reference the .env file path. Future keys (DataForSEO, etc.) can be added to the same file.
3
Open Obsidian on the vault folder
Point Obsidian to the same folder. It will start empty. Keep Obsidian open throughout — you will watch the Brain grow in real time as the agent writes markdown files.
4
Load the Brainstein skill and Claude Obsidian skill into your Claude session
Both skills must be active in the session. Brainstein handles multi-agent orchestration and domain-expert knowledge fetching. Claude Obsidian handles vault file creation, markdown formatting, and interlinking syntax.
5
Switch Claude to Plan Mode (Max model tier) and submit the mission prompt
Use Plan Mode — not ultracode — for this step. Max model tier only. Your prompt should include: (a) the target website URL, (b) the mission to scrape all text, images, video embeddings, animated SVGs, GIFs, full-page screenshots, and Design DNA, (c) instruction to use Firecrawl as the scraping engine, (d) instruction to use best practices with multi-agent parallel execution via the multi-agent wizard, (e) the .env file path. Let Claude ask its clarifying questions and answer them before proceeding.
6
Confirm scraping engine as Firecrawl and escalate to ultracode / workflow execution mode
When Claude asks which engine to use, confirm Firecrawl. When asked about output format, confirm: self-contained template scripts. Then switch from Plan Mode to workflow/ultracode mode. Claude will now spawn parallel sub-agents.
7
Monitor the vault as it populates — do not interrupt
Watch Obsidian fill with markdown files. Expected outputs: README, CODEX, per-page markdown notes (with full text, H2s, H3s, FAQs, author info, internal/external links), all PNGs/SVGs/GIFs/logos extracted, full-page screenshots embedded, video embeds captured, sitemap note, Design DNA note (colors, typography, logo set). Full scrape of a 40-60 page site typically takes 20-45 minutes. A 1,000-page site will take longer — plan accordingly.
8
Review the Obsidian Graph View to verify interlinking quality
Open Graph View in Obsidian. A healthy Website Brain shows dense, beautiful interlinking between page nodes. If the graph looks like isolated dots (ADHD Brain), prompt Claude to audit and repair internal link references. A well-linked Brain has every major page connected to relevant related pages.
9
Run a quality check prompt
After the agent signals completion, prompt: 'Do a full review and confirm all pages are scraped, all images are saved, all internal links are resolved, word counts are present, and Design DNA is documented. Report any gaps.' Address any gaps before declaring the Brain complete.
10
Optionally link the Website Brain to your Marketing Brain or other existing vaults
If you have a Marketing Brain or other Brains already built, decide whether to combine them or keep the Website Brain standalone. Combining is recommended for clients or for your own business — it enables future agents to cross-reference SEO data, marketing strategy, and website content in one session.
11
Begin layering additional skills on top of the completed Brain
The Website Brain is a foundation, not an endpoint. Recommended next layers: (1) run claude-seo for on-page audits, (2) run claude-blog for content gap analysis, (3) use Codex with the vault open to generate on-brand LinkedIn posts, images, or GIFs in parallel, (4) schedule automated page update tasks referencing the Brain. The Brain's value compounds with every additional skill run.

// What are real-world examples of using the Website Brain method?

A 40-page local service business website with a blog, landing pages, and a portfolio section needs a full AI-ready knowledge base so a Claude agent can generate on-brand blog posts and social media content without manual briefing each time.

Build the Website Brain vault for the site using Firecrawl + Brainstein + Claude Obsidian. The resulting vault contains every page as a markdown note, all portfolio images, blog cover images, the Design DNA (brand colors, fonts, CTAs), and a full internal link graph. When the Claude agent is later asked to write a new blog post, it opens the vault, finds related pages, matches the brand voice from existing content, mirrors the CTA style from the Design DNA note, and auto-inserts relevant internal links — all without a manual brief.

A web design agency wants to propose a redesign to a client but needs to audit the existing site's design consistency before presenting recommendations.

Build the Website Brain for the client site. Open Graph View to inspect interlinking. Review the Design DNA note to identify inconsistent fonts, colors, or component patterns across pages. Use the full-page screenshots embedded in each page note to visually spot pages that deviate from the design language. Feed the vault to an image generation agent (Codex) to produce concept art showing a unified redesign. Present the Brain's gap report and concept art to the client as the audit deliverable.

A content team wants to update 200 old blog posts on a large site to improve SEO without manually reviewing each one.

Build the Website Brain for the full site. Each blog post becomes a markdown note with its H2s, H3s, word count, internal links, and existing keywords captured. Prompt Claude: 'Starting from the oldest post to the most recent, create a daily scheduled task: run claude-seo review, then update the page content with current best-practice structure and high-search-volume keywords, then confirm internal links are present. Run one page per day.' The Brain ensures Claude never loses context or repeats work already done.

// What mistakes should you avoid when building a Website Brain?

Do NOT skip Plan Mode. Jumping straight to execution without letting the agent clarify scope and engine choice results in incomplete scrapes and poorly structured vaults.
Do NOT paste API keys directly into prompts or chat windows. Always use a .env file. This is a security risk and also means the key cannot be reused cleanly across sessions.
Do NOT use a weak model tier for planning. Use Max (or equivalent best available) for Plan Mode. Reserve ultracode only for the execution phase.
Do NOT treat the Website Brain as a one-time deliverable. A Brain that is never updated or extended is a dead Brain. Layer additional skills (claude-seo, claude-blog, image generation) on top of it regularly.
Do NOT ignore the Graph View after the build. An ADHD Brain (isolated, unlinked nodes) means the agent will lack context when navigating between pages. Audit and repair interlinking before using the Brain for downstream tasks.
Do NOT run automated bulk page updates without caution — Google can flag mass AI-generated content updates. Update pages thoughtfully, especially on large sites.
Do NOT create new pages on a large site before updating existing old ones. If you have hundreds of pages from 2016 that are not ranking, update those first. Google rewards updated, well-structured pages.
Firecrawl has a free tier limit of 1,000 credits. For very large sites, plan your credit usage or upgrade to a paid plan before starting the crawl.

// What key terms do you need to know for the Website Brain method?

Brain (Vault): An Obsidian markdown vault that serves as persistent, AI-readable knowledge infrastructure for a specific domain (a website, a marketing strategy, a social media system, etc.). Distinguished from a simple file folder by its dense internal interlinking, which allows AI agents to navigate context the way a human expert would.
Website Brain: A Brain built specifically from a scraped website. Contains every page as a markdown note, all images and SVGs, full-page screenshots, video embeds, Design DNA, sitemap, and a fully interlinked graph — everything a human or AI agent needs to understand, edit, or recreate the site.
Brainstein: A Claude skill that acts as a domain-expert orchestrator. It dispatches multi-agent parallel tasks, fetches best practices from the internet, and structures the resulting knowledge into a Brain vault. Used in combination with Claude Obsidian skill.
Claude Obsidian Skill: A Claude skill that handles all Obsidian vault operations: creating markdown files, formatting content correctly for Obsidian, writing internal [[wikilinks]], and embedding images and screenshots into notes.
Multi-Agent Wizard: The orchestration pattern used during Brain builds where the primary agent spawns multiple parallel sub-agents, each handling a different page, image category, or task simultaneously — dramatically reducing total build time.
Design DNA: The captured visual identity of a website: color palette, typography system, logo set, component library (buttons, CTAs, cards, heroes), and brand imagery. Stored as a dedicated note in the Website Brain so future AI-generated content always matches the site's aesthetic.
Plan Mode: A Claude operating mode used before execution, in which the agent asks clarifying questions, defines scope, and proposes a plan before writing any files or running any tools. Always use Plan Mode with the Max model tier at the start of a Brain build.
ADHD Brain: A poorly built Brain vault where files exist but are not interlinked — nodes are isolated and sporadic in the Graph View. Contrasted with a healthy Brain where all pages are densely connected, enabling AI agents to navigate context fluidly.
Firecrawl: The recommended web scraping tool (firecrawl.dev) for Website Brain builds. Provides markdown output, HTML, full-page screenshots, brand colors, OG images, and component data via API. Free tier includes 1,000 credits.
.env File: A local configuration file (named '.env') stored inside the vault/project folder that holds API keys in the format KEY_NAME=value. The secure, non-destructive, and reusable method for sharing credentials with AI agents without exposing them in chat.
claude-seo: A Claude skill for on-page SEO review and optimization. Intended to be layered on top of a completed Website Brain as a downstream task.
claude-blog: A Claude skill for blog content creation and structuring. Uses the Website Brain as its context source to generate on-brand, internally linked blog posts.

// FREQUENTLY ASKED QUESTIONS

What is the Website Brain Vault Build Method?

The Website Brain Vault Build Method is a framework for scraping an entire website and converting it into a fully interlinked Obsidian markdown vault — called a Website Brain — containing every page's text, images, screenshots, Design DNA (colors, fonts, logos), and internal link structure. It uses Firecrawl for scraping, Claude Code for orchestration, and multi-agent parallel execution to complete the build in under an hour for most sites.

What is a Website Brain in Obsidian?

A Website Brain is an Obsidian vault built from a scraped website where every page becomes an interlinked markdown note containing full text, headings, images, screenshots, and metadata. It includes a Design DNA note capturing the site's visual identity and a sitemap note mapping the full structure. The dense internal wikilinks allow AI agents to navigate and understand the entire site contextually, unlike a simple folder of downloaded files.

How do I build a Website Brain vault with Claude Code?

Create a new folder, add a .env file with your Firecrawl API key, open the folder in both Obsidian and your IDE, then load the Brainstein and Claude Obsidian skills into Claude. Start in Plan Mode with the Max model tier, submit your mission prompt with the target URL, confirm Firecrawl as the scraping engine, then escalate to ultracode execution mode. Claude spawns parallel sub-agents that populate your vault with markdown notes, images, and Design DNA in 20-45 minutes.

How do I use Firecrawl to scrape a website into Obsidian?

Sign up at firecrawl.dev for a free API key (1,000 credits), store it in a .env file inside your vault folder, and let Claude Code reference that file path during the build. Firecrawl returns markdown output, full-page screenshots, brand colors, OG images, and component data via API. Claude's orchestration agents call Firecrawl in parallel across all pages, converting each response into properly formatted Obsidian markdown notes with embedded images and wikilinks.

How does the Website Brain method compare to just saving a website as PDFs or bookmarks?

PDFs and bookmarks are static, isolated snapshots with no internal linking or structured metadata — AI agents cannot navigate between pages or understand site-wide patterns from them. A Website Brain is a fully interlinked knowledge graph where every page connects to related pages via wikilinks, Design DNA is explicitly captured, and all content is in AI-readable markdown. This lets agents cross-reference content, match brand voice, and insert internal links automatically.

When should I build a Website Brain before starting an AI project?

Build a Website Brain at the very start of any website-related AI project — before running SEO audits, content creation, redesign proposals, competitor analysis, or social media generation. It gives every downstream AI agent persistent, structured context about the entire site. Specifically, trigger it before running skills like claude-seo or claude-blog, and whenever you onboard a new client site or need to audit your own.

What results can I expect after building a Website Brain vault?

You get an Obsidian vault with a markdown note for every page (including full text, headings, word counts, and links), all extracted images, full-page screenshots, a Design DNA note with colors and typography, and a dense interlinking graph visible in Obsidian's Graph View. Downstream, AI agents produce higher-quality outputs because they reference real site content, match brand voice automatically, and insert accurate internal links without manual briefing.

What is Design DNA in the Website Brain method?

Design DNA is a dedicated note in your Website Brain that captures a site's complete visual identity: color palette, typography system, logo set, component library (buttons, CTAs, cards, hero sections), and brand imagery patterns. It ensures any future AI agent generating images, ads, blog graphics, or new page designs automatically matches the site's aesthetic without being manually briefed on brand guidelines each time.

How long does it take to build a Website Brain for a 40-page site?

A typical 40-60 page website takes 20-45 minutes to fully scrape and vault using multi-agent parallel execution. The time compression comes from Brainstein spawning multiple sub-agents that handle different pages, image categories, and tasks simultaneously rather than sequentially. Sites with 1,000+ pages take significantly longer — plan credit usage accordingly since Firecrawl's free tier is capped at 1,000 credits.

What is an ADHD Brain and how do I fix it?

An ADHD Brain is a poorly built vault where markdown files exist but lack internal wikilinks — in Obsidian's Graph View, nodes appear as isolated dots instead of a dense, interconnected web. Fix it by prompting Claude to audit all page notes and repair internal link references, ensuring every major page connects to relevant related pages. A healthy Brain shows dense interlinking in Graph View, enabling AI agents to navigate context fluidly across the entire site.

// GET THIS SKILL — FREE