Skip to main content

Getting Started

AIR Blackbox is an open-source compliance scanning platform for Python AI agents. It checks your code against EU AI Act Articles 9, 10, 11, 12, 14, and 15 and provides actionable remediation guidance.

What It Does​

  • Scans Python AI agent code for compliance gaps
  • Detects frameworks (LangChain, CrewAI, AutoGen, OpenAI, RAG)
  • Reports findings with severity levels and article references
  • Remediates by generating trust layer code you can copy-paste
  • Protects against prompt injection with 15+ detection patterns
  • Verifies audit trails via HMAC-SHA256 chains

Quick Install​

```bash pip install air-blackbox-mcp ```

Three Ways to Use AIR Blackbox​

1. MCP Server (Claude Desktop)​

The fastest way to get started. Install the MCP server and use it directly in Claude Desktop:

```bash pip install air-blackbox-mcp ```

Then add to your Claude Desktop config. See the MCP Server guide for full setup.

2. Python Library​

Import and use directly in your Python projects:

```python from air_blackbox_mcp.scanner import scan_code

results = scan_code(your_agent_code) print(results) ```

3. GitHub Action​

Add compliance scanning to your CI/CD pipeline:

```yaml

  • uses: airblackbox/air-compliance-action@v1 with: path: ./src ```

See the GitHub Action guide for full configuration.

EU AI Act Articles Covered​

ArticleRequirementWhat AIR Checks
Art. 9Risk ManagementRisk classification of tools and functions
Art. 10Data GovernanceData handling and privacy controls
Art. 11Technical DocumentationLogging and audit trail presence
Art. 12Record-KeepingAudit ledger and HMAC chain verification
Art. 14Human OversightHuman-in-the-loop and approval gates
Art. 15Accuracy & RobustnessInput validation and injection protection

Supported Frameworks​

AIR Blackbox detects and provides framework-specific guidance for:

  • LangChain / LangGraph — via `air-langchain-trust`
  • CrewAI — via `air-crewai-trust`
  • AutoGen — via `air-autogen-trust`
  • OpenAI SDK — direct integration
  • RAG pipelines — custom trust layer support

Next Steps​