Skip to main content

AI Agent Governance for Make.com, Zapier, and n8n: How to Enforce Policy and Keep Audit Trails

· 5 min read
David Sanker
David Sanker
Lawyer, Legal Knowledge Engineer & UAPK Inventor

Automation platforms like Make.com, Zapier, and n8n have made it easy to build AI-powered workflows. But "easy to build" doesn't mean "safe to deploy in production" — especially when those workflows take real-world actions: sending emails, filing documents, moving money, or updating records.

This post shows how to wire UAPK Gateway into any automation platform to get:

  • Policy enforcement (ALLOW / DENY / ESCALATE before any action runs)
  • Human approval gates (escalated actions wait for a real person)
  • Tamper-evident audit logs (hash-chained, cryptographically signed records)

Why You Need Governance Beyond What Automation Platforms Provide​

Make, Zapier, and n8n give you logs. They don't give you governance.

Logs tell you what happened. Governance decides what's allowed to happen and stops it if not.

The difference matters when:

  • A compliance team asks "who authorized this?" — not just "when did it run?"
  • A regulator asks for evidence a human reviewed high-risk actions before execution
  • A developer makes a mistake in a workflow and you need to know the $50K transfer was blocked, not just that the step errored
  • You're in a regulated industry (legal, finance, healthcare) where "the AI did it" is not a defense

UAPK Gateway is a sidecar that sits between your automation and the outside world. Every action goes through it first.

The Pattern: Evaluate → Branch → Execute​

The integration pattern is the same regardless of platform:

Trigger → Build action payload → UAPK Evaluate → Branch on decision:
  allow   → Execute action → Log result
  deny    → Stop / notify
  escalate → Notify reviewer → Wait for approval → Re-submit with override token

The evaluate call is a dry run — no side effects, instant policy check. The execute call does the same check plus runs the action.

Make.com​

UAPK Gateway has a native Make.com custom app with 8 modules.

Install: Import the app JSON from github.com/UAPK/gateway → sdks/make/app.json into your Make.com team.

Connection fields:

  • Base URL: https://api.uapk.info
  • Organization ID: from your dashboard
  • API Key: from your dashboard
  • Management Token: JWT from POST /auth/login (for approvals)

Sample scenario — email governance:

  1. Gmail Trigger → new email matching criteria
  2. UAPK: Execute Action
    • UAPK ID: email-agent
    • Action Type: send_email
    • Tool: email:send
    • Params: {"to": "{{email.from}}", "subject": "...", "amount": null}
  3. Router → branch on decision
    • allow → Gmail: Send Message
    • deny → Slack: Post Message (notify team of denial)
    • escalate → Slack: Post Message with approval_id, wait for manual step, then re-submit with override token

The key: the email doesn't send unless UAPK says allow. If the recipient is on a denylist, or the daily email budget is exceeded, it's blocked — automatically, without code changes.

Zapier​

The UAPK Gateway Zapier app is published as a private integration (request access at [email protected]).

It includes 4 creates and 3 searches:

  • Creates: Evaluate Action, Execute Action, Approve Action, Deny Action
  • Searches: Find Approval, List Pending Approvals, Find Audit Records

Sample Zap — financial workflow:

  1. Trigger: Typeform submission (payment request)
  2. UAPK: Evaluate Action — check if amount is within policy
  3. Filter: Only continue if decision = allow
  4. Stripe: Create Payment Intent — only runs after UAPK approval

For escalated actions (amount above threshold), a separate Zap handles the approval workflow:

  1. Trigger: Schedule (every 15 min) or webhook
  2. UAPK: Find Pending Approvals
  3. Filter: Has pending approvals
  4. Gmail: Send Email — alert reviewer
  5. UAPK: Approve Action — after manual confirmation step
  6. Returns override_token for the agent to re-submit

n8n​

The UAPK Gateway n8n node is published to npm as n8n-nodes-uapk-gateway.

Install: In your n8n instance → Settings → Community Nodes → n8n-nodes-uapk-gateway

The n8n node has a distinctive feature: three output pins — Allow, Deny, Escalate. This maps naturally to n8n's visual workflow branching.

[HTTP Request: webhook] → [UAPK Gateway: Execute]
                                    │
              ┌─────────────────────┼────────────────────────┐
           Allow                  Deny                  Escalate
              │                    │                        │
    [Postgres: Insert]     [Slack: notify]     [Email: alert reviewer]

Each output pin routes to a completely separate branch of your workflow. No extra router nodes, no conditional expressions to write.

Available operations:

  • Gateway: Evaluate, Execute
  • Approval: Get, Approve, Deny, List Pending
  • Records: List, Verify Integrity

Langflow​

For AI-native workflows built in Langflow, install the uapk-langflow PyPI package:

pip install uapk-langflow

Eight components appear in your Langflow sidebar under "UAPK Gateway":

  • Evaluate Action — policy dry-run, connect to Conditional Router
  • Execute Action — policy + execution, routes to Allow/Escalate branches
  • Approve / Deny Action — HITL decision nodes
  • Get Approval / List Pending Approvals — approval status queries
  • List Records / Verify Integrity — audit log access

The Langflow pattern is ideal for RAG pipelines where the agent reads documents and then needs to take regulated actions (file a brief, send a contract, initiate a payment). The Evaluate component gives compliance a checkpoint without breaking the flow.

What Goes in the Audit Log​

Every action that passes through UAPK Gateway generates a tamper-evident record:

{
  "record_id": "int-abc123",
  "uapk_id": "email-agent",
  "agent_id": "make-scenario-001",
  "action_type": "send_email",
  "tool": "email:send",
  "decision": "allow",
  "decision_reason": "All policy checks passed",
  "record_hash": "sha256:...",
  "previous_record_hash": "sha256:...",
  "gateway_signature": "ed25519:...",
  "created_at": "2026-03-28T..."
}

Each record includes:

  • The action details and policy decision
  • A SHA-256 hash of this record
  • The previous record's hash (hash chain — tampering is detectable)
  • An Ed25519 signature from the gateway

You can verify the entire chain with a single API call:

GET /api/v1/orgs/{org_id}/records/verify/integrity
# → { "valid": true, "record_count": 1842, "errors": [] }

This is what "court-admissible audit log" means in practice: not just logs, but cryptographically verified, hash-chained, independently verifiable records that you own and control.

Getting Started​

  1. Start the manifest builder at build.uapk.info — create your first agent manifest in under 5 minutes
  2. Get your API key from the UAPK dashboard
  3. Follow the integration guide for your platform:

Or contact us if you want a pilot on your infrastructure with expert setup included.