Autonomous AI development that builds features, fixes bugs, and ships code—with minimal guidance.
$ Enable Agent Mode
✓ Agent Mode activated
$ Build authentication system with JWT
Planning... 8 steps identified
Step 1/8: Creating User model...
Step 2/8: Generating auth routes...
Step 3/8: Adding middleware...
✓ Authentication system complete!
Agent Mode completes multi-step tasks without constant guidance. Describe your goal and watch AI break it down into steps, execute each one, and iterate until done.
Create, modify, rename, and delete files across your entire project. Agent Mode understands your codebase structure and makes coordinated changes.
Run build scripts, execute tests, install dependencies, and start development servers. All from natural language instructions.
When something goes wrong, Agent Mode identifies the issue and attempts to fix it automatically. Smart error handling and retry logic built-in.
Connect to databases, GitHub, Slack, AWS, and more through MCP servers. Agent Mode can query data, create PRs, and interact with external services.
Critical operations require your confirmation. You maintain control while AI handles the heavy lifting. Review and approve before destructive actions.
Tell the agent what you want to build or fix in natural language
Agent analyzes your codebase and creates a step-by-step plan
Agent writes code, runs commands, and modifies files
Agent tests its work and iterates until successful
Cmd/Ctrl + Shift + AToggle Agent ModeEscapeCancel current operationEnterConfirm step (when prompted)Cmd/Ctrl + ZUndo last AI change“Build a user authentication system with JWT tokens, login/register endpoints, and password reset functionality.”
Creates models, routes, middleware, tests, and documentation in one session.
“Debug why the checkout form is not submitting. Users report error on clicking submit button.”
Analyzes code, identifies root cause, implements fix, and adds regression test.
“Refactor the UserService class to use dependency injection and add unit tests.”
Restructures code, updates dependencies, maintains functionality, adds comprehensive tests.
“Create a REST API for managing blog posts with CRUD operations, pagination, and filtering.”
Generates endpoints, validation, database queries, and OpenAPI documentation.
“Add a new comments table with foreign keys to users and posts, including indexes.”
Creates migration, generates rollback script, updates models, tests migration locally.
“Write comprehensive tests for the payment processing module.”
Creates unit tests, integration tests, mocks external services, achieves 90%+ coverage.
Destructive actions require your explicit approval
Control what operations Agent Mode can perform
Easily revert AI changes with Git integration
Stop execution instantly with Escape key
Connect Agent Mode to external services through MCP servers for even more capabilities:
Create PRs, manage issues
SupabaseQuery and modify databases
SlackSend notifications
AWSDeploy and manage cloud
Customize Agent Mode behavior with configuration files:
.antigravityProject-specific agent rules
GEMINI.mdGlobal agent preferences
Rules GuideComplete configuration reference
Tip: Define which operations require confirmation in your .antigravity file to balance autonomy with control.
Read our comprehensive guide with advanced techniques, prompt templates, and best practices.