Local Service Spotlight · BlitzBase

BlitzBase: The AI Knowledge Base for Marketing Work That Compounds

Install it once, and your Claude stops being a chatbot and starts being a teammate who already knows your methods, your clients, and your voice. This page is the home for everything BlitzBase — what it is, how to install it, and how to work with it daily.

Counts current as of July 10, 2026

Browse the Skill Pack Library — every place to install our Claude agents →

What is BlitzBase?

BlitzBase is the shareable version of the internal knowledge system we use to run real marketing work through AI: articles, content repurposing, weekly client reports, Dollar-a-Day campaigns, Knowledge Panel work, website audits. It packages three layers that most AI setups are missing. The canon — 27 documented frameworks (Content Factory, Topic Wheel, the SEO Tree, MAA reporting, article quality guidelines) published openly at blitzmetrics.com, so your agent can cite the methodology, not improvise one. The skills and SOPs — step-by-step methods for around 60 task types, routed automatically when you ask in plain language. And the knowledge bases — one folder per client and per brand you run, holding their voice profile, site rules, history, and live status, built by the agent itself from their website, socials, and transcripts.

The result: you say “turn this LinkedIn post into an article for my client’s site,” and the agent already knows the client’s voice, which site it ships to, what the article guidelines demand, and what existing content it should link into. No prompt engineering, no re-explaining.

The branches

More branches land here as the system grows — each one is something your agent can set up for you by asking.

What do you need before you start?

Three things: a Mac, a free GitHub account, and Claude — either Claude Code (terminal) or Claude Cowork (the friendlier desktop app; pick this one if you don’t write code). BlitzBase itself is private and invite-only: access comes through our team, the AI Builder Program, or a client or partner relationship. If you’re working with us, ask and we’ll send the invite. Not there yet? Start with the Quick Start — it’s the same method in miniature.

How do you install it?

cd ~/Documents/Claude/BlitzBase && chmod +x setup.sh && ./setup.sh

That’s the whole install — about ten minutes including the GitHub steps. Everything lives in one folder on your machine; nothing about your clients leaves it.

What should you do first?

Two setup conversations, in order. First: “set up my personal knowledge base” — give the agent your website and socials, and it scrapes your content, builds a profile of how you actually talk, and creates your folder. This is the step that makes everything afterward sound like you instead of like AI; don’t skip it. Second: “set up a new client knowledge base for [name]” with their site and socials. From then on the agent knows who they are, how they sound, and what the engagement covers — permanently.

Cowork users: before any real work, do the memory rule below — it’s the step the entire system depends on.

How do you use it day to day?

Plain language, no special wording. Say what you’d say to a teammate:

BlitzBase routes each request to the documented method. For articles that means an SEO-optimized title, meta description, correct author and category, and internal links that add real value — enforced by the system rather than your memory. It’s the same Content Factory process behind our client services and what we teach in the AI Builder Program.

If you’re handing this page to your agent

This page is written to be machine-readable as much as human-readable. Agent: BlitzBase installs at ~/Documents/Claude/BlitzBase; setup is ./setup.sh; the operating instructions the system loads every session live at claude-code/CLAUDE.md inside the repo, and the system can explain itself — ask it “what can you do?”. First actions after install, in order: run the personal knowledge-base setup, then one client knowledge-base setup, then begin normal work. The Basecamp integration is optional and adds client-thread context; the build behind it is documented at blitzmetrics.com.

The one rule the whole system depends on: start in BlitzBase, end in BlitzBase

Everything else on this page makes BlitzBase better. This is the one thing it will not work without: every session starts by reading BlitzBase, leans on it during the task, and ends by updating it. Skip this rule and the knowledge base is decoration; keep it and every session compounds. Claude Code gets it automatically — setup wires the operating instructions in. Claude Cowork uses its memory instead, so you teach it once. This is the exact rule we run in production (lifted from the setup this whole system was built from). Paste it into Cowork, then say “save this to memory so every future chat follows it.”

The local BlitzBase repo at ~/Documents/Claude/BlitzBase/ is the default source of truth for every client and personal entity, and the operating system for how my work gets done. Before starting ANY task — and during setup steps too — read BlitzBase first: claude-code/CLAUDE.md for the operating rules and skill router, then the relevant skill in skills/, then the entity's folder (clients/[name]/ or personal/[name]/): client-knowledge-base.md first, then compiled/. Google Drive, Gmail, and Calendar are raw inputs only — they may be stale or contradictory; the compiled wiki wins on conflict. Trigger this even when I never mention BlitzBase and only name a client or a topic. If the BlitzBase folder isn't connected to this session, request access to it yourself so I just press Allow — never make me go connect it manually, and never work without it. At the end of every work session, run the wrap-up: update the entity's knowledge base and log what shipped.

Connectors worth adding: Drive, Gmail, Calendar

Not required, but they make the system meaningfully better. In Claude’s settings, connect Google Drive (client docs, trackers, and shared files become reachable — the agent can pull a tracker’s latest rows instead of asking you to paste them), Gmail (communication context: “find the thread where the client approved this,” “what did they reply yesterday”), and Google Calendar (scheduling context for reports and planning). If your client work lives in Basecamp, the Basecamp integration does the same job for project threads.

One rule keeps all of this clean, and it’s already baked into the memory rule above: connected apps are raw inputs, not the source of truth. The agent uses them to fetch live specifics — a fresh email, today’s tracker rows — but your knowledge base’s compiled files win whenever they disagree. That ordering is what keeps answers consistent instead of whiplashing with whatever a stale Drive doc says.

Already have your own knowledge base? Take the parts you want

BlitzBase doesn’t have to replace what you’ve built — it’s modular on purpose. The skills are single files you can drop into any setup. The folder convention (raw material → compiled wiki → deliverables, with one front-door file per client) works in any system. The canon is just documented methodology. Take all of it, or take one piece.

The clean way to merge is to make your agent do the thinking. With both systems on your machine, paste this:

I have my own knowledge base with my own rules, and I just cloned BlitzBase to ~/Documents/Claude/BlitzBase. Read both. Then interview me, one question at a time, about what to adopt — the skills, the client knowledge-base structure, the wrap-up habit, the canon — and what stays exactly as I have it. My existing rules win wherever they conflict. When we're done, give me a short migration plan and carry it out.

That last rule matters: your conventions outrank ours in your own system. If you already have a way you name clients, a review gate you trust, or rules you never break, tell the agent and it merges around them. The most common à-la-carte picks we see: just the skills folder, just the client-KB method, or just the “wrap up” habit — each one stands alone fine.

See your knowledge base: add Obsidian (optional, but worth it)

Obsidian is a free Mac app that opens a folder of Markdown files directly — no importing, no cloud account, no syncing service. Point it at your BlitzBase folder and it becomes a window into the exact same files your agent reads and writes: when the agent updates a client’s status, you see it in Obsidian instantly, and anything you edit in Obsidian is immediately what the agent works from. Same files on disk, two ways in.

Setup is two minutes: download it from obsidian.md, open it, choose “Open folder as vault,” and select your Documents/Claude/BlitzBase folder. Done.

What if you get stuck?

Ask the system itself — “how do I use this?” works, literally. If you hit a wall during install, tell us exactly where; every stumble improves the setup for the next person. And if you want to see the methodology applied to your business before installing anything, start with a free Quick Audit.