Unworldly is an advanced flight recorder designed for AI agents, providing comprehensive monitoring solutions for all activities performed by these agents within your system. With features such as real-time recording of file changes and shell commands, Unworldly functions like a DVR, ensuring accountability through tamper-proof, ISO 42001 and HIPAA-compliant audit trails. It captures critical actions, identifies potential security risks, and replays past sessions, making it an essential tool for developers and organizations utilizing AI technology.

Key Features

• Comprehensive Monitoring: Records every file modification and command executed by AI agents, allowing for precise tracking of actions.
• Real-Time Alerts: Flags potentially dangerous actions, such as network requests and credential access, with instant notifications.
• Tamper-Proof Logs: Utilizes a SHA-256 hash chain to secure recorded events, ensuring the integrity of your logs.
• Local-First Design: Operates entirely on local machines without the need for cloud storage or telemetry, enhancing privacy and security.
• Customizable Risk Patterns: Users can define allowlists and blocklists for command patterns, tailoring monitoring to specific needs.

Quick Start Guide

To begin using Unworldly, simply install via pip:

pip install unworldly-recorder
unworldly watch

This command initiates the recording process, capturing every action taken by the AI agent.

Risk Engine

Unworldly evaluates risks associated with actions by providing a risk score for each activity, categorizing them into safe, caution, and danger levels:

• Safe: Normal file edits and standard commands.
• Caution: Dependency changes and configuration file accesses.
• Danger: Accessing credentials, executing destructive commands, and making external network requests.

Compliance with ISO 42001

Unworldly adheres to the ISO 42001 standards, implementing key controls necessary for robust AI management systems. It ensures transparency, accountability, and the availability of observable behavior through detailed session replays and reports.

Agent Compatibility

Unworldly is agent-agnostic, compatible with leading AI platforms such as Claude Code, Cursor, Copilot, and more. This flexibility allows seamless integration and monitoring across diverse environments.

Who Can Benefit from Unworldly?

• Developers seeking insight into AI agents' activities.
• Security professionals auditing AI behavior in corporate settings.
• Compliance officers requiring trustworthy documentation for ISO 42001 and HIPAA standards.
• Open-source maintainers reviewing AI-generated code changes.

Unworldly equips users with the tools necessary to ensure that AI agents operate transparently and responsibly, protecting sensitive information while fostering a secure development environment.