AI Agent Control Plane

AI actions,
governed by Custos.

Custos provides policy enforcement,
auditability and runtime security
for AI agents operating against
databases, APIs and enterprise systems.

Get Started View Architecture

What Custos Does

Control. Observe. Protect.

Observe

Real-time visibility into AI agent actions and system interactions.

Decide

Policy-based decisions ensure actions are safe, compliant, and reliable.

Execute

Approved actions are executed through secure connectors and SDKs.

Protect

Built-in guardrails and audit trails keep your systems and data safe.

Architecture Overview

AI AGENTS

CUSTOS SDK

Language-ready SDKs for every stack

CUSTOS GATEWAY

Unified entry point
for all agent actions

CUSTOS CENTRAL

Policy Engine

Define and enforce policies

Decision Core

Evaluate context and decide

Audit & Logs

Record and monitor every action

SYSTEMS TO PROTECT

GET STARTED

Add verification to your AI actions

Works with your existing PostgreSQL and psycopg2 code. Just route traffic through the Custos gateway.

import psycopg2
from custos import Custos

conn = psycopg2.connect(port=9999)
custos = Custos(agent_id="my-agent")

plan = {
  "intent": "List actors",
  "action": "SELECT * FROM actor LIMIT 10;"
}

with custos.secure_action(plan) as ctx:
    conn.cursor().execute(plan.action)

Read the Docs View on GitHub

Custos doesn’t just run AI actions — it governs them in real time. Here are examples of what gets blocked:

plan = {
  "intent": "Reset database",
  "action": "DROP TABLE users;"
}
# ❌ Blocked: destructive operation detected

plan = {
  "intent": "Analyze customers",
  "action": "SELECT ssn, credit_score FROM customers;"
}
# ❌ Blocked: Dangerous hallucinated column access : sensitive fields + unauthorized access
# Policy: PII columns restricted for this agent