Open Source • Apache 2.0

Security Layer for
AI Agents

Protect agentic workflows with command risk analysis, CVE scanning, sandboxed execution, security dashboard, and session auditing — works with every AI coding assistant on macOS, Linux, and Windows.

200+ risk patterns
CVE scanning
Agent-native
Cross-platform

Open source security tool • Built for developers

Get Started in 30 Seconds

Four simple steps to secure your entire development workflow

Step 1

Install

One-line installation for macOS, Linux, and Windows. Supports automatic updates and multiple installation methods.

Step 1: Install
# macOS & Linux (recommended)
curl -fsSL https://raw.githubusercontent.com/xadnavyaai/vectra-guard/main/install.sh | bash

# Windows (PowerShell, no admin required)
irm https://raw.githubusercontent.com/xadnavyaai/vectra-guard/main/scripts/install-windows.ps1 | iex

# Or download latest release (v0.6.0)
# See https://github.com/xadnavyaai/vectra-guard/releases/latest

# Verify installation
vectra-guard version
Step 2

Initialize

Set up global configuration, enable features, and sync CVE database for vulnerability scanning

Step 2: Initialize
# Initialize global config (~/.config/vectra-guard/config.yaml)
vg init

# Or repo-scoped config (.vectra-guard/config.yaml)
vg init --local

# Enable soft delete (optional but recommended)
# Edit config.yaml:
# soft_delete:
#   enabled: true
#   auto_delete: true

# Sync CVE database for vulnerability scanning
vg cve sync --path .
Step 3

Seed Agent Instructions

Configure your AI coding assistants (Cursor, Copilot, OpenClaw, etc.) to use Vectra Guard automatically

Step 3: Seed Agent Instructions
# Seed instructions for all supported agents
vg seed agents --target . --targets "agents,cursor"

# OpenClaw: auto-detects ~/.openclaw and merges safely
vg seed agents --targets openclaw
# Use --yes to skip confirmation (CI/scripts)
vg seed agents --targets openclaw --yes

# This creates:
# ✅ .cursor/rules/vectra-guard.md (Cursor)
# ✅ AGENTS.md (Generic agents)
# ✅ ~/.openclaw/AGENTS.md (OpenClaw — auto-detected)
Step 4

Start Using

Execute commands safely with protection, CVE scanning, and automatic backups

Step 4: Start Using
# Start a tracked session
SESSION=$(vg session start --agent "cursor-ai")
export VECTRAGUARD_SESSION_ID=$SESSION

# Run protected commands (with soft delete)
vg exec -- npm install
vg exec -- rm old-file.txt  # Automatically backed up!

# Scan for vulnerabilities
vg cve scan --path .

# Restore deleted files if needed
vg restore list
vg restore <backup-id>

What's Next?

Explore advanced features like sandbox configuration, trust management, and custom security policies

Seamless Agent Integration

Seed security rules directly into your AI agent's context

OpenClaw

Featured Integration

Local AI assistant with full system access — browser automation, shell commands, and file operations. VectraGuard auto-detects your OpenClaw install and merges security instructions without overwriting your custom rules.

VectraGuard detects ~/.openclaw/ (or env override), confirms the path interactively, and wraps its section in <!-- vectraguard:begin/end --> markers. Re-runs update only the marked section — your custom instructions are always preserved.

# ~/.openclaw/AGENTS.md preview
<!-- vectraguard:begin -->
## Safe Execution
Run commands via vg exec -- <command>
## CVE Scanning
vg cve scan --path . before installs
## Soft Delete
Files backed up via vg exec -- rm
## Secret Detection
vg scan-secrets --path .
<!-- vectraguard:end -->
OpenClaw Integration
# Auto-detects ~/.openclaw and prompts to confirm
vg seed agents --targets openclaw

#   OpenClaw detected (default)
#   State directory: ~/.openclaw
#   Destination: ~/.openclaw/AGENTS.md
#   Confirm? [Y/n/custom path]: y
#
#   [~] ~/.openclaw/AGENTS.md   2.3 KB (merged)

# Non-interactive / CI — skip the prompt
vg seed agents --targets openclaw --yes

# Re-run is idempotent — updates only the marked section
#   [~] ~/.openclaw/AGENTS.md   2.3 KB (updated)

Cursor

IDE-native rules for Cursor AI assistant

.cursor/rules/vectra-guard.md

Claude Code

Project instructions for Claude Code CLI

CLAUDE.md

GitHub Copilot

Agent instructions for Copilot Workspace

.github/copilot-instructions.md

VS Code

Security instructions for VS Code agents

.vscode/vectra-guard.instructions.md

Windsurf

Rule files for Windsurf AI editor

.windsurf/rules.md

Codex

Instruction set for OpenAI Codex agent

CODEX.md
$ vg seed agents --target . --targets "agents,claude,cursor,copilot,openclaw"# seed all at once

See It In Action

Interactive demos showing how Vectra Guard protects your development workflow

Blocking Dangerous Commands

See how Vectra Guard prevents destructive operations

Terminal
$vg exec -- rm -rf /
❌ CRITICAL: Command blocked for safety Risk Analysis: • Pattern: DANGEROUS_DELETE_ROOT • Risk Level: CRITICAL • Reason: Attempting to delete root directory Recommendation: Use specific paths instead of root directory deletion. This command has been blocked to protect your system.
Try this yourself!
Install Vectra Guard →

Security Without Compromise

Everything you need to protect your development workflow — from AI agents to production deployments.

Command Risk Guard

Blocks dangerous operations automatically before they hit your system.

Smart Sandboxing

Isolates risky actions in secure environments with intelligent caching.

CVE Scanning

Flags packages with known vulnerabilities before installation.

Soft Delete

Automatic backup system with restore capability for safe file operations.

Session Auditing

Track what ran, who ran it, and why — perfect for compliance.

10x Faster Installs

Cache-mounted dependencies make sandboxed installs blazing fast.

Explainable Risk

Human-friendly explanations for why something is risky.

Trust Store

Trust common commands once to skip sandbox on repeated runs.

Security Dashboard

Live event feed, risk charts, session management, and CVE data via vg serve.

Cross-Platform

Native support for macOS, Linux, and Windows with full feature parity.

Real-Time Protection

Analyze and protect commands as they're executed.

Agent-Friendly

Works seamlessly with Cursor, VS Code, Copilot, and OpenClaw workflows.

30+
Protected Directories
200+
Risk Patterns Detected
3
Platforms Supported
10x
Faster with Cache

VectraGuard Enterprise

Security operations center for AI-powered teams

Real-time threat dashboard with OWASP LLM Top 10 coverage

Session audit trail with filtering and CSV export

Agent monitoring with risk scoring and violation tracking

Policy management — injection detection, rate limits, redaction

Attack simulator with preset injection patterns

Google SSO with admin access control

VectraGuard Enterprise Dashboard
Active Sessions
24
Threats Blocked
156
Risk Score
Low
Threat Activity — Last 7 Days
MonTueWedThuFriSatSun
Recent SessionsLive
cursor-ai
12 cmdslow
openclaw
8 cmdsmedium
copilot
23 cmdslow

Install. Protect. Done.

30 seconds to production-ready security

Open Source

Free forever. Apache 2.0 licensed.

Three Commands. Zero Config.
# 1. Install
curl -fsSL https://raw.githubusercontent.com/xadnavyaai/vectra-guard/main/install.sh | bash

# 2. Initialize
vg init --local

# 3. Protect everything
vg exec -- npm install
vg cve scan --path .

Enterprise

Cloud dashboard for teams and organizations.

Real-time threat dashboard
Session audit trail with CSV export
Agent monitoring and risk scoring
Policy management and injection detection

Questions? Custom solutions?

x@navyaai.comnavyaai.com