Your AI team,
ready to work

• 1

  Clone the repo

  Replace my-company with your company name. Fork if you want to track upstream changes.

  macOS / Linux

  Windows

  git clone https://github.com/felipeluissalgueiro/hive.git my-company
  cd my-company

  git clone https://github.com/felipeluissalgueiro/hive.git my-company
  cd my-company

• 2

  Open in Claude Code

  Open the my-company/ folder in Claude Code (CLI, desktop, or IDE extension). Claude Code reads CLAUDE.md and loads the Stamper orchestrator automatically.

  macOS / Linux

  Windows

  # CLI
  claude
  
  # Or open my-company/ in Claude Code desktop app

  # CLI
  claude
  
  # Or open my-company/ in Claude Code desktop app

• 3

  Run setup

  The setup skill (~5–10 min) asks about your company and configures each squad with relevant context.

  /hive-setup
  
  # What it does:
  # · Asks company name, industry, stage
  # · Asks your working style and preferences
  # · Lets you pick which squads to activate
  # · Fills foundation docs with your context
  # · Writes everything to the right files

# Open a squad (loads persona + STATE + foundation)
/open-squad commercial

# Work with Victor (Head of Sales)...

# Close the squad (updates STATE, propagates L1 to orchestrator)
/close-squad commercial

# See what's happening across all squads
/status

# Search past incidents before a sensitive operation
python _core/lookup.py "deploy vercel"

Core workflow

When you open a squad folder, Claude Code loads both the root CLAUDE.md (Stamper) and the squad's CLAUDE.md (specialist). You get two personas in one session.

Every squad maintains a memory/STATE.md with three layers. L1 is designed to be machine-readable for aggregation by /status.

# squads/commercial/memory/STATE.md

[L1]
Victor (Commercial) — 3 hot leads, 1 proposal pending review, pipeline $42k.
Last updated: 2026-06-02 14:30

[L2]
In progress:
- Proposal for Acme Corp — send by EOW
- Follow-up call with TechStartup scheduled Friday
- Updating ICP profile with Q2 learnings

[L3]
Backlog:
- Cold email sequence for segment B
- Competitor analysis for enterprise tier

Memory files per squad

Knowledge lookup

Before sensitive operations (deploy, DNS change, integration), search past incidents and sessions:

# Search across incidents + sessions + memory
python _core/lookup.py "deploy vercel"

# Filter by source
python _core/lookup.py "ghl webhook" --source incidents
python _core/lookup.py "auth" --source sessions --since 2026-01-01

# JSON output for scripts
python _core/lookup.py "supabase" --json --top 5

Workers live in workers/ and run on a schedule via _core/harness.sh. They are pure Python — no LLM calls. They read data sources and write to STATE.md.

Design rules

• Deterministic — same input, same output. No side effects.
• Write only to STATE.md — never CLAUDE.md, skills, or foundation docs.
• Idempotent — safe to run multiple times.
• Fail loudly — log errors, don't swallow exceptions silently.

# _core/harness.sh — add workers here
run_worker "workers/daily-revenue.py" "09:00"
run_worker "workers/health-scores.py" "every:1h"
run_worker "workers/competitor-scan.py" "every:24h"

Example workers

Foundation

Dev workflow

View full skills catalog (50 skills) →

HIVE ships 7 skills that query multiple AI models via OpenRouter. Useful for reducing model bias, validating decisions, and finding the cheapest model for recurring tasks.

Setup

1. Get an API key at openrouter.ai
2. Set OPENROUTER_API_KEY=sk-or-... in your environment
3. Run the install script: bash _core/mcp/install.sh

# macOS / Linux
export OPENROUTER_API_KEY=sk-or-your-key-here
bash _core/mcp/install.sh

# Windows (PowerShell)
$env:OPENROUTER_API_KEY = "sk-or-your-key-here"
bash _core/mcp/install.sh

Available skills

Supported models (aliases)

View full models map (30+ models) → · MCP setup guide →