Datacore

Own Your Intelligence.

An open-source framework for building AI-automated businesses. Datacore gives Claude (and other MCP-compatible agents) the context, structure, and autonomy to run day-to-day operations while you focus on strategy.

Quick Start

Option 1: Let your AI install it

Tell Claude Code (or Cursor, Windsurf, OpenClaw):

"Go to datacore.one and install Datacore."

Option 2: CLI

npx @datacore-one/cli init

Sets up ~/Data, clones modules, and configures the MCP server automatically.

Option 3: MCP server only

npx @datacore-one/mcp init

Then add to .claude/mcp.json or .cursor/mcp.json:

"datacore": {
  "command": "npx",
  "args": ["-y", "@datacore-one/mcp"]
}

Then open Claude Code and try /today or /continue. See GETTING_STARTED.md for a full walkthrough.

What is Datacore?

It starts as an extended mind. It becomes an autonomous business.

Stage 1 — Extended mind        ← start here
  AI that knows your work, remembers your decisions, surfaces what matters.
  Persistent memory via PLUR. GTD task management. Zettelkasten knowledge base.

Stage 2 — Autonomous business  ← where most users end up
  Agents run day-to-day operations: content, research, outreach, coordination.
  Queued during the day. Executed overnight. Reviewed in your morning briefing.

Stage 3 — AI business network  ← the horizon
  Agents from different businesses collaborating and exchanging value.

Your data stays on your drive. You control the agents. You set the direction.

At its core, it provides:

• Autonomous execution -- Delegate tasks to AI agents overnight; wake up to a quality-evaluated briefing
• GTD task management -- Capture, organize, and delegate tasks using Getting Things Done methodology with org-mode
• Knowledge management -- Zettelkasten-style notes, wiki-links, and semantic search across your knowledge base
• Modular architecture -- Install only what you need; extend with community or custom modules
• Persistent memory -- Powered by PLUR (preinstalled): corrections, preferences, and decisions survive across sessions

How It Works

You capture ideas and tasks
        |
Datacore organizes, links, and indexes them
        |
AI assistants access your knowledge and context via MCP
        |
Agents execute delegated work overnight
        |
You review results in your morning briefing

Prerequisites

• Claude Code -- AI coding assistant
• Git and GitHub CLI
• Python 3.8+

Architecture

~/Data/
|
+-- .datacore/                    # System core
|   +-- agents/                   # AI agent definitions
|   +-- commands/                 # Slash commands (workflows)
|   +-- modules/                  # Installed modules
|   +-- lib/                      # Python utilities
|   +-- specs/                    # System specifications
|   +-- dips/                     # Design proposals
|   +-- registry/                 # Agent, command, source registries
|   +-- state/                    # Runtime state (gitignored)
|   \-- env/                      # Secrets (gitignored)
|
+-- 0-personal/                   # Personal space
|   +-- org/                      # GTD system (org-mode)
|   +-- notes/                    # PKM (Obsidian)
|   +-- code/                     # Personal projects
|   \-- content/                  # Generated content
|
+-- [N]-[name]/                   # Team spaces (separate repos)
|
+-- CLAUDE.md                     # AI context (layered, auto-generated)
+-- install.yaml                  # Installation manifest
\-- sync                          # Multi-repo sync script

Key Concepts

Spaces -- Isolated workspaces for different contexts (personal, teams, organizations). Each space has its own GTD system, knowledge base, and journal. Team spaces are separate git repos.

Agents -- AI agent definitions that handle specific types of work: inbox processing, content writing, data analysis, research orchestration, project management, and more.

Commands -- Slash commands that orchestrate multi-step workflows: /today (morning briefing), /continue (resume work), /tomorrow (end-of-day delegation), /wrap-up (session close).

Modules -- Optional extensions that add domain-specific functionality. Install only what you need.

Layered Context -- Configuration files use a four-layer privacy model (public, org, team, private) so you can contribute improvements upstream without exposing personal data.

Memory -- Persistent memory is handled by PLUR, an open-source engram engine that comes preinstalled. Corrections, preferences, and decisions survive across sessions and are injected automatically — no setup needed.

Modules

Public modules available for community use:

See the Module Catalog for installation instructions and the full list of available modules.

Documentation

Contributing

Datacore uses a fork-and-overlay contribution model. Fork the repo, make improvements to public layer files (.base.md), and submit a PR upstream. Your private configuration stays local and is never shared.

See CONTRIBUTING.md for full guidelines.

License

MIT License -- see LICENSE for details.

Datacore is built by Datacore. The AI system that bootstraps itself into existence.

datacore.one · github.com/datacore-one/datacore