Claude Code and Obsidian: memory in 30 minutes
Three markdown files take Claude Code from a stranger to a colleague who knows your company: one about the business, one per service and one about your tone of voice. A Claude Code and Obsidian setup keeps those files in a folder you can read and edit like any notebook, so you stop repeating the same background in every session. The foundation takes thirty minutes, and it is the difference between asking an AI for advice and handing it the keys to the office.
The agent knows nothing, yet
The first thing people notice when they open Claude Code is the questions it asks back. What do you sell, who are your customers and what is the rule for discounts are all fair questions, exactly the ones a new hire asks in the first week.
The difference with an agent is that you only explain once. Claude Code reads a file called CLAUDE.md at the start of every session, so whatever you write there is known in every conversation that follows.1 That folder of files is the context library, and building it changes what you can hand over.
Is Claude Code only for developers?
Claude Code is sold as a coding tool, and it does that job well. Anthropic's growth marketing team feeds it a CSV file with hundreds of ads, lets it find the ones that underperform and has it write new variants within the character limits in minutes instead of hours.2 Its legal team built a prototype phone tree with it that routes colleagues to the right lawyer, without a developer involved.2
The point is plain. An agent is only as good as the context it gets, and you control the context.
What is a context library?
Picture a folder on your computer with markdown files inside, meaning plain text files with a few symbols for headings and lists. One file per customer, one per service, one for your search strategy, one for onboarding and one that explains how you want emails written.
Claude Code loads the CLAUDE.md files in the folder and the folders above it when it starts, and CLAUDE.md can pull in other files with an import line such as @services/overview.md.1 Imported files can import further files up to four hops deep, so a small CLAUDE.md can point to the whole library.1 Anthropic suggests keeping each CLAUDE.md under 200 lines, since longer files take up context and Claude follows them less consistently.1
With the library in place you can ask for real work. Draft a reply to this customer email, write a blog post that follows our search strategy, prepare a proposal for the client in Rotterdam based on her notes. The agent already has the background and only needs the instruction.
Why Claude Code and Obsidian fit together
Obsidian is a note-taking app that stores every note as a plain markdown file in a folder on your own disk, with no proprietary format and no cloud lock-in.3 Claude Code reads, edits and creates those same files, so the two work on one folder without any export or sync step.
Obsidian gives you the comfortable interface for building and keeping the library tidy, with a link graph that shows how files connect, live preview of the formatting, tags and folders for hundreds of documents. The librarian's desk is Obsidian, while Claude Code is the reader who uses the library. One weak spot remains, since Obsidian will not stop you from writing a customer note full of personal data, so the library needs the same rules as any other shared document.
Six kinds of files to start with
You do not need everything on day one. Start with the files that answer the questions the agent asks most.
1. Customer folders
One file per customer, with contact details, history, what they bought, their preferences and open issues, plus short notes after every meeting. When you ask for a follow-up email to the customer in Lyon, you skip the background, because the agent reads the file.
2. Service descriptions
One file per service, covering what is included, the price, the arguments, the usual objections and how you differ from competitors. The agent can then draft proposals, answer product questions and write landing page copy that matches what you actually sell.
3. Search strategy
One file with your priority keywords, which pages they belong to, how you want to be positioned and which topics you own. A blog post drafted by the agent then uses the right keywords, links to the right pages and keeps your tone. We run our own site this way, with the search strategy as one markdown file the agent reads before it writes a word.
4. Subagent instructions
Slightly more advanced and worth it. You can define specialist agents as markdown files, for example one that always writes in your brand voice, one that checks a text for GDPR risks and one that reads everything from the customer's side. Saved as files, the same instructions give the same quality every time you call them.
5. Process documentation
Onboarding steps, the routine for a new customer, the rules for support tickets and who approves what. Written down in markdown, the agent can follow the process, spot a missing step and carry out parts of it.
6. Content calendar
Topics, channels, deadlines and the approval routine. With these written down, the agent can suggest the next post, write a first draft and check that it fits the plan.
CLAUDE.md: where everything starts
The core of the whole system is one file. CLAUDE.md sits in the folder and loads at every new session, and it works as the agent's onboarding document.1 Running /init in Claude Code writes a first version for you, which you then trim.1
A good CLAUDE.md for a business library can look like this:
# [Company name]: context for Claude Code
## About us
We are a [industry] firm focused on [audience]. Read @context/about-us.md for the full background.
## Customers
Customer folders live in @customers/. Each one contains a client-brief.md.
## Services
See @services/overview.md for every service with prices and arguments.
## Tone of voice
Warm, direct and concrete. No buzzwords. See @brand/tone.md.
## Search
Read @seo/search-strategy.md before writing any web copy.
## Important
Never contact customers directly. Every message is a draft I approve.
The last line matters most. You set the boundaries, and the agent works inside them.
What can the agent do with all this?
Concrete examples say more than principles.
Ask for "a welcome email to our new customer Lisa, who just bought the starter package", and the agent reads Lisa's file and the tone file first. The draft comes out personal and on brand, so you check it and press send.
A request for "a blog post about AI for accounting firms that follows our search strategy" makes the agent read the strategy and pick the keywords. It delivers a draft with the right structure and internal links in about three minutes, where the first draft used to take an afternoon.
Meeting notes work the same way, for example "we met the logistics team yesterday, update the customer file and draft a summary". The agent updates the file and writes the summary, and a person still reads it before it goes out, because the agent sometimes credits a decision to the wrong person when the notes are thin.
A checklist to get started
In the right order:
- Create a folder on your computer called "context-library"
- Install Obsidian and open the folder as a vault
- Create CLAUDE.md with a short overview of what sits where
- Write one file about your company
- Write one file per service you sell
- Create a customer folder with one file per active customer
- Put your tone of voice and writing style in one file
- Test it by asking Claude Code for one concrete task
- Add context whenever you notice the agent lacks information
The foundation takes an afternoon at most, and it pays back every working day after that.
How satori. helps you get started
The context library is one of the most useful ideas we set up when a European company starts using Claude for real work. It rarely needs new software, only the knowledge you already have, written down in one place. Claude Code comes with Pro, Max, Team and Enterprise, so a company on Claude Team already has it on every seat.4
satori-launch sets up the Claude accounts, the folder structure and the first five skills, then trains your team in a webinar and a full-day workshop, for €560 per user from eight users. Teams that want the licences, support and one invoice in euros handled can use satori-claude, and more on agents in daily work is in our guide to agentic AI with Claude Code. All satori. prices are fixed and exclude VAT.
Create a folder called context-library, write a CLAUDE.md with five lines about your company and ask Claude Code to draft a reply to the last customer email you received. In twenty minutes you see how much the answer changes once the agent knows who you are.
Sources
Footnotes
-
Anthropic (2026). How Claude remembers your project. Claude Code Docs. https://code.claude.com/docs/en/memory ↩ ↩2 ↩3 ↩4 ↩5 ↩6
-
Anthropic (2025). How Anthropic teams use Claude Code. https://claude.com/blog/how-anthropic-teams-use-claude-code ↩ ↩2
-
Obsidian (2026). Obsidian Help. https://help.obsidian.md/ ↩
-
Anthropic (2026). Plans and Pricing. https://claude.com/pricing ↩