Upgrade to Pro Builder (L2)

L1 gives you governance inside Claude Code. L2 adds multi-tool orchestration: file locking, task planning, knowledge capture, and a TUI dashboard that coordinates Claude and Codex (as a task runner) as a team. Antigravity CLI (agy) support is in development (build B1, target 2026-10-27); today the orchestrator launches agy as a task runner.

When You Need This

You need L2 when you're running multiple AI tools on the same codebase and hitting these problems:

• File conflicts — Claude refactors a module while Codex updates tests against the old interface
• Lost context — decisions made in one tool aren't visible to the other
• No coordination — each tool works on whatever seems important, with no shared plan
• Session amnesia — patterns and learnings evaporate between sessions

If you're using Claude Code solo, L1 is sufficient. L2 is for multi-tool workflows.

Install

curl -fsSL https://forge.nxtg.ai/install.sh | sh

Single binary. 6.4 MB. No runtime dependencies. Works on Linux and macOS.

Or via Homebrew:

brew install nxtg-ai/tap/forge

Quick Demo

Four commands to see what the orchestrator does:

# 1. Initialize Forge in your project
forge init
# Creates .forge/ directory with state, event log, and knowledge base

# 2. Configure the AI brain
forge config brain openai
# Sets up GPT-4o for task planning (or use 'rule-based' for free heuristics)

# 3. Generate tasks from your project
forge plan --generate
# Reads your SPEC.md or README, decomposes into dependency-aware tasks

# 4. See the task board
forge status
# Shows tasks, assignments, dependencies, and project health

The Dashboard

forge dashboard --pty

This launches a TUI dashboard with:

• Task board — all tasks with status, assignment, dependencies
• Agent panes: live interactive terminals for each AI tool (Claude, Codex; today, Antigravity CLI (agy) support is in development; build B1, target 2026-10-27)
• Event log — real-time audit trail of every action
• Health status — project health score in the header

Each agent runs in its own PTY session. You see their actual TUI, not piped text. Press Tab to switch between panes, i to attach to a pane (type directly into the agent), and Esc to detach.

What You Get Over L1

How It Connects to L1

When you install the orchestrator binary, the L1 plugin auto-detects it. The plugin's .mcp.json registers both MCP servers:

1. governance-mcp (Node.js, 8 tools) — always available
2. orchestrator-mcp (Rust, 12 tools): available when forge is in PATH

No manual configuration. The plugin starts calling orchestrator tools (forge_get_tasks, forge_claim_task, forge_complete_task) automatically. This is the "Lego Snap" — two independent pieces that click together via MCP.

Key Commands

forge init                     # Initialize Forge in a project
forge config brain openai      # Configure AI brain (or 'rule-based' for free)
forge plan --generate          # Generate task plan from spec
forge status                   # Task board + health summary
forge dashboard                # TUI dashboard (text output)
forge dashboard --pty          # TUI dashboard (interactive agent panes)
forge run                      # Headless autonomous execution
forge mcp                      # Start MCP server (stdio)

File Locking in Practice

When the orchestrator assigns a task to an agent, it locks the task's target files. If another agent tries to claim a task that touches those same files, the claim is rejected until the lock is released.

Agent A claims T-001 (editing src/auth.rs, src/auth_test.rs)
  → Files locked to Agent A

Agent B claims T-002 (editing src/auth.rs, src/db.rs)
  → REJECTED: src/auth.rs is locked by Agent A

Agent A completes T-001
  → Files unlocked

Agent B claims T-002
  → Success: all files available

No configuration. No manual lock management. The orchestrator handles it.

When to Upgrade to L3

L2 gives you everything you need for multi-tool orchestration from the terminal.

L3 (Ship Lord) adds a visual web dashboard with real-time graphs, the Infinity Terminal (sessions that survive browser close and network drops), and full mission control. You need it when:

• You want visual oversight of agent activity across projects
• You need a terminal that persists across browser sessions
• You're managing a team and want a shared dashboard

Next Steps

• CLI Commands Reference — All orchestrator commands in detail
• MCP Tools Reference: 20 tools across both MCP servers
• AI Brain Configuration — Rule-based vs. OpenAI, API keys