Agent Modes

Agent Modes

Lantern supports multiple modes for different coding scenarios. Choose the mode that best fits your workflow.

Mode 1: Default Squad (Headed Orchestrator + Headless Specialists)

The most common mode: a conductor role managing a fleet of specialized agents.

Command

lantern startwork myproject
# or explicitly:
lantern startwork myproject 1 --agent claude

What Launches

• Orchestrator pane (headed): The conductor role, running claude agent CLI interactively
• 8 specialist panes (headless ACP): AI, Data Architecture, Security, Operations, Platform, UI, Documentation, QA
• 9 total panes in a 4×2+1 grid layout, each with a unique color, role label, and dedicated git worktree

When to Use

Multi-agent projects needing coordination across diverse specialties (e.g., building a full-stack feature with separate code review, data, security, and UI concerns).

Example Workflow

lantern startwork my-app 1 --agent claude
# Terminal opens with 9 panes, orchestrator pane is interactive
# Type instructions in the orchestrator; team members execute tasks in parallel

Mode 2: Solo Goose (Single Focused Agent)

A single agent with no orchestrator overhead, for focused work.

Command

lantern startwork myproject --agent goose

What Launches

• Single Goose pane (headed): Full-featured native Goose session
• 1 worktree for the session
• No specialist fleet — just you and Goose

When to Use

Single-agent focused fixes, prototyping, or when you want Goose's native features (readline, keyring, full interactive capabilities) without the multi-agent overhead.

Example Workflow

lantern startwork my-app --agent goose
# Terminal opens with single Goose pane
goose> help

Mode 3: Squad with Different Agent CLI

Use a different agent CLI instead of claude for all roles.

Commands

lantern startwork myproject 2 --agent codex
lantern startwork myproject 3 --agent kimi
lantern startwork myproject 4 --agent agy
lantern startwork myproject 5 --agent gemini

What Launches

Same 9-pane grid layout as Mode 1, but with the specified agent (codex, kimi, agy, or gemini instead of claude).

When to Use

Testing different AI providers, or when your project requires a specific agent.

Supported Agents

• claude (default)
• codex (OpenAI)
• kimi (Kimi/MoonshotAI)
• gemini (Google)
• agy (AGI Lab)
• goose (solo mode only)

Example Workflow

lantern startwork my-app 1 --agent kimi
lantern startwork my-app 2 --agent codex
# Run two squads in parallel with different agents
lantern status  # see both sessions

Switching Between Modes

You can run multiple squads simultaneously with different modes and agents:

lantern startwork project-a 1                    # Default squad, claude
lantern startwork project-a 2 --agent goose      # Solo goose
lantern startwork project-b 1 --agent codex      # Default squad, codex

lantern status  # shows all 3 sessions active

Each squad is independent and can be stopped separately:

lantern stopwork project-a-1
lantern stopwork project-a-2
lantern stopwork project-b-1