> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.lakera.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.lakera.ai/_mcp/server.

# Networking

Check Point AI Red Teaming makes outbound HTTPS calls to your target, so the target must be reachable
from the public internet. "Target" here means whatever Red calls — your agent endpoint
directly, or a [wrapper](/docs/red/creating-a-wrapper) in front of it.

## How Red reaches your target

Red is a SaaS hosted on [Convex](https://www.convex.dev/). During a scan, Red's backend
makes outbound HTTPS calls to the target URL you configured.

```
Red backend (Convex, AWS us-east-1)
        │   outbound HTTPS
        ▼
Your target endpoint
```

Red does not currently support native PrivateLink, VPC peering, or running inside your
VPC.

## Exposing a private target

If your AI workload runs in a private network (for example, EKS inside a private VPC),
common patterns include:

* **Public hostname on your existing ingress.** Add an ALB or NLB plus a DNS record
  (e.g. `https://redteam-ai.customer.com`) pointing at the Service or Ingress for the
  agent you want to test. Scope this hostname to the agent under test so you're not
  exposing broader control-plane infrastructure.
* **Mirrored staging environment.** Deploy a copy of the workload in a separate VPC or
  account with its own public ingress.
* **Reverse proxy in a DMZ.** Run a small public-facing service — a Fargate task, EC2
  instance, or pod behind a public ALB — that forwards requests from Red to your private
  endpoint. Also the right pattern if you need to control the egress IP yourself.

## IP allowlisting

If your firewall or WAF only accepts traffic from specific sources, allow these ranges:

| Protocol | Addresses                                             |
| -------- | ----------------------------------------------------- |
| IPv4     | `216.235.222.0/24`                                    |
| IPv6     | `2600:1f18:77e:2300::/56`, `2600:1f10:4a09:7c00::/56` |

The source of truth is the
[Convex networking documentation](https://docs.convex.dev/production/networking).

> **Warning**
>
> **IP allowlisting is not authentication.** These ranges are shared across all Convex
> tenants in the region, not just Check Point's.

## Scan load

A full scan sends thousands of prompts, often in parallel.

* Point Red at a **non-production or red-team environment** if at all possible.
* Check that any WAF, rate limiter, or upstream LLM provider quota in front of the
  endpoint can absorb the traffic.
* If the target is resource-constrained, lower **Max Concurrent Requests** in the
  target's **Advanced settings** so Red throttles from its side.

## Next steps

* [Agent Overview](/docs/red/connect-to-your-agent) for supported contracts and auth.
* [Creating a wrapper](/docs/red/creating-a-wrapper) if your API doesn't match Red's
  contracts.