Model & CLI Commands

Commands for switching AI models, switching between CLI providers, configuring fallback chains, forwarding requests across providers, toggling Gemini thinking mode, and flushing model assignments. These commands let you dynamically reconfigure your AI backend mid-session without restarting.

Command Summary

/model [name]

Switch the active AI model within the current CLI provider. The model name must be a valid identifier recognized by the active provider. When called without arguments, displays the currently active model.

Arguments

Supported Models (examples)

Usage Examples

# Show current model
/model

# Switch to Claude Opus 4.8
/model claude-opus-4-8

# Switch to Sonnet for lighter tasks
/model claude-sonnet-4-6

# Switch to Gemini 2.5 Pro
/model gemini-3-flash-preview

/cli [name]

Switch the active CLI provider. This changes the entire backend — the API endpoint, authentication, available models, and tool capabilities all change with the provider. When called without arguments, displays the current CLI.

Arguments

Supported CLIs

Usage Examples

# Show current CLI
/cli

# Switch to Claude
/cli claude

# Switch to Gemini
/cli gemini

# Switch to Codex
/cli codex

# Switch to Grok
/cli grok

What Happens on Switch

When you switch CLIs, CLI-JAW:

• Gracefully disconnects from the current provider
• Initializes the new provider with its default model (unless a /flush pin exists)
• Carries over the conversation context where supported
• Updates the status bar and dashboard to reflect the new provider

/fallback [cli1 cli2 ...|off]

Configure a fallback chain of CLI providers. If the primary CLI fails (rate limit, network error, API outage), CLI-JAW automatically tries the next provider in the chain. This ensures uninterrupted operation even when a single provider is down.

Arguments

Usage Examples

# Set fallback: try Gemini first, then Codex, then Grok
/fallback gemini codex grok

# Set Claude as the only fallback
/fallback claude

# Disable fallback (only use the active CLI)
/fallback off

# Show current fallback chain
/fallback

How Fallback Works

1. A request is sent to the active CLI
2. If the active CLI returns an error (rate limit, timeout, 5xx), the request is retried on the first fallback CLI
3. If that also fails, it moves to the next CLI in the chain
4. If all fallbacks are exhausted, the error is surfaced to the user

/forward [on|off]

Enable or disable request forwarding to another provider. When forwarding is on, certain requests can be routed to a secondary provider for specialized handling — for example, sending a code generation request to Claude while using Gemini as the primary for conversation, or routing image analysis to a vision-capable model.

Arguments

Usage Examples

# Enable forwarding
/forward on

# Disable forwarding
/forward off

# Show current forwarding status
/forward

When to Use Forwarding

/thought [on|off]

Toggle Gemini's thinking (extended reasoning) mode. When enabled, Gemini uses additional compute to reason step-by-step before responding, similar to Claude's extended thinking. This is useful for complex tasks like multi-step math, logic puzzles, or architectural decisions. Only applies when Gemini is the active CLI.

Arguments

Usage Examples

# Enable Gemini thinking
/thought on

# Disable thinking (faster responses, lower cost)
/thought off

# Show current thinking status
/thought

When to Use Thinking Mode

• Enable for: complex reasoning, multi-step problems, architecture decisions, debugging intricate issues, math/logic
• Disable for: simple questions, quick edits, file operations, routine tasks (saves time and tokens)

/flush [cli] [model] | off

Pin a specific model to a CLI provider, or clear an existing pin. When a flush pin is set, switching to that CLI automatically loads the pinned model instead of the provider's default. Use off to clear all pins or a specific pin.

Arguments

Usage Examples

# Pin Claude to always use Opus 4.8
/flush claude claude-opus-4-8

# Pin Gemini to always use 2.5 Pro
/flush gemini gemini-3-flash-preview

# Clear the Claude pin (reverts to default model on next switch)
/flush claude off

# Clear all pins
/flush off

# Show current flush pins
/flush

How Flush Pins Work

Flush pins are stored in the session and persist until explicitly cleared or the session ends. They override the provider's default model selection:

1. You set /flush claude claude-opus-4-8
2. Later, you switch to Gemini with /cli gemini
3. When you switch back with /cli claude, it automatically uses claude-opus-4-8.

Natural Language Mapping

CLI-JAW's command parser understands natural language in both Korean and English. Here are common phrases and their command mappings:

Common Workflows

Cost-Optimized Setup

Use a lightweight model for routine tasks and pin a powerful model for when you need it:

# Default to Opus 4.8 for strongest Claude work
/model claude-opus-4-8

# Pin Opus for when you switch to heavy tasks
/flush claude claude-opus-4-8

# Set Gemini Flash as fallback for rate limits
/fallback gemini

Multi-Provider Resilience

Ensure uninterrupted service by chaining fallbacks:

# Primary: Claude, Fallback: Gemini -> Grok -> Codex
/cli claude
/fallback gemini grok codex

# Pin each provider to its best model
/flush claude claude-opus-4-8
/flush gemini gemini-3-flash-preview
/flush grok grok-3

Gemini Deep Reasoning Session

Switch to Gemini with thinking enabled for complex analysis:

# Switch to Gemini Pro with thinking
/cli gemini
/model gemini-3-flash-preview
/thought on

# When done, switch back to Claude
/cli claude
/thought off