# What Is Guardian? A Plain-English Guide to Approval Gates for AI Agents > What Guardian is, how its policies answer Allowed, Needs approval or Denied before a workflow acts, how approvals and the audit log work, and n8n setup. Source: https://learn.ynteractive.com/content/explainers/what-is-guardian-ai-approval-gate · Updated 2026-10-20 · Free from Stepthrough (https://learn.ynteractive.com) By [Robert Breen](https://learn.ynteractive.com/content/about) · Published October 20, 2026 · Updated October 20, 2026 · 6 min read Guardian (guardian-safety-gate.com) is an approval and audit layer for actions that cannot be undone. It sits inside your workflow, right before a step like sending an email, making a payment or deleting a record, and answers one of three ways: Allowed, Needs approval or Denied. Your workflow only acts on Allowed, or after a named person approves. Every decision is logged. You can try it hands-on in the free lesson [Guardian: Make Your n8n AI Agent Ask Before It Emails a Customer](https://learn.ynteractive.com/content/guardian-n8n-ai-agent-approval-before-sending-email), the first module of the free [Guardian Jump Start](https://learn.ynteractive.com/courses/guardian-jump-start). This page is for owners, operations people and anyone who runs, or is about to run, an [AI agent](https://learn.ynteractive.com/content/glossary/ai-agent) or [automation](https://learn.ynteractive.com/content/glossary/automation) that can send, pay or delete on its own. It is my own explainer, not an official Guardian guide. The facts come from guardian-safety-gate.com, Guardian's n8n guide and the Guardian app, all read in October 2026. ## What problem does Guardian solve? An AI agent that drafts client emails or sorts [invoices](https://learn.ynteractive.com/content/glossary/invoice) saves real time, right up to the moment it does something you cannot take back: an email with the wrong amount, a payment to changed bank details, a deleted record. You cannot unsend an email. So most owners end up checking every draft by hand, which gives back most of the time the agent saved. The usual answer is a line that every small office already draws in its head. The agent can read, sort and draft on its own. Anything that goes to a customer, moves money or changes a record waits for a person. Some things never happen at all. Guardian turns that line into rules the workflow cannot skip. This idea is called [human in the loop](https://learn.ynteractive.com/content/glossary/human-in-the-loop), and the rules are a kind of [guardrails](https://learn.ynteractive.com/content/glossary/guardrails). ## How does Guardian work? Five ideas explain what you see in the Guardian app. **Action types.** Each kind of action has a name in lowercase dot notation, such as email.send or record.delete. Your workflow sends that name, plus the details of the action (the payload), to Guardian before it acts. **Policies and rules.** A policy applies to one action type and holds one or more rules that check fields in the payload. In the lesson, a rule checks the recipient field: team addresses are allowed, anything else needs approval. Guardian's rule builder offers scenarios such as Set a number limit, Only allow specific values, Block specific values and Set an aggregate limit. When several rules apply, the strictest result wins. **Three verdicts.** Guardian answers ALLOW, REQUIRE_APPROVAL or DENY. Allowed actions run. Actions that need approval wait for a person. Denied actions do not run, and the attempt is recorded. **Deny by default.** Under Settings > Organization, Default Policy Behavior can be set to Deny unmatched actions, which Guardian marks Recommended. Any action type without a policy is then blocked, so you do not need a rule for "never delete": you just never write an allow policy for it. **Approvals and the audit log.** Waiting requests land in the Approval Inbox with the reason approval was required. An approver can use Edit Payload & Approve to fix a wrong amount first; the original payload is shown beside the edited one, and the edited version is checked against the policy again. Every decision goes into a hash-chained, tamper-evident [audit trail](https://learn.ynteractive.com/content/glossary/audit-trail). ## How does Guardian fit into n8n? Guardian has an n8n community node, an SDK and an MCP server. In n8n, following Guardian's own guide: 1. Put the Guardian node right before the action node, after the step that writes the email. 2. Set Action Type (for example email.send) and Payload, a JSON object whose field names match your policy, such as recipient, subject and body. 3. Connect only the Allowed output to the send. Point Denied and Needs Approval at a note or leave them unconnected. Never connect all three to the action. 4. For the approval path, add a Guardian Approval Trigger, then a Guardian node set to Enforce, then the send, which reads the approved payload. The node needs a Guardian API credential: an API key from the dashboard settings, a signing secret and a base URL. Keep only one active workflow listening for the same action type, or an approved email can go out twice. ## How do you test a policy safely? Test before anything real can happen: - **Test This Policy** in the policy form runs sandbox tests with a payload you paste. No request is created and nothing is logged. - **The policy tester** (the flask icon in the policy list) does the same for a saved policy. Guardian's guide says to run every policy through it with the real payload, because a rule whose field name differs from the payload, even by a typo, silently never fires. - **Listen Mode**, under the Organization tab, records what Guardian would have decided without blocking anything. It is advisory, so also disable the real send step while you test (select it in n8n and press D). ## What do businesses use Guardian for? - **Client emails wait for a person.** An AI agent drafts invoice reminders; team emails go out alone, client emails wait in the Approval Inbox. This is the free [Guardian n8n lesson](https://learn.ynteractive.com/content/guardian-n8n-ai-agent-approval-before-sending-email). - **Payment limits.** Payments under a set amount run; anything over it, or to changed bank details, needs approval. See [approval threshold](https://learn.ynteractive.com/content/glossary/approval-threshold). - **No deletes, ever.** With Deny unmatched actions on, an agent that asks to delete a record is refused. - **A record for the auditor.** Every allowed, approved and denied action is in one log, with who approved what. ## What are Guardian's limits? Guardian's own guide is clear about them. Guardian never performs the action itself: your workflow does the sending, and only its Allowed and approved paths should reach the send step. The tamper-evident chain covers the submitted request and the decision; approver identity, approval time and edited payloads are stored and fingerprinted but not yet chained. Listen Mode results are advisory. A per-action setting for what happens when Guardian cannot be reached is planned, not built. Check the guide for the current list. ## Is Guardian free? Guardian's site said in October 2026 that it is free to start, with no credit card. In the app, some features, such as the AI Wizard for writing policies and Allow unmatched actions, need a paid plan. Check the Pricing page in the app for current plans. This page does not list prices because they change. ## How do you start learning Guardian? Start with the free [Guardian Jump Start](https://learn.ynteractive.com/courses/guardian-jump-start): 1. [Guardian: Make Your n8n AI Agent Ask Before It Emails a Customer](https://learn.ynteractive.com/content/guardian-n8n-ai-agent-approval-before-sending-email): turn three plain lines into a policy, test a team email and a client email, save it, and see where the Guardian node and the Approval Inbox fit in n8n. About 14 minutes. New to AI agents? The free [AI Agents 101](https://learn.ynteractive.com/courses/ai-agents-101) course explains what an agent is and when it needs limits. More Guardian modules will be added to the Jump Start as they are built and tested in the real app. ## Key takeaways - Guardian checks an action before your workflow runs it and answers Allowed, Needs approval or Denied. - Policies apply to one action type, such as email.send, and check fields in the payload. - Deny unmatched actions blocks anything without a policy, which is how you make "never delete" automatic. - Approvers can fix a payload before approving, and every decision is recorded in a tamper-evident log. - Test with the real payload before you rely on a policy, and connect only the Allowed output to the action. ## Frequently asked questions **Does Guardian send the email?** No. Guardian only decides and records. Your n8n workflow does the sending, and only its Allowed and approved paths should reach the send step. **What happens to an action with no policy?** With Deny unmatched actions, the recommended default, it is blocked and the attempt is recorded. **Do I need to code to use Guardian?** No for n8n: you build policies in the Guardian app by filling in forms, and add the Guardian node in n8n. Guardian also has an SDK and an MCP server for developers. **Is this an official Guardian course?** No. Guardian is a Stepthrough partner, and this is my own explainer and lesson, built in a real Guardian account. ## Sources - [Guardian](https://guardian-safety-gate.com) home page (read October 20, 2026) - [Integrating Guardian into an n8n workflow](https://guardian-safety-gate.com/n8n), Guardian's n8n guide (read October 20, 2026) - The Guardian app, used to build the linked lesson on October 19, 2026 ## Practice it free **Guardian: Make Your n8n AI Agent Ask Before It Emails a Customer**. A business case for a small firm whose n8n agent drafts client emails: turn "team emails go out alone, client emails wait for a person, deletes never happen" into a Guardian policy, test it, and see… You do every step yourself in a practice copy of Guardian, it checks your work as you go, and “Do it for me” finishes any step you get stuck on. Free, and nothing touches your real accounts. [Start the free lesson](https://learn.ynteractive.com/modules/guardian-n8n-approval-gate) [Or read the full step-by-step guide](https://learn.ynteractive.com/content/guardian-n8n-ai-agent-approval-before-sending-email) ## Free lessons for this - [Guardian: Make Your n8n AI Agent Ask Before It Emails a Customer](https://learn.ynteractive.com/content/guardian-n8n-ai-agent-approval-before-sending-email): A business case for a small firm whose n8n agent drafts client emails: turn "team emails go out alone, client emails wait for a person, deletes never happen"… ## Related terms Plain-English definitions, each with an example from a free lesson. [AI agent](https://learn.ynteractive.com/content/glossary/ai-agent) · [Automation](https://learn.ynteractive.com/content/glossary/automation) · [Invoice](https://learn.ynteractive.com/content/glossary/invoice) · [AI agent permissions](https://learn.ynteractive.com/content/glossary/agent-permissions) · [Policy engine](https://learn.ynteractive.com/content/glossary/policy-engine) · [Wait node](https://learn.ynteractive.com/content/glossary/wait-node) · Agent orchestration · [Agentic AI](https://learn.ynteractive.com/content/glossary/agentic-ai) [The full AI glossary](https://learn.ynteractive.com/content/glossary) ## More AI explainers - [AI Document Processing for Small Business: A Free 10-Module Course on OCR, Parsing, PDFs and E-Signatures](https://learn.ynteractive.com/content/explainers/ai-document-processing) - [AI Transcription for Meetings: A Free 9-Module Course to Turn Talk Into Work](https://learn.ynteractive.com/content/explainers/ai-transcription-for-meetings) - [How to Build an AI Support Bot: A Free 10-Module Course, from Your FAQ to Human Handoff](https://learn.ynteractive.com/content/explainers/build-an-ai-support-bot) [All AI explainers](https://learn.ynteractive.com/content/explainers) · [All free tutorials](https://learn.ynteractive.com/content)