Package Information
Documentation
@appomni/n8n-nodes-agentguard
AppOmni AgentGuard provides runtime security for n8n workflows that leverage AI agents.
n8n is a fair-code licensed workflow automation platform.
Installation
Operations
Credentials
Compatibility
Usage
Resources
Installation
Follow the installation guide in the n8n community nodes documentation.
Operations
| Resource | Operation | Description |
|---|---|---|
| Prompt | Classify | Send a prompt to AppOmni for DLP and/or prompt firewall analysis. The node has two outputs (Allowed, Blocked). |
Credentials
Create an AgentGuard API credential with:
- Tenant URL — the URL of your AppOmni tenant (e.g.
https://example.appomni.com). - Ingest Token — your AppOmni-issued ingest token; sent as
X-AppOmni-Ingest-Token. - Host override (optional) — overrides the
X-Hostheader; leave blank unless instructed by AppOmni support.
Compatibility
Last tested against n8n 2.19.2.
Usage
Wire the node before any LLM call you want to protect:
Trigger → AgentGuard ─┬─ Allowed → LLM call
└─ Blocked → log / alert / refuse
The output item carries an agentguard object with the verdict, the original prompt, and the full API response, ready for downstream nodes.
If the AppOmni tenant's policy includes a user_quarantine classifier, repeat-offender users can be temporarily locked out after a set number of strikes within a rolling window. For Chat Trigger workflows the node auto-populates User ID with the chat session ID, so users are quarantine-eligible by default — note that this scopes quarantine to the conversation, so starting a fresh chat session resets the strike count. For persistent identity across sessions (or for non-chat triggers), populate User ID, Username, or User Email under Additional Metadata explicitly — these resolve in that priority order, and any value you set there overrides the chat-session default. When a user is locked, AgentGuard returns response_action: "block" with block_reasons containing the quarantine reason — same shape as any other block, so the existing Blocked output handles it without extra wiring. Quarantine windows, strike counts, and lock duration are configured server-side per tenant; nothing needs to change in the workflow.