Open. Portable. In your hands.

AI First.
From Day 0.

Customer work starts with what you want to achieve. A CRM designed around conversation from the outset, with knowledge, contacts and next steps in your own workspace.

Get started ↗Explore the project

Apache 2.0 · Python · Markdown · Local agent skill

YOUR REQUEST → YOUR CRM
“Import this customer list and show me the duplicates first.”
The import is ready for review. You see new contacts, changes and possible duplicates before any data is written.
Review the previewYou decide →

Illustrative workflow

The project trailer

AI First CRM in 60 seconds.

From your request through verified changes to connected Markdown files. With music, animated examples and the BETA label. The trailer uses German on-screen text.

Download trailerOriginal soundtrack version

60 seconds · Full HD · 16:9

Designed differently from the outset

You describe the goal.
Your agent organizes the work.

01 / CONVERSATION

Your goal comes first.

Import customer lists, organize opportunities and prepare next steps. Describe the task in your own words.

02 / KNOWLEDGE

Connections stay intact.

Companies, people, deals and tasks connect with evidence-backed knowledge. Pipelines and dashboards make the current state visible.

03 / CONTROL

Your data stays portable.

Markdown files, open exports and local HTML views. Your workspace can be read, versioned and taken with you.

Changes with your approval

Understand first. Then change.

Plan → Preview → Confirm → Apply

Review what will change. Checksums bind execution to the plan you saw. If the data changes, a new preview is required.

Inside your workspace

One file per record.
Connections you can read.

Companies, contacts, deals and tasks are stored as Markdown files in your workspace. Stable identifiers connect them. Choose a view to explore the storage, relationships and presentation of the same example.

The same documented example in three views

How it is stored.

Each record has its own file. Its UUID remains stable even when its display name changes.

Your workspace/ └── records/ ├── company/ │ └── 61e37838….md ├── task/ │ └── 0d5b33f5….md └── workspace-member/ └── 57f057d5….md
title: "Angebot nachfassen"
status: "TODO"
assignee: "[[records/workspace-member/57f057d5-431e-490a-9aee-984a166d9b62|Anna Admin]]"
targets:
  - "[[records/company/61e37838-3154-4acc-b2c9-62d6ea20ab04|Acme Holding GmbH]]"

The excerpt shows the task's metadata block. File names in the folder illustration are shortened; full UUIDs are stored.

How the files connect.

The task stores two references: to the company and to the assigned team member.

Task · Markdown

Angebot nachfassen

records/task/0d5b33f5….md
status: TODOStable identifier: crm_id
targets · belongs to
Company · Markdown

Acme Holding GmbH

records/company/61e37838….md
assignee · assigned to
Team member · Markdown

Anna Admin

records/workspace-member/57f057d5….md

Discoverable from the company, too. The inverse relationship is computed from the links. The task does not need to be stored a second time under the company.

Solid arrows represent stored references. The dashed note explains the computed inverse relationship.

How it becomes a readable view.

The generated browser view shows a record and its relationships. The Markdown files remain the underlying data.

CRM · CompanyRead-only view
Company

Acme Holding GmbH

Linked taskAssigned toStatus
Angebot nachfassenAnna AdminTODO

The task references this company, so it can be shown here together with its assignee.

A simplified illustration of existing relationships, not a screenshot of the CRM interface. The actual browser view is generated under graph/crm/.

A documented example with sample data. It illustrates CRM links, not a complete customer record. Stored sample names and values retain their original language. Explore the file format and relationships ↗

Optional Git team mode · GitHub.com

Work together.
In the same skill.

You and your team stay in your respective AI systems. Your agents prepare changes independently and share verified states through private Git storage. No separate CRM application is required.

Agent A

Your AI system

Skill + isolated draft

Agent B

Your colleagues' AI system

The same skill + a separate draft

Preview and confirmation
Preview and confirmation

Protected Git proposals and verification

Identity, write permissions, relationships and the complete data state are checked. Stale proposals are not silently applied.

Published state in private Git

Connected Markdown files, a verified version and generated views.

Your interface stays the same

Describe tasks in conversation. Your agent handles Git steps, previews, checks and verified feedback.

Newer changes are protected

Every proposal is bound to a specific starting state. A changed base requires a new preview instead of an unchecked overwrite.

Reading stays available

Other agents read immutable published states while private drafts are being edited.

Requirements: Python, Git, authenticated GitHub CLI and a private repository with enforced branch checks. All repository readers can copy files and history.Repository writers must be trusted to preserve the skill workflow and its checks; skill roles do not replace GitHub access roles. Ordinary erasure tools do not remove old Git versions; team mode makes this boundary explicit. Explore team mode and setup ↗

Ten everyday tasks and six deeper examples

How it would work for you.

Choose a topic and follow the steps. These animated examples explain how the project works. They do not execute CRM actions.

Open full diagram ↗

Select nodes, explore connections or open the workflow in its own view. The first ten topics cover everyday work. Six more explore data maintenance, recovery and customization.

How the project fits together

Your agent plans. You review and confirm. Local tools write files, build views and publish a verified state.

Open full architecture ↗

Your first workspace

Start small.
Grow with you.

01

Install the skill in an agent with local file access. The runtime requires Python 3.9 or later.

02

Choose a folder and describe your purpose, language and team. Your agent guides you through setup.

03

Import contacts, build your pipeline and open the generated views in your browser.

How the project works today

Clear capabilities.
Clear boundaries.

  • Conversation for interaction, HTML for read-only views.
  • Workflows run on the next invocation; there is no background service.
  • Messages and events come from imported files. Outgoing messages remain drafts.
  • Access control belongs in your file storage. Git team mode uses authenticated GitHub access and protected checks. How the agent processes data depends on your chosen provider.

Read the full capability guide ↗

Open Source · Apache License 2.0

Your CRM. Your workspace.
Your next step.

Installation and documentation ↗