Using ProceedGate with n8n

Place an HTTP Request node immediately before expensive or sensitive nodes (OpenAI, HTTP Tool, SerpAPI, CRM, etc.). Route with an IF node on allowed.

Try it first — no account needed

Sanity-check routing before creating a key. Want to see allow/deny branching in n8n without signing up? Import the no-account demo workflow. It calls POST /v1/check/demo — demo-only, no API key, no credits consumed, and does not protect real production tools.

First run with the same task_hash → allowed: true
Repeat the same task → allowed: false (demo storm path)

Download no-account demo workflow →

Demo vs production: The demo endpoint is for routing tests only. For real workspace tracking, credits, and logs, create a free key and use POST /v1/check (5,000 checks/month).

Verified API fields for POST /v1/check: agent_id, task_hash, optional action (tool_call · model_call · retry · …), optional step_hash. See Docs. Fields like workflow_id or reason_code in examples below are illustrative n8n patterns unless marked otherwise.

For real usage — need an API key? Create a free ProceedGate workspace to get 5,000 checks/month (no credit card). Paste your workspace key into the n8n HTTP Request node as:

Authorization: Bearer pg_ws_...

Do not hard-code the key in shared workflow exports. Get free API key →

HTTP Request node (production payload)

POST https://governor.proceedgate.dev/v1/check
Authorization: Bearer pg_ws_...
Content-Type: application/json

{
  "agent_id": "n8n-lead-enrich",
  "task_hash": "sha256-of-lead-id-or-stable-input",
  "action": "tool_call"
}

IF routing (illustrative)

Map the JSON body — production uses allowed and HTTP status (429 when blocked). On the ProceedGate HTTP Request node, enable Never Error (options.response.response.neverError: true in the repo export) so non-2xx responses like 402 or 429 with allowed: false reach the IF node instead of failing the workflow.

// IF node (example expressions; matches proceedgate-guard-sub-workflow.json)
{{ $json.allowed === true }}  → continue to OpenAI / SerpAPI / CRM
{{ $json.allowed === false }} → stop branch, Slack alert, dead-letter

// Optional: treat zone === "gray" as manual review (Planned workflow guard fields)

Native n8n guard vs ProceedGate-powered guard

Native n8n: Guard sub-workflow + Code node (e.g. “Evaluate Budget Limits”) + your DB/Redis + custom policy logic you maintain per workflow.

ProceedGate-powered: n8n keeps orchestration, approval UX, and workflow wiring; ProceedGate adds a shared guard decision/check layer. Same Guard sub-workflow shape, but replace the custom budget/repetition Code node with an HTTP Request to POST /v1/check. ProceedGate can replace that custom check with a managed allow/deny decision — then route primarily on allowed or HTTP 429. zone can be used for optional logging or review patterns where available, with shared decision logs across agents and workflows.

Illustrative workflow JSON (not production-tested). Download/import examples:

Import path A — native n8n: Import Guard Sub-workflow JSON → test mock Code budget logic.

Import path B — ProceedGate-powered: Import ProceedGate-powered Guard Sub-workflow JSON → replace REPLACE_WITH_PG_WORKSPACE_KEY → route on allowed/denied.

Then: Import Main AI Agent Flow and point Call n8n Workflow Tool to whichever guard workflow you want to test.

Shared Guard Sub-workflow pattern

Reuse one guard across many main workflows by calling a sub-workflow before any costly tool batch:

Main workflow → Call n8n Workflow Tool → Shared Guard Sub-workflow → ProceedGate Check → allow / warn / block → optional Human Review → costly tool (SerpAPI, OpenAI, CRM, …)

On block, the main branch stops before the next paid node. On warn, route to human review before CRM/SAP or other sensitive writes.

Example flow

Webhook → ProceedGate check → IF allowed → SerpAPI → Firecrawl → OpenAI → CRM. On block, skip paid nodes.

Walk through a labeled demo: Runaway workflow simulator.

Planned workflow guard fields

The fields below are not live /v1/check fields today; they are shown to explain the n8n workflow pattern we are exploring.

Not returned by /v1/check today — useful for future n8n node or policy packs:

reason_code: repeated_task, budget_threshold, too_many_tool_calls, sensitive_endpoint_guard
workflow_id, node_name for audit exports

Markdown in repo: examples/n8n/proceedgate-pre-call-guard.md · examples/n8n/native-vs-proceedgate-guard-workflow.md · JSON: guard-sub-workflow.json · main-ai-agent-flow.json · proceedgate-guard-sub-workflow.json

Privacy (n8n integration telemetry)

What we store (operator-only, not public): hashed API key fingerprint, hashed IP prefix, workspace id, agent id, action, task/step hashes, allow/deny outcome, zone, HTTP status category, user-agent string, request id, and timestamp for POST /v1/check. Workflow downloads via /dl/n8n/* log asset id + hashed IP only. We do not store prompts, raw payloads, raw IPs, or API keys.