Authentication Options

• None—No authentication. The webhook accepts requests from any caller. Use only in isolated test environments.
• Header Auth—AI Red Teaming sends a secret value in a custom HTTP header with every request. You define the header name and value in both n8n and AI Red Teaming. Available on the Webhook node only.
• Basic Auth—AI Red Teaming sends the username and password in the standard Authorization: Basic header using base64 encoding. You provide the credentials in plain text when creating the target; AI Red Teaming encodes them automatically.

Configure Your n8n Workflow

1. In your n8n instance, create a new workflow or open an existing one.
2. Add a Webhook trigger node as the first node:
   • Set HTTP Method to POST.
   • Set Respond to Using Respond to Webhook Node for REST mode, or to Streaming for streaming mode.
   • Configure Authentication to match the method you plan to use in AI Red Teaming (either Header Auth or Basic Auth).
3. (For text-only attacks) Add the AI model node (for example, Google Gemini — Message a Model or OpenAI — Message a Model) and connect it to the Webhook node. In the model node, set the Message field to reference the prompt from the request body:

   {{ $json.body.chatInput }}

4. (For multi-turn conversations) Add a Simple Memory node before the AI model node and key the memory on the session ID from the request:

   {{ $('Webhook').item.json.sessionId }}

   AI Red Teaming generates a stable session ID per conversation and sends it in the sessionId field. The Simple Memory node uses this ID to maintain context across turns while keeping separate conversations isolated from each other.
5. (For multimodal file attack support with Supports File Upload Option) Add an IF node after the Webhook node to check whether $json.body.files has items, then configure each branch:
   IF node (contains a file):
   1. A Convert to File node (operation: Move Base64 String to File) with Base64 Input Field set to body.files[0].data.
   2. An Extract From File node (operation: Extract From PDF) with Input Binary Field set to data.
   3. Add the AI model node (for example, Google Gemini — Upload a media file) and connect it to the Extract From File node. Set the Message field to the extracted text:

      {{ $json.text }}

   IF node (doesn't contain a file):
   1. Connect the Webhook node directly to the AI model node you added in Step 3.
   The IF node is required. Without it, an empty files array on a text attack causes the Convert to File node to error and the entire target fails validation.
6. (For REST mode) Add a Respond to Webhook node as the last node in the workflow. Set Respond With to JSON and map the AI model's output to the output field, for example:

   {{ { "output": $json.content.parts[0].text } }}

7. Toggle the workflow to Active (Published) in the n8n editor.
8. The n8n provides two webhook URLs for every Webhook node. Only the production URL works for AI Red Teaming scans. In the Webhook node, copy the Production URL. It uses the path pattern /webhook/<path>. You will enter this URL when creating the target in AI Red Teaming.
   Do not use the test URL (/webhook-test/<path>). It accepts only one request before it stops listening, so the first attack in a scan succeeds and all subsequent attacks fail with a connection error.

Response Modes

Multi-turn Conversations

• In stateful mode (used by some other connection types), the server returns a session ID that AI Red Teaming reads and chains into subsequent requests. n8n never returns a session ID, so stateful mode does not apply.
• In stateless mode, AI Red Teaming resends the full conversation history in each request. n8n agents that use a memory node expect a session ID, not a full transcript. Resending history would double-count against the memory node.

Multimodal File Attacks

AI Red Teaming
  → POST {"chatInput": "...", "sessionId": "...",
          "files": [{"mime_type": "application/pdf", "data": "<base64>"}]}
  → n8n Webhook node
  → IF node         (check: $json.body.files has items)
  → Convert to File (Move Base64 String to File; Base64 Input Field: body.files[0].data)
  → Extract From File (Extract From PDF; Input Binary Field: data)
  → AI Agent node   (prompt contains the extracted document text)
  → Respond to Webhook node

Deployment Considerations