Google Antigravity 2.0 (CRAZY Updates & FULLY FREE): These NEW Updates are CRAZY!

Key Concepts

• Anti-Gravity: A Google-developed agentic IDE (Integrated Development Environment) designed for autonomous coding and task execution.
• MCP (Model Context Protocol): A standard for connecting AI assistants to systems, data, and tools.
• Agentic IDE: An IDE that allows AI agents to autonomously perform tasks like coding, terminal execution, and browser interaction.
• Strict Mode: A security feature requiring human review for autonomous agent actions to prevent unauthorized exploits.
• Sandboxing: A security mechanism for running code or browser tasks in an isolated environment to protect the host system.
• Rate Limits: Transparent quotas for AI model usage, now visible to users to help with project planning.

---

1. Overview of Anti-Gravity Updates

Contrary to the perception that Google’s "Anti-Gravity" product has been abandoned, the platform is undergoing significant "hardening." While it lacks flashy marketing, recent updates focus on core infrastructure, stability, and security. The platform is increasingly integrated into the broader Google ecosystem, including Google AI Studio and Google Cloud.

2. Settings and Configuration

The settings panel has been overhauled to provide granular control over agent behavior:

• Agent Settings: Includes "Strict Mode" (enforces human review for actions), terminal command permissions, and shell integration settings.
• Model Transparency: Users can now view specific rate limits for each model, which refresh every four hours. The platform allows for the addition of AI credits to extend usage beyond free-tier limits.
• Customizations: Supports the integration of MCP servers and "Skills." Users can point the agent to specific directories to automatically discover and utilize custom skills.
• Browser Control: Agents can be granted browser access to interact with URLs. Security policies for JavaScript execution include:
  • Disabled: No JS execution.
  • Request Review: Agent prompts for permission.
  • Always Proceed: Maximum autonomy (highest security risk).

3. Agent Management and Workflow

The Agent Manager has been redesigned to improve project navigation and environment management.

• Environment Support: Currently supports local environments, with potential for remote configurations in the future.
• Linux Support: Recent updates have introduced custom provider and model support for Linux users, allowing for the configuration of local models like Llama.
• Conversation History: A new "Changes Pane" allows users to view a diff of all modifications made during a session, enabling users to review and comment on specific agent actions for iterative refinement.

4. Security and Reliability Improvements

A significant portion of recent development has focused on "core plumbing" to make the tool viable for professional projects:

• Terminal Sandboxing: Implemented for both macOS and Linux to ensure agent-run commands do not compromise the host machine.
• Bug Fixes: Addressed critical issues such as the accidental deletion of files during reverts and failures in loading MCP servers.
• Unified Permission System: Introduced to create clear boundaries between what an agent can do autonomously versus what requires human oversight.

5. Development Timeline and Evolution

The platform’s evolution is documented through a series of incremental updates:

• February: Introduced the model quota screen, terminal integration, and the transition from "Secure Mode" to "Strict Mode."
• April (7th & 16th): Focused on the unified permission system, bug fixes for MCP servers, and workspace-specific settings.

6. Synthesis and Conclusion

Anti-Gravity is currently in a phase of "quiet hardening." While it may not yet match the innovative pace of competitors like Verdant—which the author notes is currently more feature-rich—Anti-Gravity has successfully transitioned from a "fun demo" to a tool capable of handling real-world development tasks.

Main Takeaway: The product is not abandoned; rather, it is being systematically improved to address the "boring" but essential requirements of an agentic IDE: safety, reliability, and transparency. Users who previously dismissed the tool are encouraged to revisit it, as the current build is significantly more stable and secure than its initial launch version.