The setup flow
Capture, then tech stack, then first ingest, then connect. Four pillars, in order, each with a concrete deliverable. Roughly an hour of work for the first ingest on the self-serve path.
Brain 101 / The full method
A brain is a folder of your work that any AI can read.
Not another app. Not another login. It is a markdown brain you own, hosted in our cloud, downloadable any time. LAN indexes how you think and serves it to the AI client you already pay for. We never copy your data and we never hold your tokens.
This page is the full conceptual ground before you pick a path. It carries everything the in-product tour used to teach, in one place, before you buy anything.
What a brain is
Your context as a graph of markdown: people, decisions, processes, notes, all linked. Your brain lives in our cloud by default; download a zip any time. Cancel any month and keep it.
Markdown you can read without us
Every page is plain text. Open it in any editor. There is no proprietary format and no lock-in. If you stop paying, the files are still yours and still readable.
A graph, not a pile
Notes link to notes. People link to decisions. Decisions link to the work they produced. The structure is what makes an AI answer from your context instead of guessing.
Your brain, your control
Your brain lives in our cloud by default, with a one-click download (.zip). The LAN connector reads it. Cancel any time and keep it.
Three roles, three tools
Claude Chat is the orchestrator: you tell it what you want and what you do not know. Claude Code is the builder: it reads your brain, runs the ingest, edits files. Cowork is the executor: it uses your computer to act in apps you already use.
How it works
You connect your sources. It learns your context. It returns action: follow-ups, summaries, documented decisions, reports. Three moves, in order.
You connect your sources
Existing material goes in as-is. Documents, exports, notes. Whatever shape it is in is fine. You never edit the brain files by hand: everything structured flows through a conversation, and a raw/ folder holds the originals.
It learns your context
An indexing pipeline (graphify) walks through your material, works out what is a person, a decision, a concept, a process, and writes a structured wiki. The first pass is a draft you react to. Two passes is normal.
It returns action
Once wired, any AI client you use can query the brain. Ask a question that needs your context and it answers from your work, not from a generic guess. From there it drafts follow-ups, summaries, and documented decisions.
The methodology backbone
Four Setup pillars, then three Endgame stages. The same spine you will see in your dashboard once you start. Setup gets you to a working brain. Endgame is the loop that compounds it.
Setup
- 1. Capture and collect
Gather what you already have. You decide what goes in; LAN provides the seed and the method.
- 2. Tech stack
A small set of tools: Claude Code, the graphify skill, Obsidian to view the result. Or hand the first ingest to LAN.
- 3. First ingest
The pipeline turns your raw folder into a brain you can read. You will not love the first pass. That is the point: you iterate.
- 4. Connect
Your brain is live in our cloud. Install the LAN connector, wire MCP into your AI clients, verify it answers from your content.
Endgame
- 1. Automate
Climb from connect to teach to trigger. Connect tools, teach skills, set triggers that run without you typing.
- 2. Iterate
The loop never closes. New context arrives, you drop it in, the brain grows. A single ingest is a draft; six months of iteration is the value.
- 3. Maintain
Structural hygiene, light touch. Backups, health checks, keeping the brain lean. Run the checklist monthly.
The architecture
Your brain is a server your AI client queries over MCP. Your brain in our cloud on one side, the LAN connector on the other. Tokens never leave you.
Your brain is the source of truth
The brain content lives in your private LAN-hosted bucket. It is the stable address and the source the connector reads from. Nobody but the apps you authorize sees it. Download a zip any time.
The LAN connector is read-only and scoped
The LAN connector reads from your private brain bucket only. It is the connector, not a copy. Your brain stays yours; download a zip any time.
MCP wires the brain into your AI client
You paste one MCP URL into Claude Desktop, Claude Code, ChatGPT, or any MCP-aware client. The OAuth flow handles auth. You never see or paste a token.
Lightweight and full-power modes
Lightweight mode (Claude Chat, ChatGPT, Gemini): add the MCP URL once and the brain is available in any conversation. Full-power mode (Claude Code, Cowork): local agents read, write, edit, and run skills with your authorization.
Skills and automation
A skill is a recorded process plus the critical thinking patterns you use. Automation is a three-step climb you take one step at a time.
1. Connect
Add the brain into every AI surface where you work. Chat, Code, Cowork. Day-zero coverage.
- Skill gap
- None to low. Most clients have a settings panel for MCP.
- Learning curve
- Hours.
2. Teach
Skills are processes you do, written down so the AI can repeat them. You teach a skill by doing the work once alongside Claude Code and asking it to write the file.
- Skill gap
- Medium. Write the skill once, the brain repeats. Skills compose: one calls another.
- Learning curve
- Days.
3. Trigger
A trigger is an event outside your invocation: a new email matching a pattern, a new commit, a new calendar invite. Triggers wrap skills so work happens without you typing.
- Skill gap
- Higher. Each trigger needs an event source plus a skill to wrap.
- Learning curve
- Weeks.
You climb one step at a time. Most people connect tools for weeks before writing their first skill, and teach skills for months before automating the first trigger.
Worked examples
The artifacts the in-product tour used to carry, in one place. The setup flow, what a healthy graph looks like, and the shape of a real wiki entry.
A healthy graph
Open Obsidian on the brain folder after the first ingest and click the graph view. Healthy means a few hubs and a long tail of smaller nodes. Disconnected dust means re-tweak and re-run.
A sample wiki entry
Each entry is markdown with frontmatter (type, status, links) and a body. A decision entry records the context, the decision, the alternatives, and the tradeoff. The links are what make it queryable.
type: decision status: active links: [[ai-client-strategy]], [[connector-scope]] Context. We need every AI client to read the brain without a custom integration per client. Decision. Expose the brain over MCP. One URL, the OAuth flow handles auth, no tokens shown to the user. Tradeoff. MCP support varies by client today. We accept that: the clients that matter most support it now and the rest are adding it.
When to use Chat, Cowork, or Claude Code
All three reach the same brain over the same MCP connection. None of them can do something the others fundamentally cannot. The split is about ergonomics and depth, not capability walls. Pick the one that fits the shape of the work in front of you, and switch freely.
Chat
The claude.ai web app. Lowest friction. Open a tab, ask a question, read what comes back. Best when you are thinking out loud, checking a fact in your brain, or pulling one answer.
- Best for
- Ask my brain a question.
Cowork
Structured task workflows against the brain. Multi-document operations, longer reasoning that spans several steps, work you want to hand off and come back to. Best when one prompt is not enough and you want the brain to carry a task through.
- Best for
- Run a multi-step task on my brain.
Claude Code
The deepest integration, for builders. Repositories, file operations, agentic development. It reads and writes the brain as files and runs work end to end. Best when the work is code, or when you are editing the brain itself.
- Best for
- Build with my brain in the loop.
If you only learn one, learn Chat: it covers most days. Reach for Cowork when a task outgrows a single message, and Claude Code when you are in a repository.
The full FAQ
Everything the in-product tour answered, before you buy.
What exactly is a brain?
A private store of your context, written as linked markdown: people, decisions, processes, notes. An indexing layer turns it into something any AI client can query over MCP. It is your data, in our cloud, in a format you can read and download without us.Do I own my data?
Yes. Your brain lives in our cloud by default and is downloadable as a zip any time. You own the data. We do not use your content for training.Do you hold my AI provider tokens?
No. The brain runs on your AI client. You bring your tokens. We never see them. The connector uses an OAuth flow, so no token is ever shown to you or stored by us.What if I cancel?
Cancel any month. The MCP endpoint stops responding. You keep your brain: download the zip any time and keep it. The markdown is still readable in any editor.How is this different from Notion AI, Glean, or mem.ai?
Those aggregate your data into their store and answer from the copy. LAN indexes how you think and answers through the AI client you already pay for. We never duplicate your data and we never hold your tokens.How is this different from Obsidian with plugins?
Obsidian is local-first storage without an institutional indexing layer or an MCP surface. The brain method adds the structured ingest, LAN-hosted storage you can download any time, and the connector that exposes it to any AI client.Why use Sonnet and not Opus for the first ingest?
The first-ingest prompt is tuned for Sonnet. It is faster on the long context the pipeline needs and the quality is fine for a draft. You iterate from there.Do I have to edit markdown files by hand?
No. Everything structured flows through a conversation with the AI. A raw/ folder holds your originals. The brain itself is written by the pipeline, not by you.What AI clients work with the brain?
Claude (chat and Code), ChatGPT, Gemini, and any MCP-aware runtime. Lightweight clients add the MCP URL once; full-power clients (Claude Code, Cowork) can also write and run skills with your authorization.How long does the first setup take?
Complete the intake form, upload your material, and your brain is built automatically. Most builds finish in minutes to a few hours. We email you when it is ready and you pick up at the connect step.What is a skill?
A recorded process plus the critical thinking patterns you use. You teach it by doing the work once alongside Claude Code and asking it to write the skill file. Skills compose: one can call another.What is a trigger?
An event outside your invocation, like a new email matching a pattern or a new calendar invite, wrapped around one or more skills so work happens without you typing.Is my brain private?
Yes. Your brain lives in a private tenant-scoped store. Only you can access it. Only the apps you authorize via OAuth can read it.