Open Source

Crow

A modular, agentic framework for assistance, research, projects, and home. Build and run your own AI agents, keep your data on hardware you control, and work with any AI assistant through the open Model Context Protocol. Your AI, your devices, your data.


Works With

Crow connects to any AI client that supports the Model Context Protocol. One platform, every assistant.

Claude ChatGPT Gemini Grok Cursor Windsurf Cline Claude Code

Key Features

Everything you need for persistent, cross-platform AI work.

Build Your Own Agents

The Bot Builder lets you compose an agent from a persona, skills, tools, and permissions, then run it over email, Discord, or Meta glasses. Each agent is scoped to the tools you grant it, and opt-in self-authoring stays behind an operator-approval gate.

Persistent Memory

Your AI assistant remembers across sessions. Full-text search, categorized recall, and context that persists no matter which platform you use.

Research Pipeline

Save sources, generate APA citations, build bibliographies, and manage research projects. Academic-grade source management built into your AI workflow.

P2P Encrypted Sharing

Share memories, research, and messages directly between Crow users. End-to-end encrypted via NaCl, no central server, no accounts, no metadata leaks.

Cross-Platform

The same memory and research data available whether you are in Claude, ChatGPT, Gemini, Cursor, or any MCP-compatible client.

20+ Integrations

Connect GitHub, Slack, Notion, Gmail, Trello, Discord, and more. Crow proxies external services through a single authenticated gateway, and every integration's tools become available to your agents.

Custom Skills

Behavioral prompts that teach your AI new workflows: research methodology, code review, writing, project management. No code required to create your own.

Your Data Stays Yours

Crow stores your data on infrastructure you control. Pair it with a local model and nothing ever leaves your network. Connect a cloud assistant and only what you send that provider goes out, on your terms.


Crow for Your Field

The same framework points in very specific directions depending on what you do. Each cut below describes capabilities Crow ships today.

Education

A private AI workspace for students, teachers, and researchers. Crow remembers your work across sessions, manages sources with real citations and bibliographies, and connects to course tools and public datasets. Built for FERPA-sensitive settings, your data stays on infrastructure you control.

Law

A self-hosted agentic assistant for legal work: matter organization, document research, and citation tracking, with privileged material kept on hardware you own. No third-party AI vendor sees the file.

Home Entertainment

Turn a home server into an agentic hub for your media. Voice-controlled music and video across your devices, even your glasses, drawing on your own library. Your collection, your rules, nothing tracking your habits.

Enterprise

A self-hosted agentic platform for teams that handle regulated data. Build internal assistants, connect the tools your team already uses, and keep every byte inside your own network, with operator-approval gates on anything an agent does.


Open Research Infrastructure

Crow is not just an AI tool. It is infrastructure for democratizing access to education data.

  1. Access the database

    Maestro Press is building a statewide education data platform covering every Texas district and campus: funding, accountability, demographics, and public records. Independent researchers and advocates can connect to this database through Crow's MCP protocol from their preferred AI assistant, without needing to build their own data pipelines.

  2. Perform independent research

    Crow's research pipeline (persistent memory, auto-citations, source management) lets users conduct their own analyses on the Maestro Press dataset. A parent advocate or policy researcher can ask questions of the data, build bibliographies, and develop findings with full APA citation support built in.

  3. Share results peer-to-peer

    Crow's encrypted P2P sharing lets researchers share findings, memories, and data directly with each other. No central server, no accounts, no metadata leaks. A researcher in El Paso can share their analysis with a colleague in Houston without either trusting a third-party platform.

  4. Contribute back

    Users can contribute their own research, cleaned datasets, and findings back to the Maestro Press database through Crow, enriching the shared resource for the entire community. The open protocol means contributions flow both ways.


Quick Start

Desktop (Claude Desktop)

Clone and run the setup script. Copy the generated config into Claude Desktop settings.

git clone https://github.com/kh0pper/crow.git && cd crow npm run setup npm run desktop-config
Desktop setup guide →

Developer (Claude Code)

Clone and setup. Claude Code auto-detects the MCP configuration and project context.

cd crow npm run setup claude
Claude Code guide →

Developer Program

Crow is open to contributions. Build integrations, skills, tools, and deployment bundles for the ecosystem.

MCP Integrations

Connect new services: Linear, Jira, Todoist, and any API with an MCP adapter.

Skills

Write behavioral prompts that teach the AI new workflows. No code required.

Core Tools

Add MCP tools to crow-memory, crow-research, or crow-sharing.

Self-Hosted Bundles

Create Docker Compose configs for specific use cases and deployment targets.


Crow is free, open source, and ready to use.