Endpoint agent that discovers, scores, and enforces policy on AI agents, MCP servers, and skills

Optimus Labs was founded by Nipun Gupta, a security executive with more than fifteen years across large enterprises and startups, and its team draws on people from Carnegie Mellon, Mindgard, and GreyNoise. Its investors include Arka Benhamou Global Ventures, Executive Venture Fund, and an a16z scout fund. The company's core argument is architectural: the most consequential actions an AI coding agent takes, reading a private repository, executing a shell command, writing to a local file, never cross the network, so a proxy or gateway sitting at the network edge cannot see them. The control point has to be the machine.

A lightweight endpoint agent deploys through existing MDM infrastructure and builds an inventory of every agent, MCP server, and skill present, including installations nobody approved. Each is assessed against its permissions, memory configuration, and data access patterns, with scoring aligned to the OWASP Top 10 for agentic systems and specific attention to what the company calls the lethal trifecta, the combination of untrusted input, access to sensitive data, and the ability to act externally. At runtime it intercepts actions before they execute, allowing a policy to alert, require a justification from the user, or block outright. Coverage spans coding agents such as Cursor, Claude Code, GitHub Copilot, and Devin, desktop assistants, and custom internal frameworks, with anonymized logs of intent, tool calls, and data flows retained for ninety days.