Resources
The library.
The dictionary, the workbook, and the free course — everything we'd hand an operator before they hire us. The method in the open. Take it.
The workbooks, given away.
The workbooks from the Lab courses, yours to keep. No gate, no email — every PDF links straight to the file. Start with the full Install Track Workbook, or grab just the piece you need.
Install Track Workbook
Every chapter, the whole curriculum — the same one every course and every install runs on. The audit in paper form. Free · PDF.
Printing it? Toner-friendly print version (PDF) →
Security Workbook
Working safely with AI — what to lock down before you hand it the keys. Free · PDF.
Code Workbook
Reading agent work — how to follow what the machine actually did. Free · PDF.
Code Glossary
The terminology reference card — 29 terms, plain language, one page. Free · PDF.
How to Think About AI in Your Business
The operator’s foundations. Not theory: every lesson ends in something you do. You leave with a written baseline, your own operating rules, five moat scores, and your first real delegation. 3 hours · Beginner. Free, no card.
The rest of the Track runs inside the Lab as lessons and in-account worksheets. See the full curriculum →
The dictionary.
Every term we use across the site, in plain language. The same definitions that pop up when you hover a dotted term in the copy — gathered here to read straight through.
- Branded subdomain
- Your install lives at a subdomain you brand and gate (e.g. yourname.brain.example.com). Token-protected, scoped to you. The data and the endpoint move with you across platforms and over time.
- Company Brain
- A Company Brain install — same install pattern as the Personal version, scoped to a team or function instead of one human. Custom at scale rather than productized at the unit.
- Embedded corpus
- Your work files chunked and embedded so the system can find what you mean, even when the keywords don’t match. Yours. Gated to you. Owned end-to-end. Walks with you when you leave any platform.
- Embedding
- A numeric representation of a piece of text. Lets the system find conceptually-similar text even when the keywords don’t match — "layoff conversation" and "termination meeting" have similar embeddings.
- Install pattern
- The repeatable architecture behind every Personal Brain install: corpus → retrieval → role skills → MCP endpoint → branded subdomain. Same pattern across every tier; what changes is scope (one person vs. team).
- LLM · Large Language Model
- The underlying AI model (Claude, GPT-4, Gemini, etc.). Your install is platform-agnostic — any MCP-compatible LLM client can call into the same corpus and skills.
- Markdown bundle
- A folder of plain Markdown files (.md) — human-readable, AI-readable, future-proof. The format your corpus ships in. No proprietary database, no vendor lock-in.
- MCP · Model Context Protocol
- The standard way AI apps (Claude Desktop, ChatGPT, etc.) talk to external tools and data sources. Your install runs as an MCP server — the LLM sees your corpus and skills as another set of tools, no platform lock-in.
- Personal Brain
- A Personal Brain install — the productized version of the install pattern, sized to one human and one role. Distinct from a Company Brain (team-scoped, same architecture, different scope).
- Retrieval
- How the system pulls the right slice of your corpus into the AI’s context — at the moment the LLM needs it. Different roles need different retrieval shapes (semantic, lexical, hybrid, graph, tabular).
- Retrieval chunk
- A small slice of a document — typically a section or a conversation turn — that gets indexed individually. Chunking lets retrieval return a precise snippet instead of an entire 30-page memo.
- Retrieval shape
- The pattern of retrieval that fits a role’s actual work. A researcher needs semantic search over prose; a finance lead needs tabular retrieval; a relationship-manager needs a graph of who-knows-whom. The shape decides what the install actually feels like.
- Role-tuned skills
- Claude Code skills installed alongside the corpus — proven prompt-and-workflow recipes for the work you actually do. The skill set evolves as the role does.
- Semantic vs. lexical search
- Semantic = matching by meaning (uses embeddings). Lexical = matching by exact keyword (uses BM25 ranking). Your install runs both at once and merges the results, so you don’t have to choose.
- The four layers
- Four kinds of working intelligence that live inside any AI account: memory (what it remembers about you), context (what you give it per session), skills (the recipes it runs), and corpus (the documents it can search). Naming them is the first step to making them portable.
- Vault
- The folder of Markdown files that holds your corpus — conversation history, notes, methodology, references. Browse it directly in Obsidian, VS Code, or any text editor. Portable when you leave.
Reading
Long-form essays on memory, ownership, and the human side of installing AI. Mirrored from the M.L. Sebastian Substack.
Case material
Public-shareable case slices from real installs. Annotated to show the architecture decisions — what got installed, what got measured.
Want something specific?
A framework, a doc, a download that would help you think through an AI install? Ask. Most of what ends up in this library started as a question someone needed answered.