Prompts for confirmation before modifying files or running commands
Accept Edits
--mode accept-edits
Automatically applies edits and writes files without prompting
Plan Mode
--mode plan
Formulates step-by-step implementation plans before touching code
Sandbox
--sandbox
Restricts shell execution inside an isolated terminal sandbox
Unrestricted
--dangerously-skip-permissions
Auto-approves all tool permission prompts
Key Concepts
Agentic Loop: Dynamically inspects files, runs shell commands, views diffs, and self-corrects based on tool results.
Progressive Disclosure: Skills and contextual rules are loaded on demand rather than cluttering token context.
Asynchronous Tasks: Background commands and multi-agent runs execute concurrently without blocking turns.
CLI Options
Core Flags
Flag
Short
Description
--print <prompt>
-p
Run prompt non-interactively and print response
--prompt-interactive <prompt>
-i
Run initial prompt and remain interactive
--continue
-c
Resume the most recent conversation
--conversation <id>
Resume specific conversation by ID
--model <model>
Specify model for the session
--effort <level>
Set reasoning effort (low, medium, high)
--mode <mode>
Agent mode (accept-edits, plan)
--sandbox
Enable terminal sandbox restrictions
--dangerously-skip-permissions
Auto-approve all tool permissions
--add-dir <path>
Add directory to workspace (repeatable)
--project <id>
Google Cloud / Vertex project ID
--new-project
Create a new project for this session
--output-format <format>
Output format (text, json, stream-json)
--input-format <format>
Input format (text, stream-json)
--json-schema <schema>
Enforce structured JSON schema for outputs
--print-timeout <duration>
Timeout limit for print mode (default 0s)
--log-file <path>
Override CLI log file path
--remote-control
Connect to remote control background daemon
--disable-slash-commands
Disable slash command expansion in print mode
Subcommands
# Manage MCP servers
$ agy mcp [list|add|remove|enable|disable]# Manage plugins
$ agy plugin [list|install|uninstall|enable|disable|import|validate]# List supported AI models
$ agy models
# List available agents
$ agy agents
# Manage remote control background daemon
$ agy remote-control [start|status|stop]# Serve microphone input to a CLI on another host
$ agy mic-serve
# Update agy CLI
$ agy update
Headless & CI/CD Formats
# Plain text output
$ agy -p "check repo for vulnerabilities" --output-format text
# JSON output for tooling pipelines
$ agy -p "extract exported API endpoints" --output-format json
# NDJSON streaming for real-time consumers
$ agy -p "generate OpenAPI spec" --output-format stream-json
# Enforce structured output via JSON Schema
$ agy -p "list todos" --json-schema schema.json --output-format json
Model Selection & Reasoning
Models & Reasoning Effort
# Check all available models
$ agy models
# Run with Gemini 3.8 Flash with high reasoning (fast & deep)
$ agy --model gemini-3.8-flash-high
# Run with Gemini 3.1 Pro for complex architectures
$ agy --model gemini-3.1-pro-high
# Run with Claude Sonnet 4.6 (Thinking)
$ agy --model claude-sonnet-4-6
# Specify reasoning effort level
$ agy --effort high
$ agy --effort medium
$ agy --effort low
Model Reference
Model Identifier
Display Name
Recommended Use
gemini-3.8-flash-high
Gemini 3.8 Flash (High)
Daily coding, rapid agent turns, default
gemini-3.8-flash-medium
Gemini 3.8 Flash (Medium)
Balanced speed and thinking tokens
gemini-3.8-flash-low
Gemini 3.8 Flash (Low)
Fastest latency, lightweight queries
gemini-3.7-flash-high
Gemini 3.7 Flash (High)
Previous generation high reasoning
gemini-3.1-pro-high
Gemini 3.1 Pro (High)
Massive refactors, cross-repo design
claude-sonnet-4-6
Claude Sonnet 4.6
Thinking-enabled coding assistant
claude-opus-4-6-thinking
Claude Opus 4.6
Deep complex multi-step reasoning
gpt-oss-120b-medium
GPT-OSS 120B
Open-weights reasoning tier
Interactive Slash Commands
Slash Commands /
Slash Command
Purpose
/help
Show available slash commands and keybindings
/plan
Formulate a multi-step design & plan before execution
/goal
Autonomous mode: runs until goal is completed
/boost
Deep thinking, multi-angle exploration, rigorous checks
/browser
Web automation, page inspection, and live testing
/grill-me
Interactive interview to settle design decisions
/schedule
Set a one-off timer or recurring cron schedule
/teamwork-preview
Autonomous multi-agent collaboration
/learn
Persist developer preferences and lessons
/exit
Exit the CLI session
/quit
Alias for /exit
Goal & Boost Modes
# Inside interactive agy:
> /goal Implement end-to-end tests for billing API and verify they pass
# Boost mode for mission-critical changes:
> /boost Review security implications of session token rotation
Interactive Interviews
# Challenge your assumptions before writing code
> /grill-me I want to migrate our database from MongoDB to Postgres
Keyboard Shortcuts
General Controls
Keybinding
Action
Ctrl+D Ctrl+D
Exit agy CLI session
Ctrl+C
Cancel currently running agent turn or streaming response
Enter
Submit prompt / send message
Shift+Enter
Insert a newline into multiline input
Alt+Enter
Insert a newline into multiline input
Input Navigation
Keybinding
Action
Up / Down
Browse prompt history
Ctrl+A
Move cursor to beginning of line
Ctrl+E
Move cursor to end of line
Ctrl+U
Delete from cursor to start of line
Ctrl+K
Delete from cursor to end of line
Ctrl+L
Clear terminal screen
Agent Control
Action
How
Accept Edit
Press Enter or click Approve on tool prompt
Reject Edit
Press Esc or type feedback
Interrupt
Ctrl+C halts tool call immediately
Subagents & Delegation
Subagent Architecture
Antigravity CLI can invoke specialized subagents that execute concurrently with their own contexts and workspaces: