Back to blog
NewProduct · Guard

Claix Guard Now Sends Alerts and Approval Requests to Slack

New in Claix Guard: Needs approval + Slack. A third verdict for changes that are allowed in principle but deserve a human look. Connect Slack with one click, pick a channel, and approve or reject AI agent INSERT, UPDATE and DELETE statements with a button. Each verdict can go to Slack or to your own endpoint.

What's new

Claix Guard now has a third verdict, Needs approval, and every verdict can be routed to an endpoint of your choice or to Slack. When an AI agent proposes an INSERT, UPDATE or DELETE that is allowed in principle but should be confirmed by a person, Claix Guard posts it to your Slack channel with Approve and Reject buttons. Connecting Slack takes one click: Add to Slack, pick a channel, done. There is no Slack app to create and no token to paste.

Slack is optional. The same popup can send a verdict to any public HTTPS URL you already run: an n8n Webhook, a Make or Zapier webhook, a serverless function, or your own API. You do not need Slack to receive alerts.

The update at a glance

BeforeNow
Two verdicts: allow or denyThree verdicts: Allowed, Denied and Needs approval
One alert endpoint for denied statementsA separate destination for each verdict: your endpoint or Slack
Slack meant creating your own app and pasting tokensAdd to Slack, choose a channel, send a test message
A person could not weigh in on a borderline changeApprove and Reject buttons in Slack, each calling the endpoint you configured

Existing alerts you already had on an endpoint keep working exactly as before. Adding Slack does not replace them unless you change that verdict's destination.

A free endpoint, or Slack

Each verdict has one destination. It is either a free public HTTPS endpoint, or the Slack channel you connected. You can mix them: Allowed can stay on the endpoint you already use, while Needs approval goes to Slack.

  • Endpoint. Claix POSTs the check result, including the proposed SQL and the verdict. This works for Allowed, Denied and Needs approval. A Needs approval alert sent to an endpoint has no buttons.
  • Slack. Allowed and Denied arrive as plain messages. Needs approval arrives with Approve and Reject. The message shows the proposed SQL so a person can read it before deciding.
  • No app to build. Add to Slack opens Slack's own screen. You pick the workspace and the channel there. Claix stores the connection. You never paste a bot token, a signing secret, or a channel id.

The step-by-step is in the guide Connect Slack to Claix. The check itself is POST /check-sql.

Why a third verdict

An agent that proposes database changes usually produces three kinds of statements:

  • Clearly fine. A single order changes status. Let it through.
  • Clearly not allowed. A DROP, a delete on a protected table. Block it.
  • Allowed in principle, but worth a second look. A refund above a threshold, a change that touches more rows than usual, an update on a sensitive table.

With only allow and deny, the third kind has to be forced into one of the other two. Either you block legitimate work or you let risky changes run unseen. Needs approval is the verdict for the third kind: the statement is paused until a person decides.

VerdictUse it whenExample
AllowedThe change is inside your limitsOne reservation changes from pending to confirmed
Needs approvalThe change is permitted, but a person should confirm this specific caseA refund above your threshold
DeniedNobody should be able to do thisA DROP, or a delete on a protected table

A practical rule: deny what nobody should ever do, ask for approval on what a person can reasonably judge, and allow the rest. If almost every approval request is accepted, the rule is probably too strict and can become an Allowed rule with a limit. If almost every one is rejected, it should probably be Denied. Too many approvals lead to people approving without reading.

How it works

Claix Guard is a checker. It decides whether a proposed statement is allowed, and Claix never executes the SQL. Your own code or workflow does. Approvals follow the same principle.

Agent proposes SQL
  -> your code calls /check-sql
  -> Claix Guard returns a verdict
       |- Allowed         -> alert to your endpoint or Slack
       |- Denied          -> alert to your endpoint or Slack
       |- Needs approval  -> endpoint, or a Slack message with Approve / Reject
                               |- Approve -> calls your approve endpoint
                               |- Reject  -> calls your reject endpoint

When someone clicks a button in Slack, Claix calls the endpoint you configured for that button. The Slack message already shows the proposed SQL. The call sends the decision, an approval id, a hash of the statement, the time, and the Slack user who clicked. Turn on one option in the same popup if that call should also include the SQL. The endpoint is what acts: your backend or your automation runs the statement on Approve, or notifies the agent or the user on Reject. Claix holds no write access to your database, so clicking Approve in Slack does not change anything by itself unless your endpoint does the work.

Set it up in three steps

  • Open the Guard alerts popup in your Claix dashboard.
  • Click Add to Slack, choose the channel on Slack's permission screen and accept. Back in Claix you will see the connected channel and can send a test message, change the channel or disconnect.
  • Choose a destination for each verdict: a free HTTPS endpoint or Slack. For Needs approval on Slack, set the endpoint that Approve calls and the endpoint that Reject calls.

Only Needs approval messages carry buttons. Allowed and Denied alerts arrive in the channel as plain messages. If you would rather not use Slack for a verdict, leave it on an endpoint. That POST is the check result, SQL included.

Example: approving a refund with n8n

Say an agent proposes an UPDATE that issues a refund above your threshold.

  • Your workflow sends the statement to Claix Guard and receives Needs approval.
  • Claix posts the request in Slack. The finance lead reads the SQL and clicks Approve.
  • Claix calls your approve endpoint, which is an n8n Webhook trigger followed by a Postgres node that runs the statement, and then confirms to the customer.
  • If the lead clicks Reject, Claix calls your reject endpoint, which sends the agent a message that the change was declined.

The same pattern works with Make, Zapier, a serverless function or any backend that can receive an HTTP request. If nobody needs a button, point Needs approval at the webhook directly and skip Slack.

Good practices

  • Pick the channel carefully. Alerts can include the proposed SQL, so send them where the right people, and only they, will see them.
  • Make the approve endpoint do the whole job, including running the statement, so the button always leads to an outcome.
  • Keep Denied for the things nobody should be able to do, and use approvals for judgment calls.
  • Review your approval rate from time to time and move rules between verdicts when they are always accepted or always rejected.
  • Send the test message before you rely on a channel.

Frequently asked questions

What is the Needs approval verdict in Claix Guard?
It is a third verdict, alongside Allowed and Denied, for statements that are permitted in principle but should be confirmed by a person before they run.
Does Claix run the SQL after someone clicks Approve?
No. Claix never executes SQL and has no write access to your database. Approve calls the endpoint you configured, and that endpoint, your backend or automation, runs the statement.
Do I need to create a Slack app?
No. Click Add to Slack, choose a channel and accept the permissions. You do not create an app or paste tokens.
Can I use a free endpoint instead of Slack?
Yes. Each verdict goes to one destination: a public HTTPS endpoint, or Slack. You can mix them. Your existing endpoint alerts keep working until you change that verdict.
Which Slack messages have buttons?
Only Needs approval messages have Approve and Reject buttons. Allowed and Denied alerts are sent without buttons.
What do the endpoints receive?
A verdict sent to an endpoint receives the check result, including the proposed SQL. A button click sends a signed POST with the decision, an approval id, a hash of the statement, the time, and who clicked. The SQL is added to that call only if you turn the option on.
Does it work with n8n, Make and custom backends?
Yes. Any endpoint that can receive an HTTP request works, including an n8n Webhook trigger, a Make or Zapier webhook, a serverless function or your own API.
Can I change the channel or disconnect Slack?
Yes. You can send a test message, change the channel or disconnect Slack from the same popup.
When should a rule return Needs approval instead of Denied?
Use Denied for what nobody should be able to do. Use Needs approval when the action is legitimate in some cases and a person can judge this specific one, for example a refund above a threshold.
Do Allowed and Denied alerts also go to Slack?
Yes. Each verdict can be routed to Slack or to an endpoint. Only Needs approval messages include buttons.

Try it

Open the Guard alerts popup in your Claix dashboard, click Add to Slack and send a test message. If you are new to Claix Guard, the free tier includes 30 successful calls per month, enough to test the whole flow. Product page: Claix Guard. Setup: Connect Slack to Claix.