Oculi Launches Security Layer for AI Coding Agents with Runtime Policy Enforcement

Summary

Oculi Security has announced a new security platform designed to provide runtime observability and policy enforcement for AI coding agents. The system intercepts every tool call made by AI agents — including shell commands, file operations, and Model Context Protocol (MCP) calls — and enforces customizable security policies before execution. The platform addresses growing concerns about AI agent security as developers increasingly rely on autonomous coding assistants like Claude Code, Cursor, and Windsurf.

The system works by hooking into IDEs and sitting between the development environment and AI agents, with minimal performance impact. Developers define security policies using simple YAML configuration files that specify which actions should be allowed, warned against, or denied. Key security features include shell command control to block dangerous operations like 'rm -rf', file access guards to protect sensitive files like .env configurations, and MCP tool gating to prevent unauthorized network calls. The platform provides comprehensive audit trails with real-time telemetry through the 'oculi tail' command and generates detailed reports on agent activity.

Oculi offers both local policy enforcement for individual developers and an optional enterprise gateway with centralized policy management, JWT authentication, and integration with Open Policy Agent (OPA). The platform supports major AI coding assistants including Claude Code (with first-class hook integration), Cursor, Windsurf, and any MCP-compatible client. The company is currently accepting early access signups through a waitlist as it prepares for commercial launch.

• The system integrates with major AI coding assistants including Claude Code, Cursor, Windsurf, and any Model Context Protocol-compatible client
• Enterprise features include centralized gateway with JWT authentication, OPA policy engine integration, rate limiting, and organization-wide policy management

Editorial Opinion

Oculi addresses a critical gap in the rapidly evolving AI coding agent ecosystem. As developers grant increasing autonomy to AI assistants, the lack of granular security controls has been a legitimate concern — particularly in enterprise environments with strict compliance requirements. The policy-as-code approach is particularly smart, allowing security policies to be versioned, reviewed, and deployed using familiar DevOps workflows. However, the platform's success will depend heavily on how it balances security with developer productivity; overly restrictive policies could frustrate developers and lead to policy bypass attempts, while too-lenient defaults might not catch genuine risks.