Learn

Build a CRM your agents can use (it's just a folder)


A CRM that AI agents can use can be as simple as a shared folder with one Markdown file per company, a small CSV for the pipeline, and a SKILL.md file that tells any agent how to read and update the records. Put that folder in a shared Fabric drive and it works for everyone at once: a person opens the files in Fabric, Claude Code reads them on the mounted drive, and Claude or ChatGPT reach them through MCP. Someone builds it once and the whole team has it.

This page shows a working layout, an example skill, and how the same pattern covers a monthly report and a knowledge base.


Why a folder works as a tool

When one person builds a useful agent workflow on their laptop, it usually stays there. Their prompt, their notes file and their scripts live in a home directory nobody else can see. A colleague who wants the same thing starts again from scratch, with slightly different field names, and soon the team has three half-compatible versions of the same contact list.

A shared folder avoids that, because the tool and the data live in the same place. The records are plain text that both people and agents can read.

The skill file describes the rules in plain language, so any agent that reads it follows the same conventions. And because it's a folder in a shared drive, a fix to the skill or a new record is there for everyone the next time they open it. Nobody has to export, import or sync anything.

There's no app to integrate either. If an agent can open files, it can use the CRM.


The folder layout

Here's an illustrative structure for a small B2B team:


crm/
  SKILL.md
  pipeline.csv
  companies/
    northwind-dental.md
    halden-studio.md
    ortega-freight.md
  people/
    priya-nair.md
    tom-okafor.md
  notes/
    2026-11-14-ortega-call.md
crm/
  SKILL.md
  pipeline.csv
  companies/
    northwind-dental.md
    halden-studio.md
    ortega-freight.md
  people/
    priya-nair.md
    tom-okafor.md
  notes/
    2026-11-14-ortega-call.md
crm/
  SKILL.md
  pipeline.csv
  companies/
    northwind-dental.md
    halden-studio.md
    ortega-freight.md
  people/
    priya-nair.md
    tom-okafor.md
  notes/
    2026-11-14-ortega-call.md

Each company gets one file in companies/, named with a lowercase slug so agents can find it by name. Contacts live in people/ and link back to their company. pipeline.csv holds one row per open deal, which keeps stage and value easy to sort and easy for an agent to summarize. Call notes and meeting write-ups go in notes/, dated so they sort in order.

A company record needs only a few fields at the top and a running activity section below:


markdown

# Ortega Freight

Stage: proposal
Owner: Sam
Value: 18,000
Next step: send revised scope by 2026-11-21

## Activity
2026-11-14 Call with Tom Okafor. Wants phased rollout, Q1 start.
2026-11-02 Intro from Priya Nair at Halden Studio.
# Ortega Freight

Stage: proposal
Owner: Sam
Value: 18,000
Next step: send revised scope by 2026-11-21

## Activity
2026-11-14 Call with Tom Okafor. Wants phased rollout, Q1 start.
2026-11-02 Intro from Priya Nair at Halden Studio.
# Ortega Freight

Stage: proposal
Owner: Sam
Value: 18,000
Next step: send revised scope by 2026-11-21

## Activity
2026-11-14 Call with Tom Okafor. Wants phased rollout, Q1 start.
2026-11-02 Intro from Priya Nair at Halden Studio.

That's enough for a team of a few people. You can add fields later, and because the rules live in the skill, every agent picks up the change.


A short example SKILL.md

The skill is where the CRM's rules live. Skill files follow the Agent Skills format: a short YAML header with a name and a description of when to use it, then instructions in Markdown.

markdown

---
name: crm
description: Read and update the team CRM in crm/. Use when logging a call, adding a company or contact, or asking about a deal.
---

# Team CRM

Before creating a company, search companies/ for its name.
Update the existing file if there is one.

Name new files with a lowercase slug, e.g. ortega-freight.md.

Add activity as a new dated line at the top of the Activity
section. Don't edit old lines.

When a deal's stage or value changes, update the company file
and its row in pipeline.csv.

Save call notes in notes/ as YYYY-MM-DD-company-topic.md.

Don't delete records. If something looks wrong, add a line to
notes/questions.md and leave it for a person.
---
name: crm
description: Read and update the team CRM in crm/. Use when logging a call, adding a company or contact, or asking about a deal.
---

# Team CRM

Before creating a company, search companies/ for its name.
Update the existing file if there is one.

Name new files with a lowercase slug, e.g. ortega-freight.md.

Add activity as a new dated line at the top of the Activity
section. Don't edit old lines.

When a deal's stage or value changes, update the company file
and its row in pipeline.csv.

Save call notes in notes/ as YYYY-MM-DD-company-topic.md.

Don't delete records. If something looks wrong, add a line to
notes/questions.md and leave it for a person.
---
name: crm
description: Read and update the team CRM in crm/. Use when logging a call, adding a company or contact, or asking about a deal.
---

# Team CRM

Before creating a company, search companies/ for its name.
Update the existing file if there is one.

Name new files with a lowercase slug, e.g. ortega-freight.md.

Add activity as a new dated line at the top of the Activity
section. Don't edit old lines.

When a deal's stage or value changes, update the company file
and its row in pipeline.csv.

Save call notes in notes/ as YYYY-MM-DD-company-topic.md.

Don't delete records. If something looks wrong, add a line to
notes/questions.md and leave it for a person.

Six rules is plenty. The search-before-create rule prevents duplicate companies, the append-only activity log keeps history intact, and the last rule sends anything unclear to a person instead of letting an agent guess. Share skills with your whole team through a shared drive covers how teams keep skills like this one in a single place, and agent skills in Fabric shows more examples.


How every person and agent uses it

People use the CRM like any set of documents. They open a company file in Fabric, read the latest activity before a call, and type a line afterwards. Fabric's own AI assistant can answer questions about the folder if you open it there.

Agents reach the same files whichever way they connect. With the Fabric desktop app, Fabric is mounted as a drive on Mac, Windows or Linux, so Claude Code or Codex on that computer can read crm/SKILL.md and edit records like local files. Claude and ChatGPT connect through the MCP server, where you might say "read crm/SKILL.md, then log this call with Ortega Freight." How AI agents connect to a shared drive compares the routes. A line in your CLAUDE.md or AGENTS.md pointing to the skill saves repeating that instruction.

A few properties of the drive make this hold up when the whole team leans on it.

Agents can search across the folder on the server with grep, regular expressions or full-text search, without downloading anything. Finding every company where Stage: proposal appears, or every note that mentions a competitor, is a single search. Why AI agents give up searching your files explains why that matters.

Two agents will sooner or later update pipeline.csv within minutes of each other. If one saves a change based on an older version of the file, Fabric refuses the write, and that agent re-reads the file and applies its change again. Both updates end up in the file, and no stray copy appears. When two AI agents edit the same file walks through an example.

Agents also can't change SKILL.md in Fabric, even with write access to the folder. They can add and update records, but the rules stay as a person wrote them. When the team decides to track a new field, someone edits the skill by hand, and every agent follows the new version from its next read. Your agent instruction files are a security risk explains why that protection exists.


Controlling what each agent can do

Not every agent needs the whole CRM. With Agent Vault you can give an agent access to the crm/ folder alone, allow reading, searching and creating documents, and leave deleting off. You can set that access to last for a single task, a day, a week or indefinitely, and every action it takes is logged. Agent Vault is on the Pro and Team plans.

If a record does get damaged, earlier versions are kept in version history and deleted files go to a trash you can restore from.


The same pattern for a monthly report

A monthly client or board report works the same way. Make a reports/monthly/ folder with a SKILL.md that names the sources (say, pipeline.csv, the month's files in notes/ and an exported spreadsheet of revenue), describes the sections and links a template. Each month, anyone on the team can ask their agent to run it, and the finished report lands in the same folder next to last month's. Fabric's background agents can run a task like this on a monthly schedule, on Pro and Team plans. Ask your files, don't build a dashboard makes the wider argument for keeping raw exports next to a short notes file.


And for a knowledge base

A team knowledge base is a folder of articles plus a skill that says how to add one: check whether the topic already exists, use the house template, link related articles, put open questions in a separate file. Agents and people both add to it, and every agent that answers a question reads from the same pages. How to create a shared knowledge base covers the structure.


When a folder CRM is enough

A folder CRM suits small teams, early pipelines and anyone who wants their agents to do the updating. It's quick to change and every record is readable without special software. A larger sales team that needs forecasting, sequences and detailed reporting will likely still want a dedicated CRM, and our comparison of CRMs for small teams is a reasonable place to start.

For the bigger picture of teams and agents working from one set of files, see your AI agents need a shared drive, not a group chat.


Frequently asked questions

Can I build a CRM with just Markdown files?

Yes. One Markdown file per company, a CSV for the pipeline and a skill file describing the rules is enough for a small team, and both people and AI agents can read and update it.

What is a SKILL.md file?

It's a Markdown file with a short header giving the skill's name and when to use it, followed by instructions. Agents read it to learn how to do a specific task, such as updating a CRM.

How do AI agents read a CRM stored in Fabric?

Agents on your computer read it through the mounted Fabric drive. Others, such as Claude and ChatGPT, read it through Fabric's MCP server.

Can an AI agent change the rules in my SKILL.md?

No. Fabric protects instruction files such as SKILL.md, CLAUDE.md and AGENTS.md, so agents can't change them even with write access. A person edits them.

What happens if two agents update the CRM at the same time?

If an agent saves a change based on an older version of a file, Fabric refuses the write. The agent re-reads the file and reapplies its change, so neither update is lost.

Can I stop an agent from deleting CRM records?

Yes. In Agent Vault, deleting is a separate permission, so you can allow reading, searching and creating documents without it. Deleted files also go to a trash you can restore from.

Does every teammate need the same AI tool to use it?

No. The CRM is a set of files, so it works with Claude, ChatGPT, Claude Code, Codex, Cursor and VS Code through the same folder.

Can I use the same approach for reports and documentation?

Yes. A monthly report or a knowledge base can be a folder plus a skill describing its sources, structure and rules.


Related pages


The workspace that thinks with you.

Ready when you are.

The workspace that thinks with you.

Ready when you are.

The workspace that thinks with you.

Ready when you are.