n8n Radware Agentic AI Protection

This package provides the Radware Chat Model community node for n8n AI Agents.

It is intentionally positioned as an in-path integration only. In n8n, this is the only clean customer deployment pattern that can protect the AI Agent interaction as a full model path: AI Guardrails for prompts and responses, plus Behavioral / Agentic Protection when tool context is present in the model request.

Out-of-path is not exposed as a public node in this package. n8n community nodes cannot globally intercept every native AI Agent tool or workflow action before execution. An explicit guard node can protect only the actions a customer manually routes through it, which is not a full one-go n8n AI Agent protection deployment.

Installation

Install in n8n (recommended)

1. In n8n, go to Settings -> Community Nodes.
2. Click Install.
3. Enter the package name: @radware/n8n-nodes-radware-agentic-protection.
4. Agree to the community node risk notice.
5. Click Install and wait for n8n to finish installing the package.

Self-hosted / manual install (alternative)

For self-hosted instances where you manage packages directly:

npm install @radware/n8n-nodes-radware-agentic-protection

Restart the n8n instance after a manual install so it picks up the new node.

Migrating from the legacy package

The official Radware package replaces the legacy unscoped package n8n-nodes-radware-agentic-protection.

1. Export or back up important workflows before changing community packages.
2. Uninstall the legacy package from Settings -> Community Nodes.
3. Install @radware/n8n-nodes-radware-agentic-protection.
4. Open each protected workflow and confirm Radware Chat Model is connected to the AI Agent.
5. Re-select the existing Radware In-Path API credential if n8n asks for it, then run a benign test and a policy-block test.

The Radware API credential itself does not change. The package scope changes the n8n node type identifier, so existing workflows must be checked after migration.

Credential

Create a Radware homegrown agent with in-path enforcement in Radware Cloud:

https://console.radwarecloud.com/

Getting your Radware API key

Your Radware API key is created in the Radware Cloud portal (console.radwarecloud.com) when you add a Homegrown Agent with In-Path Enforcement. The key starts with sk-rdwr- and is displayed only once, at creation time. Copy it and store it securely (a secret manager or your n8n credential store) immediately — Radware cannot show it to you again.

Then create Radware In-Path API credentials in n8n:

• Radware API Key: API key copied from the Radware in-path homegrown agent.
• Provider: provider segment Radware forwards to, such as openai, anthropic, google, or custom.
• Resolved Base URL: optional full base URL. If empty, the node uses https://api.agentic.radwarecto.com/v1/<provider>.

Do not store Radware keys in workflow JSON, screenshots, Git, or validation reports.

Node

Radware Chat Model

Use this node as the LLM model connection for n8n AI Agents. The AI Agent sends model traffic to Radware instead of directly to OpenAI, Gemini, Anthropic, or another provider. Radware applies Agentic AI Protection and proxies the request to the provider configured for the Radware homegrown agent.

Typical canvas:

Chat Trigger -> AI Agent
                ^
                Radware Chat Model

Direct OpenAI/Gemini/etc. model credentials are not used by that AI Agent path. The customer configures the upstream provider in Radware Cloud.

Coverage

• AI Guardrails: prompt, response, topic, HAPBlocker, and PII controls through the in-path model request/response.
• Behavioral / Agentic Protection: unsafe tool/action detection when the AI Agent's model request includes the relevant tool definitions and tool-call context.

For Behavioral validation, use a deterministic tool-call case where the model request includes the prior tool output and all relevant tools, for example read_email and send_email.

Example

• examples/in-path-agent-model-example.json: AI Agent using Radware Chat Model as its in-path model.

Validation

Minimum validation:

• Benign prompt is allowed.
• Credit-card PII prompt is blocked by AI Guardrails.
• HAPBlocker prompt is blocked by AI Guardrails.
• Medical/medicine blocked-topic prompt is blocked by AI Guardrails.
• Benign tool action is allowed.
• Malicious tool action is blocked by Behavioral / Agentic Protection when full tool context is present.

See docs/validation.md.

Security Notes

• The package has no runtime dependencies.
• The package exposes only the in-path language model node for n8n customers.
• In-path is fail-close by default: if Radware/provider connectivity fails, the AI Agent model turn fails rather than silently calling an unprotected provider.

Resources

• npm package
• GitHub repository
• Radware Cloud portal
• Customer guide
• In-path setup
• Validation guide