ClaudeGuard is currently in beta testing. Features and pricing may change as we improve the product.
Security for AI Agents

Catch cloud mistakes before they hit production

ClaudeGuard is a team-wide safety layer for cloud work. Every AWS and Google Cloud command is reviewed before it executes, so you catch problems early, not after production breaks.

Create Team See How It Works

Up and running in 60 seconds

No config files. No API keys. Just install, login, and go.

cloudguard | setup + live session
# 1. Install
$ pip install git+https://github.com/roeygraif/ClaudeGuardPublic.git
 
# 2. Join your team
$ claude-guard login
Team invite code: cg_aBcDeFgH...
Email: alice@acme.com
✓ Logged in to Acme Corp.
 
# 3. Launch, that's it
$ claude-guard
━━━ CLOUDGUARD ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Connected to Acme Corp (server mode)
All cloud commands will be supervised.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 
# Agent tries to delete a production database
$ aws rds delete-db-instance --db-instance-identifier prod-postgres
 
━━━ CLOUDGUARD ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Risk: [CRITICAL]
Cost: Savings: ~$1,250/month
 
Blast Radius:
→ web-api: SQL queries will fail
→ batch-processor: nightly ETL broken
→ analytics-dashboard: reports offline
 
This is a production database with 3 active
dependents. Deletion is not reversible.
Recommendation: Create a snapshot first.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 
Confirm to proceed? (yes/no) _

Not just dangerous commands

ClaudeGuard understands your environment: what's production vs. dev, which resources are sensitive, and how services connect.

🛡

Environment-Aware Analysis

When a change looks harmless but could expose data, open the network, weaken IAM, or break dependencies, ClaudeGuard flags it instantly with a clear explanation and safer alternatives.

💥

Blast Radius Preview

Before anything runs, ClaudeGuard shows what this action might affect and what could go wrong. Fewer surprise outages and far fewer "who did this and why?" moments.

💰

Cost Estimates Up Front

Every resource creation, resize, or deletion includes a cost estimate. Catch expensive changes before they land on next month's bill.

🧠

Shared Infrastructure Memory

As your team works, ClaudeGuard builds shared memory: who created a resource, what it's used for, and what depends on it. If someone tries to modify something another developer just shipped, ClaudeGuard warns them before the damage is done.

📋

Team Audit Trail

Every analyzed command is logged with its risk assessment. Review your team's cloud activity from the dashboard. Full history of what was attempted and what was flagged.

🤖

Built for AI Agents

AI agents with real cloud credentials are powerful but risky. ClaudeGuard is the practical way to use powerful automation without losing control.

How ClaudeGuard works

Agent runs a cloud command

Claude Code (or any terminal agent) tries to execute an AWS or GCP CLI command.

ClaudeGuard intercepts

A PreToolUse hook catches the command before it runs. Read-only commands pass through instantly.

Infrastructure context gathered

ClaudeGuard looks up the target resource, walks its dependencies, and checks environment, IAM, and network state.

AI risk analysis

The full context is sent to the analysis engine. It returns risk level, blast radius, cost estimate, and recommendation.

Developer decides

The warning is displayed inline. The developer (or agent operator) confirms or cancels. Everything is logged.

Simple pricing

Free during beta

We'll announce pricing before the beta ends. Early teams keep their rate.

Create Team

How your data is used

ClaudeGuard performs read-only scans of your cloud infrastructure (AWS and GCP) to build a live map of your resources, dependencies, and permissions. This is how we provide accurate blast radius analysis and environment-aware risk scoring.

We never modify, create, or delete any resources in your cloud accounts. All infrastructure data is isolated per organization and never shared across teams. ClaudeGuard is currently in beta — we're actively improving accuracy and coverage.