ClawVault provides visual monitoring and atomic control for AI agents, enabling users to secure AI interactions, manage costs, and gain real-time visibility into asset access. It targets developers and organizations seeking to enforce granular security policies and detect threats within their AI ecosystems.

How It Works

ClawVault operates via a transparent proxy gateway that intercepts AI tool traffic to external APIs. A detection engine analyzes this traffic for sensitive data, prompt injections, and dangerous commands. Users define fine-grained access using composable "atomic capabilities" and can leverage generative policies to automatically create security rules via natural language. The system provides real-time monitoring, audit trails, and token budget tracking.

Quick Start & Requirements

• Install: pip install -e .
• Run: claw-vault start (proxy + dashboard), claw-vault scan "...", claw-vault demo.
• Deploy: Use ./scripts/deploy.sh <ip> [user] for server deployment, followed by ./scripts/setup.sh and ./scripts/start.sh on the server.
• Prerequisites: Python 3.x environment.
• Documentation: Full documentation index available in the doc/ directory.

Highlighted Details

• Comprehensive Sensitive Data Detection (15+ patterns: API keys, PII, credit cards).
• Robust Prompt Injection Defense against role hijacking and data exfiltration.
• Dangerous Command Guard intercepts high-risk shell commands like rm -rf or curl|bash.
• Granular Token Budget Control with daily/monthly limits and cost alerts.
• Real-time Dashboard for monitoring, configuration, and quick testing.
• Generative Policies enable natural language creation of adaptive security rules.

Maintenance & Community

Community support is available via GitHub Issues for bug reports and feature requests, and a dedicated channel for Security Issues.

Licensing & Compatibility

Licensed under the MIT license, which is permissive for commercial use and integration into closed-source projects.

Limitations & Caveats

File-side monitoring and generative policy orchestration are currently in progress. Agent-level atomic control capabilities are expanding.