Lovable Webhooks: Keep Customer Data Out of Prompts and See Every Publish
Since October 27, 2026, Lovable can send webhooks for publishes and other workspace events on the Business and Enterprise plans, and it can let your own server check and block chat messages. The hook only sees the project chat, not the API, MCP or the first prompt of a new project, and by default a server outage blocks the entire workspace.
What are Lovable Events & Hooks?
Since October 27, 2026, Lovable can send messages to your own server whenever something happens in your workspace. The Lovable changelog lists the feature for the Business and Enterprise plans. There are two kinds:
- Events report something after the fact: a project was created, published, transferred to another workspace or deleted. According to the Lovable documentation, they contain only IDs and timestamps, no project content.
- Hooks ask for permission first. Right now there is exactly one:
message.before_submit.v1. It sends every message from the project chat to your server before Lovable's agent receives it. Your server answers withallowordenyand can add a reason of up to 1,024 characters that the user sees.
Under the hood these are webhooks: Lovable calls an HTTPS URL you define and signs every request using the Standard Webhooks spec. Only admins and owners can set them up under Workspace settings, Customization, Events & Hooks.
What are they good for in a company?
Two things: an overview of which apps your team builds and publishes, and a gate for data that has no business being in an AI prompt.
| Goal | Trigger | How |
|---|---|---|
| See every publish in your team channel | project.published.v1 |
Webhook in n8n, message to Teams or Slack |
| Keep a list of all Lovable apps | Created, transferred, deleted | Webhook writes to a spreadsheet or database |
| Alert when a project leaves the workspace | Transferred | Email or ticket to the admins |
| Keep customer data, IBANs and credentials out of prompts | message.before_submit.v1 |
Pattern check, deny response with a hint |
The first row sounds trivial, but it answers a question every team with several Lovable users eventually asks: who actually made which app public?
How do you build a data filter with n8n?
With two nodes and a fast check in between. Here is how:
- Add a Webhook node and set the Respond option to "Using 'Respond to Webhook' node", as described in the n8n documentation.
- Verify the signature. Lovable sends the headers
webhook-id,webhook-timestampandwebhook-signature. Verification runs on the unmodified raw request body; if you parse the JSON first and serialize it again, the check fails. Ready-made libraries exist for JavaScript and Python. - Check the message. Patterns for IBANs, email addresses, tax IDs and common API key prefixes are enough to start.
- Respond with the Respond to Webhook node: status 200, JSON with
decisionand, fordeny, areasonsuch as "Please don't paste customer data into the chat, use test data." - Choose the failure behavior and trace a test message in the destination's delivery history.
Our take: don't run a language model for the check. Lovable gives all destinations five seconds combined and does not retry hooks. A model call with any wait time breaks that limit quickly, and then the failure behavior kicks in. A few regular expressions answer in milliseconds.
Where does the hook have gaps?
It is a safety net, not a guarantee. The documentation names the limits itself:
- Only the project chat is checked. Messages sent through the Lovable API or the MCP server, in the Chats area, by voice, when creating or remixing a project and when editing a queued message bypass the hook. If someone pastes a customer list into the first prompt of a new project, nothing stops them.
- Your server sees exactly the sensitive data. The hook receives the full message text plus download links to attachments that anyone with the link can use for five minutes. Your filter server therefore processes personal data. Host it in the EU, keep message text out of your logs and clear it with your data protection officer.
- If your server goes down, the workspace stops. The default is "Block submission": if the endpoint does not answer in time or answers incorrectly, every message in the entire workspace is blocked until you turn the hook off. The alternative, "Allow submission", lets everything through during an outage, so you filter nothing.
- Below Business, no hook runs. If the workspace drops below the Business plan, hooks stop, including those set to "Block submission". A downgrade silently switches the filter off.
- Events are not real-time alerts. On errors, Lovable tries at most eight times within 24 hours, and the first retry comes after 15 minutes. Deliveries can arrive twice or out of order; the
webhook-idstays the same and lets you drop duplicates. - The signing secret is shown only once. It cannot be revealed again or rotated. If you lose it, delete the destination and create it again. Store it in your password manager right away.
What does it cost?
Events & Hooks require the Business plan. According to the Lovable plan overview, Business on monthly billing costs between $50 (100 credits) and $4,300 (10,000 credits), while Pro with 100 credits costs $25. On annual billing it is $500 instead of $250. Per credit, Business costs you roughly twice as much.
In return, Business brings more than webhooks: single sign-on, a Data Processing Agreement, the Lovable API and the Security Center. Since October 16, there is also the Action Center, which lists open publish requests, projects of departed employees and sudden spikes in credit usage. If you want to hard-block publishing or tie it to an approval, you need workspace policies, and according to the changelog those are Enterprise only.
What does this mean for you?
- Two of you build on Pro: No action needed. A clear rule of "no real customer data in prompts" plus test data does more than a plan change.
- Several departments build with Lovable: Business often pays off anyway because of SSO and the Data Processing Agreement. Then set up the publish event first. In n8n it takes just one webhook and a message to your team channel, and from then on you see what goes live.
- You work with customer or patient data: Use the hook as a second line of defense, not the only one. The gaps for the API, MCP and new projects remain, so you also need training and a rule for how projects get started.
Our take: the hook is the first tool that lets a company decide for itself what Lovable's agent gets to see. If you are already thinking about data protection in Lovable, read our overview of Lovable and GDPR alongside it, and if you are running the numbers on a plan change, see our article on Lovable credits and limits. If you want to set up the filter or the publish log with n8n, get in touch.
Frequently asked questions
Can a Lovable hook stop an app from being published?
No. According to the documentation, the only hook right now runs before a chat message is submitted. To block publishing or require an approval, you need workspace policies, which are only available on the Enterprise plan.
Do Lovable events give my server access to a project's code or data?
No. Event payloads contain only IDs and timestamps.
Does my webhook server have to be publicly reachable?
Yes, Lovable needs a public HTTPS URL. Requests come from fixed IP ranges, so you can allow those in your firewall and block everything else.
What happens if several filter servers are set up?
They all share the same five seconds. As soon as one answers with deny, the message is blocked, and Lovable shows one of the reasons.
Sources
Tell us what you're planning. You'll get an honest assessment within 24 hours.
More articles