gitmyhub

hive

HTML ★ 19 updated 16d ago

Run your company with AI agents. Open source harness for operating a full company with specialized Claude Code squads.

An open-source framework that gives you a team of 11 named AI business agents inside Claude Code, each with a defined role, persistent memory, and 50 built-in commands covering sales, marketing, finance, dev, and ops.

PythonClaude Codesetup: moderatecomplexity 3/5

HIVE (Harness for Intelligent Virtual Enterprises) is an open-source framework for running a company using AI agents inside Claude Code, the command-line version of the Claude AI assistant. The idea is that instead of asking a single AI to handle everything, you maintain a team of specialized agents, each focused on one part of the business.

The system comes with 11 pre-built "squads": an orchestrator called Stamper who acts as Chief of Staff, plus named personas for Sales (Victor), Customer Success (Leah), Marketing (Maya), Product (Owen), Finance (Clara), Development (Ethan), Infrastructure (Dean), Operations (Harper), Quality (Nora), and Intelligence (Rex). Each squad has its own instructions defining its role and boundaries, its own memory that persists between sessions, and its own set of commands for the tasks it handles. You open a squad's folder in Claude Code and the AI immediately knows who it is and what it is responsible for.

Memory works through a three-tier system. The current status of each squad (what is happening right now, what is in progress, and what is queued up) is tracked in a STATE.md file. The orchestrator can pull a summary from all squads at once with a single command.

The framework includes 50 built-in commands covering things like onboarding a new company, opening a squad, logging incidents, creating issues, writing a technical plan, running a sales call using a structured selling framework, debugging a problem, and conducting a competitive analysis. It also has five automation hooks that handle session management automatically: creating a branch when you start working, tracking changes, and committing everything to the main branch when you stop, without any manual steps.

Background tasks (things like data pulls or status updates that do not require AI) are written as plain Python scripts that run on a schedule through the operating system's built-in job scheduler. The project is MIT licensed and built specifically to run inside Claude Code.

Where it fits