Your clients have a knowledge problem. Claude + Obsidian fixes it. $2,200 setup. $9,000/month.

Every company has a wiki. A Google Drive folder nobody opens. A Notion workspace that was organized once in 2022 and never touched again. A Confluence page where processes go to die.

The problem is not the tool. The problem is that the tool does not talk back.

Obsidian connected to Claude talks back. It reads every note, every process, every client file, every meeting summary, and answers questions about them instantly. Ask it anything about the business and it responds like a colleague who has read everything and forgot nothing.

Six clients pay $1,500/month each to have this running. Each one paid $2,200 to get it set up. That is $9,000/month in retainers plus setup fees every time a new client joins.

Here is the complete system.

What Obsidian actually is

Obsidian is a note-taking app where everything is stored as plain markdown files on the local machine. No cloud. No subscription required for the basic version. No data sent anywhere.

The reason it works for this system: Claude can read markdown files. Feed it the contents of an Obsidian vault and it has full context on everything inside - processes, clients, team members, meeting notes, decisions made six months ago.

Standard company knowledge setup:
  Google Drive:    files nobody can find
  Notion:          pages nobody updates
  Slack history:   searchable but overwhelming
  Email threads:   context buried and lost
  New employee:    spends 3 months asking questions

Obsidian + Claude setup:
  One structured vault per person or team
  Claude reads all of it
  Anyone asks a question → gets a specific answer
  New employee: full context on day one

The difference is not storage. It is retrieval. Every tool stores information. Obsidian plus Claude retrieves it conversationally.

Personal vault vs enterprise vault

This is the distinction that creates the business model.

A personal vault is one person, one vault. It knows their notes, their projects, their tasks. Useful for individuals. Does not scale to teams.

An enterprise vault is different:

The enterprise version is what gets sold. Every exec who sees a demo wants one. Most companies are still trying to solve internal knowledge with shared logins and a Notion that one person updates once a quarter.

How it works technically

Three components. Each one has a specific job.

Obsidian holds all the knowledge. Claude reads and responds. n8n or a Telegram bot is the interface employees actually use.

The employee opens Telegram or Slack. Types a question. Claude reads the relevant vault files and responds with a specific answer including the source note and the date it was last updated.

No more "ask Sarah." No more digging through Drive. No more onboarding that takes three months.

Building the vault structure

The vault structure is the product. A well-structured vault is what makes Claude's answers specific and trustworthy. A poorly structured vault produces vague answers nobody trusts.

Use this prompt to generate the initial vault structure for any client:

You are helping set up an Obsidian knowledge vault
for a [business type] with [number] employees.

The vault will be connected to Claude API.
Employees will ask questions and Claude will answer
using vault contents as context.

Generate a complete folder structure and list of
core files to create. For each file include:
- File name and path
- What content it should contain
- Who should have access (all / role-specific)
- How often it should be updated

Business context:
- Industry: [describe]
- Team size: [number]
- Departments: [list]
- Main processes that need documentation: [list]

Output as a structured list.
Prioritize files that answer the most common
questions new employees ask in the first 30 days.

After generating the structure, use this to write the first version of each file:

You are writing the [file name] file for an
Obsidian knowledge vault at [company name].

This file will be read by Claude API to answer
employee questions. Write it so that Claude can
extract specific, accurate answers from it.

Rules:
- Use clear headings (## and ###)
- Include specific names, dates, numbers
- No vague statements — if it can't be specific,
  leave it out
- End each section with "Last updated: [date]"
  and "Owner: [name]"

Topic to document: [describe the process or information]

Write the complete file content.

The system prompt that makes it work

Every employee gets a different system prompt. The system prompt is what tells Claude who it is talking to, what vault files to reference, and what it cannot discuss.

You are the internal AI assistant for [Company Name].

YOUR IDENTITY:
You work exclusively for [Company Name].
You have read and memorized every document
in the company knowledge vault.
You answer questions as if you are a senior colleague
who has been at the company for 5 years.

THE EMPLOYEE YOU ARE TALKING TO:
Name: [employee name]
Role: [job title]
Department: [department]
Access level: [standard / manager / executive]

WHAT YOU KNOW (vault contents injected here):
{{vault_contents_for_this_role}}

RULES:
- Only answer using information from the vault
- If the answer is not in the vault, say:
  "I don't have that documented yet.
   Ask [relevant person] and I'll remember it
   once it's added to the vault."
- Never guess. Never invent information.
- Always cite which document your answer comes from
- For sensitive topics (HR, legal, financial),
  confirm with the relevant department head

TONE:
Direct. Specific. Like a knowledgeable colleague,
not like a chatbot.
No "certainly" or "great question."
Just the answer.

The n8n workflow that connects everything

n8n sits between the employee's Telegram and the Claude API. It handles reading the vault, injecting the system prompt, and storing conversation history.

The n8n HTTP node that calls Claude API:

{
  "method": "POST",
  "url": "https://api.anthropic.com/v1/messages",
  "headers": {
    "x-api-key": "{{ $env.ANTHROPIC_API_KEY }}",
    "anthropic-version": "2023-06-01",
    "content-type": "application/json"
  },
  "body": {
    "model": "claude-opus-4-6",
    "max_tokens": 1024,
    "system": "{{ $json.system_prompt }}",
    "messages": [
      {
        "role": "user",
        "content": "{{ $json.employee_question }}"
      }
    ]
  }
}

What is included in the $2,200 setup

The setup fee covers five deliverables. Each one is documented so the client understands exactly what they are paying for.

Deliverable 1 — Vault architecture (Day 1):
  Complete folder structure for their business
  List of all files to create
  Access levels per department

Deliverable 2 — Core vault content (Day 2):
  10–15 core files written and formatted
  Processes, team directory, client overview,
  FAQs, onboarding guide

Deliverable 3 — Claude integration (Day 2–3):
  System prompts per role
  n8n workflow connecting Telegram to Claude
  Access filters working correctly

Deliverable 4 — Team setup (Day 3):
  Each employee added to their Telegram bot
  Test session with 2–3 team members
  Vault corrections based on their questions

Deliverable 5 — Documentation (Day 3):
  How to add new notes to the vault
  How to update existing files
  What to do when Claude gives a wrong answer

One person can deliver this in three days. After the first client the process takes two days.

What is included in the $1,500/month retainer

The retainer is what makes this a real business instead of a one-time project.

Weekly (every Monday, 30 minutes):
  Review questions Claude could not answer
  Add missing information to vault
  Update any outdated files flagged by employees

Monthly:
  Add new processes documented during the month
  Update team directory (hires, departures)
  Review Claude API costs (usually $30–60/month)
  Send client a usage report

Quarterly:
  Full vault audit — what is outdated, what is missing
  New employee onboarding if team grew
  System prompt updates based on how team uses it

What the client does not need to do:
  Touch n8n
  Update system prompts
  Manage Claude API
  Think about the technology at all

The retainer is pure maintenance. No new deliverables. No project management. Thirty to sixty minutes of work per week per client.

Who pays for this

Three types of businesses pay quickly and without negotiation.

Law firms. Every associate asks the same questions about billing codes, client protocols, and matter procedures. The vault answers them without a partner's time. One hour of partner time saved per week justifies $1,500/month immediately.

Agencies. Onboarding new employees costs real money. A vault that answers every process question cuts that cost in half. Any agency that hires more than two people per year sees the ROI instantly.

Medical practices. Protocols, insurance procedures, patient communication scripts - all of it documented once and accessible instantly. HIPAA compliance means this runs locally with Ollama instead of Claude API, but the setup and retainer pricing holds.

The pitch in one sentence:

Your new employee will read every process, every client file, and every decision made in the last three years before their first day. And they'll never forget any of it.

The sales process

No cold outreach is needed. The demo does the selling.

Build a demo vault using publicly available information about the prospect's industry. Show them the bot answering questions about their type of business. Let them ask it something. It answers specifically. They ask how soon it can be set up.

Week 1: build demo vault (3 hours)
  Use their website, LinkedIn, public info
  Create 5–6 realistic files
  Set up Telegram bot with system prompt

Demo call (30 minutes):
  Let them talk to the bot
  Ask: "What questions does your team ask
         most often in their first month?"
  Watch them realize the vault answers all of them

Close: "Setup takes three days. $2,200.
        Retainer is $1,500/month.
        Want to start this week?"

Conversion rate on demos: roughly 60–70%

The demo vault takes three hours to build. A sixty percent close rate means two demos equals one client. Two demos per week equals one new client per week at full conversion.

The math - six clients in

Month 1:
  Client 1 setup:      $2,200
  Client 1 retainer:   $1,500
  Total:               $3,700

Month 2:
  Client 2 setup:      $2,200
  2 retainers:         $3,000
  Total:               $5,200

Month 3:
  Client 3 setup:      $2,200
  3 retainers:         $4,500
  Total:               $6,700

Month 6 (6 clients):
  Setup fees:          $2,200 (new client)
  6 retainers:         $9,000
  Total:               $11,200

Active time at 6 clients:
  Retainer work:       6 × 45 min = 4.5 hours/week
  New client setup:    6 hours/week when onboarding
  Total:               ~10 hours/week at capacity

At ten hours per week and $11,200/month that is $1,120 per hour of work. The leverage comes from the retainer - clients stay because the system keeps working and they do not want to manage it themselves.

How to start this week

Build a demo vault for a business type familiar from personal experience. A previous employer, a client, an industry known well. Use their publicly available information.

Connect it to Claude API. Set up a Telegram bot through n8n. Write a system prompt for a typical employee at that business.

Send a message. Get an answer. That is the demo.

Find one business owner in the network. Offer to show them something. Run the demo. If they ask how much, the answer is $2,200 setup and $1,500/month.

The vault took three hours to build. The first client meeting takes thirty minutes. Everything else is retainer.

That is the system. Obsidian stores the knowledge. Claude retrieves it. Employees use it daily. A retainer keeps it current.

Every exec who sees a demo wants one. Most companies are still solving this with a Notion page nobody updates.

Client seven is already scheduled.

→ Follow practical AI systems posted every week
→ Save this every prompt is copy-paste ready above
→ Like if you're booking the first demo this week