Endor Labs puts security checks in Google Antigravity agents

Endor Labs makes tools that find bugs in software code. The company is a partner at the start of the new Plugin marketplace. It is in Google Antigravity, the Google platform where AI agents plan and write code. The plugin installs with one click. It checks packages and code while the agent writes. The company says only about 10% of AI code is correct and safe.

How it works

The plugin gives the agent security data while it works. Before the agent adds a package, it checks the package for known bugs and malicious code. It also examines the code for secrets. The plugin has 4 agents:

• One chooses the safest upgrade for a weak package.
• One writes patches for each scan result that is correct.
• One tells if a new package is safe to add.
• One explains a known bug with a short text.

The numbers

Endor Labs says that in its benchmark, its agents do security tasks 2.8 times as fast as a general agent. They use 12 times less tokens. A token is a unit of text that an AI model reads or writes. The company gave no test data and no external test.

What is not known

The company gave no price. It did not show if the checks find more bugs in the code of users.

Insight: Google builds a marketplace where security tools compete

In May 2026, Google was a partner for Endor Labs tools that control agents. At this time Endor Labs is a partner for the Google marketplace, which packs agents and MCP servers into bundles. Descope sells an approval step for each agent. This reads as a race to be the primary door for agent tools. The owner of the door sets the rules. Tell your suppliers which marketplace they must use.