For AI Agent Builders

Built for AI Agents

EU AI Act compliance designed for how you actually build.

from langchain.agents import AgentExecutor
from protectron.langchain import ProtectronCallback

# Add Protectron to your existing agent
callback = ProtectronCallback(system_id="my-agent")

executor = AgentExecutor(
    agent=agent,
    tools=tools,
    callbacks=[callback]  # That's it
)

No refactoring. No infrastructure changes. Just add the callback.

The Agent Compliance Challenge

AI agents are different. They're autonomous. They make decisions. They use tools. They collaborate with other agents.

Traditional compliance tools don't understand this. They treat AI as a black box—input in, output out. But regulators want to see how your agent reached its decisions.

Protectron was built for agents from day one.

What We Capture

Every Agent Decision

Trace: trc_abc123
│
├── 10:23:01 agent_action
│   ├── thought: "User needs order status, I'll search..."
│   ├── action: search_orders
│   └── confidence: 0.92
│
├── 10:23:02 tool_call
│   ├── tool: search_orders
│   ├── input: {"order_id": "ORD-123"}
│   └── output: {"status": "shipped", "eta": "Dec 18"}
│
├── 10:23:03 agent_action
│   ├── thought: "Found the order, composing response..."
│   └── action: respond
│
└── 10:23:04 llm_call
    ├── model: gpt-5.2
    └── output: "Your order #123 has shipped..."

Multi-Agent Collaboration

For CrewAI and multi-agent systems:

Crew Execution: crew_xyz789
│
├── Agent: researcher
│   ├── task: "Analyze market trends"
│   ├── tools_used: [web_search, data_analysis]
│   └── output: "Market analysis complete..."
│
├── Agent: writer
│   ├── task: "Draft report"
│   ├── delegated_from: researcher
│   └── output: "Report draft ready..."
│
└── Agent: reviewer
    ├── task: "Quality check"
    ├── human_approval: required
    └── status: approved_by_supervisor@company.com

Human-in-the-Loop

When humans intervene:

# Automatically captured
protectron.log_human_approval(
    action="high_value_refund",
    approved_by="supervisor@company.com"
)

protectron.log_human_override(
    original="deny_request",
    override="approve_request",
    reason="Loyal customer exception"
)

Framework Support

LangChain

Native callback integration for Agents, Chains, Tools, Retrievers, and Memory.

View Guide →

from protectron.langchain import ProtectronCallback

callback = ProtectronCallback(
    system_id="langchain-agent",
    log_llm_content=True,
    pii_redaction=True
)

CrewAI

Complete multi-agent crew logging with per-agent tracking, task delegation, and collaboration.

View Guide →

from crewai import Crew
from protectron.crewai import ProtectronCallback

callback = ProtectronCallback(
    system_id="my-crew",
    log_agent_thoughts=True
)

crew = Crew(
    agents=[researcher, writer, reviewer],
    tasks=[research_task, write_task, review_task],
    callbacks=[callback]
)

AutoGen

Microsoft AutoGen conversation tracking for agent messages, code execution, and human messages.

from protectron.autogen import ProtectronLogger

logger = ProtectronLogger(system_id="autogen-app")

Vercel AI SDK

Edge-ready TypeScript integration.

import { streamText } from 'ai';
import { withProtectron } from '@protectron/vercel-ai';

const result = await withProtectron(
    streamText({
        model: openai('gpt-5.2'),
        prompt: userMessage,
    }),
    { systemId: 'my-app' }
);

Custom Agents

Building something custom? We've got you:

from protectron import Protectron

protectron = Protectron(system_id="custom-agent")

with protectron.trace("session-123") as trace:
    trace.log_agent_action(
        action="decide_next_step",
        reasoning="Based on user input...",
        decision="search_database"
    )
    
    trace.log_tool_call(
        tool="database_search",
        input={"query": "..."},
        output={"results": [...]}
    )

Why Agents Need This

Article 12 Compliance

The EU AI Act requires high-risk AI systems to have automatic logging that enables traceability. For agents, this means:

Requirement	What It Means for Agents
Automatic logging	Every decision logged without manual code
Traceability	Reconstruct how any decision was made
Input recording	What triggered each action
Duration tracking	When the agent was active

Beyond Compliance

Good logging makes your agents better:

Debug faster

See exactly what went wrong

Improve performance

Analyze decision patterns

Build trust

Show customers how decisions are made

Reduce liability

Evidence when disputes arise

Built for Production

Zero Performance Impact

Async by default — Non-blocking event delivery
<1ms overhead — No perceptible slowdown
Buffered uploads — Batch events efficiently
Graceful degradation — Agent works even if logging fails

Privacy First

PII redaction — Automatically detect and redact sensitive data
Content control — Choose what to log
EU data residency — All data stays in Frankfurt
Encryption — TLS 1.3 in transit, AES-256 at rest

Enterprise Ready

SOC 2 compliant — Enterprise security standards
SSO support — SAML 2.0, OIDC
Role-based access — Control who sees what
Audit exports — Ready for regulators

Get Started in 5 Minutes

Install the SDK

pip install protectron[langchain]  # or crewai, autogen

Get Your API Key

Add the Callback

from protectron.langchain import ProtectronCallback

callback = ProtectronCallback(system_id="my-agent")
# Add to your agent executor

Deploy

Your agent now has Article 12 compliant audit trails.

See It in Action

Dashboard View

Once integrated, your dashboard shows:

Real-time traces — Watch agent executions as they happen
Decision trees — Visual representation of agent reasoning
Tool usage — Which tools are called most
Performance metrics — Latency, token usage, success rates
Anomaly detection — Unusual patterns flagged automatically

Compliance Reports

Generate Article 12 reports with one click:

Event counts by type
Human oversight statistics
System availability logs
Exportable for auditors

Pricing for Agent Teams

Plan	Agents	Events/mo	Price
Starter	3	10,000	€99/mo
Growth	10	100,000	€299/mo
Scale	25	500,000	€599/mo
Enterprise	Unlimited	Unlimited	Custom

All plans include a 14-day free trial.

From the Community

"We added Protectron to our LangChain agents in an afternoon. Now we have complete audit trails for every customer interaction."
— CTO, AI Customer Service Startup

"The CrewAI integration is exactly what we needed. We can finally show enterprise customers how our multi-agent system makes decisions."
— Founder, AI Automation Platform

Resources

Quick Start Guide

Get started in 5 minutes

LangChain Guide

Detailed LangChain integration

CrewAI Guide

Multi-agent crew logging

SDK Overview

Complete SDK documentation

Article 12 Guide

Understand logging requirements

Ready to Make Your Agents Compliant?

Add EU AI Act compliance to your agents today. No infrastructure changes. No code refactoring. Just add a callback.

Start Free Trial View Documentation →

Questions? Email us at hello@protectron.ai