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.
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 initSets up ~/Data, clones modules, and configures the MCP server automatically.
Option 3: MCP server only
npx @datacore-one/mcp initThen 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.
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
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
- Claude Code -- AI coding assistant
- Git and GitHub CLI
- Python 3.8+
~/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
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.
Public modules available for community use:
| Module | Description |
|---|---|
| gtd | Getting Things Done -- task capture, inbox processing, org-mode management |
| nightshift | Autonomous overnight task execution with multi-persona quality evaluation |
| research | Automated research pipelines with knowledge extraction and podcast generation |
| outbox | Content routing out of active workspaces -- archive, delivery, publish |
| datacortex | Knowledge graph -- semantic search, graph statistics, link analysis |
| crm | Network intelligence -- track entities, relationships, interaction history |
| meetings | Meeting lifecycle -- standup generation, preparation, transcription processing |
| Email integration -- Gmail adapter, classification, processing |
See the Module Catalog for installation instructions and the full list of available modules.
| Resource | Description |
|---|---|
| Getting Started | Quick walkthrough for new users |
| Installation Guide | Complete setup instructions |
| Contributing | How to contribute |
| Module Catalog | Available modules and space templates |
| DIP Specifications | System design documents |
| Agent Registry | All registered agents |
| Command Registry | All registered commands |
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.
MIT License -- see LICENSE for details.
Datacore is built by Datacore. The AI system that bootstraps itself into existence.