Deterministic runtimegovernance for autonomousAI agents.

Gateplex sits between your agents and the systems they act on. Each action routed through Gateplex is checked against your rules before it runs. Each decision is recorded.

Request an executive briefing Explore Shadow Audit

ALLOW / read_account_balance

BLOCK / export_customer_records

PENDING_APPROVAL / transfer_funds

Illustrative

Private VPC deployment: Runs inside your own cloud account.
One-line agent intercept: Route agent actions through Gateplex with a single change.
Two patent applications filed: Provisional applications, not yet granted.

How a decision is made

An agent acts.

Your AI agent decides to move money, export data or call an API. Gateplex sits in the path, so the action reaches it first.

Illustrative diagram

Agents act faster than anyone can review.

Three problems show up once agents touch real systems.

Runaway operations

A misconfigured or manipulated agent can move money, send data or call production systems before a person notices. Gateplex checks each action routed through it before it executes.

Regulatory and legal exposure

Regulators ask who authorised an action and what evidence exists. Gateplex records each decision and holds sensitive actions for human approval.

An audit trail you cannot trust

Logs written by the agent itself are easy to dispute. Gateplex keeps its own append-only, hash-chained record that can be verified for tampering.

See it running.

Live verdicts, the approval queue and the verified log, in one console.

Prompt Injection Detector

Rate Limiter

Repetition Detector

Spend Limit

PII Protection

Rate Limiter

PHI Keyword Block

Sample data.

Built to sit inside your perimeter.

Gateplex is a control point in the path of agent actions, not a reporting tool that reads logs afterwards.

Inline interception

Agent actions are routed through Gateplex with a one-line change, so each one is checked before it reaches the system it targets.

Deterministic policy gate

Rules decide, not a model. The same action under the same policy always gets the same verdict.

Your perimeter

In VPC and air-gapped deployments, evaluation is designed to run inside the customer's environment.

Each action routed through Gateplex gets one of four verdicts.

Select a verdict to see what an illustrative response looks like.

{  "verdict": "BLOCK",  "action": "export_customer_records",  "rules_triggered": ["PII Shield"],  "reason": "Output contains personal identifiers",  "logged": true}

Illustrative response. Field names in the API reference may differ.

Start from a compliance pack.

Add a ready-made set of rules in one click. You see exactly which rules a pack adds before you confirm, and you can switch off or delete any of them afterwards.

Prompt Injection DetectorDetects attempts to override the agent's instructions.

Rate LimiterCaps actions at 100 per hour.

Repetition DetectorCatches an agent repeating the same action more than 3 times in 60 seconds.

A pack is a starting point for your compliance team. It is built from rule patterns and is not a certification of compliance.

Ten rule types are available: PII Shield, Spend Limit, Keyword Block, Keyword Flag, Rate Limit, API Scope, Regex Match, Output Length, Prompt Injection and Repetition Detect.

For developers

One API call. A verdict your agent can act on.

Integrate with one API call or the Python SDK. Each response carries the verdict, the reasoning and the rules that matched.

agent.pyExample

from gateplex import GateplexClient

client = GateplexClient(api_key="gplx_xxxxxxxxxxxx")

response = client.log_intercept(
    agent_id="agent_example",
    event_type="tool_call",
    input="transfer 4,800 USD to vendor-042",
)

if response.is_blocked:
    raise RuntimeError(response.reasoning)▍

ResponseExample

{
  "intercept_id": "int_01h8x2k9example",
  "verdict": "BLOCK",
  "reasoning": "Spend limit exceeded",
  "triggered_rules": ["Spend Limit"]
}▍

Run it where your data lives.

Hosted

The fastest way to start. Gateplex runs the service and you connect your agents to it.

VPC Edition

Runs in your own cloud account. Evaluation is designed to run inside your environment.

Air-gapped

For environments with no external connection. Available on request.

Mapped to the questions examiners ask.

Source

The question

What Gateplex provides

EU AI Act, Article 12

Can you show a record of what the system did?

An append-only log where each entry is chained to the last with SHA-256, verifiable for tampering.

EU AI Act, Article 14

Can a person oversee and step in?

A PENDING_APPROVAL verdict holds an action until a person decides.

SAMA and QCB

Who authorised the action, and what controls ran first?

Pre-execution checks and a verifiable record of each decision.

HIPAA

Is protected data leaving in agent output?

PII detection. The HIPAA Safety pack is built from rule patterns only.

This mapping is informational. It is not legal advice or a certification of compliance.

See what your agents would have triggered, in 30 days.

The Shadow Compliance Audit is a fixed-fee diagnostic. Gateplex checks your agents' actions against your rules and reports what would have been flagged or blocked. Nothing is blocked during the audit, and the report links each finding to a tamper-evident log.

Request a Shadow Audit

Talk to us about the agents you are about to put into production.