create-forge-team

Build a team of AI agents for Claude Code. Describe your situation — Forge interviews you and generates personalized agents with skills that evolve over time.

Prerequisites

• Node.js 20+ (node --version)
• Claude Code installed and working
• Anthropic API key set as ANTHROPIC_API_KEY

Quick Start

# 1. Run the interactive team builder
npx create-forge-team

# 2. Use your agents in Claude Code
# Type /summon-{agent-name} in any Claude Code session

# 3. (Optional) Start the dashboard for a browser UI
forge bridge

The concierge interview takes 3-5 minutes. It creates 3-6 agents tailored to your goals, each with a system prompt, skills, and tool connections.

What You Get

• Personalized agents — each agent has a role, personality, and domain expertise
• Persistent memory — agents remember context across sessions via structured knowledge files
• Cross-agent knowledge — agents learn from each other's work through a shared activity feed
• Skills — slash commands (/summon-agent, /close-agent) for Claude Code integration
• MCP tool connections — connect calendar, email, databases, or any MCP server
• Background knowledge processing — transcripts are extracted into decisions, facts, and open threads

CLI Commands

| Command | Description | |---------|-------------| | forge create | Create a new agent team via conversation | | forge agent list | List all agents | | forge agent create | Add a single agent interactively | | forge bridge | Start the dashboard bridge server | | forge tools list | Show configured MCP tools | | forge memory setup | Configure the knowledge system | | forge memory status | Show knowledge queue and stats | | forge council "query" | Run a multi-model council query | | forge status | Show workspace overview |

Dashboard

The Forge Dashboard is a browser UI that connects to your local bridge server for chat, tool management, and session history.

1. Run forge bridge in your terminal
2. Copy the connection token displayed
3. Open the dashboard URL and paste the token

Memory System

Forge includes an optional memory agent that processes session transcripts into structured knowledge:

# One-time setup (needs an OpenRouter API key — free models available)
forge memory setup

# Check knowledge stats
forge memory status

After setup, session transcripts are automatically queued for processing. The memory agent extracts decisions, facts, open threads, and tool data into per-agent knowledge files that are injected into future sessions.

License

MIT