Integration Guides

Add EU AI Act compliance to your AI agents in 5 minutes. Drop-in trust layers for every major framework, backed by a centralized policy gateway.

How It Fits Together

Your Agent → Trust Layer → Gate → Allow / Block / Approve

LangChain • Google ADK • CrewAI • AutoGen • OpenAI Agents • RAG
🔗

LangChain / LangGraph

Add a callback handler to any LangChain agent or chain. Works with LangGraph too. Automatic audit trails, PII detection, consent gates, and injection scanning.

⏱ 5 min v0.2.0 + Gate support
🤖

Google ADK

Plugin for Google's Agent Development Kit. Drop into any ADK agent with one line. Automatic tool risk classification, audit chains, and policy enforcement.

⏱ 5 min v0.2.0 + Gate support
🛡

Gate Server

Centralized policy enforcement for multi-agent architectures. Docker Compose stack with real-time dashboard and Jaeger tracing. One command to start.

⏱ 2 min Docker + Dashboard

End-to-End Demo

See the full flow: start Gate, connect an agent, watch policy decisions happen in real time, and verify the tamper-evident audit chain.

terminal
# 1. Start the full stack $ git clone https://github.com/airblackbox/air-gate.git $ cd air-gate && docker compose up -d # 2. Run the end-to-end demo $ pip install air-langchain-trust air-adk-trust $ python examples/e2e_demo.py AIR Blackbox — End-to-End Demo Gate status: healthy [Agent 1: Research] web_search → auto_allow [Agent 1: Research] read_database → auto_allow [Agent 2: Outreach] send_email → require_approval [Agent 2: Outreach] update_crm → require_approval [Agent 3: Rogue] drop_table → block [Agent 3: Rogue] change_perms → block ✓ Chain verified: 6 entries, 0 tampered # 3. View the Dashboard $ open http://localhost:3000
1

Gate starts with Docker Compose

Brings up the Gate API (port 8000), real-time Dashboard (port 3000), and Jaeger distributed tracing (port 16686) in one command.

2

Agents submit tool calls to Gate

Each trust layer checks with Gate before executing a tool. Gate evaluates the action against your policy rules and returns allow, block, or require_approval.

3

Every decision is logged in a tamper-evident chain

HMAC-SHA256 hashes link each audit entry to the previous one. If anyone modifies or deletes an entry, the chain breaks and verification fails.

4

Humans approve pending actions via Dashboard

High-risk actions wait for human approval. The Dashboard shows pending actions in real time. Approve or reject with one click.

5

Export the audit chain for regulators

The full audit trail is exportable as JSON — timestamped, signed, and verifiable. Ready for EU AI Act Article 12 compliance submissions.

PyPI Packages

air-blackbox
Core Scanner
Compliance scan engine
air-compliance
CLI Tool
air-compliance scan .
air-langchain-trust
v0.2.0
LangChain + Gate
air-adk-trust
v0.2.0
Google ADK + Gate