Lovable Buyer's Checklist: Security, Export, Credits and Data

Is Lovable safe to build on? For most web apps, yes, provided you check six things before you pay. Lovable lists SOC 2 Type II, ISO 27001:2022 and AIUC-1 among its certifications, but it offers no HIPAA path, and the apps you build inherit none of those badges. By Lovable's own account, between February and April 2026 the chat history and source code of public projects could potentially be opened by any logged-in user with a project link. Code leaves through Git sync on every plan. Data does not travel with it. And Free and Pro prompts may train Lovable's models unless you opt out.

Cap the credit spend before the first prompt

Lovable bills each Build message on the work done (a Plan message costs 1 credit plus any subagent research), and failed or unwanted output is charged like a success; reverting to an earlier version refunds nothing. That makes spend control the first check, not the last. A single Build message can run for up to 10 hours. Credit check-ins, still in beta, pause a long message at 20 credits by default, but they are a notification you can wave through rather than a hard cap. Real limits arrive with Pro with 100 credits, at $25 per month billed monthly or $21 per month billed annually ^[1].

The trap is what those limits cannot see. Per-member limits count build credits only, so hosting, the built-in backend and the AI features of a live app draw on the workspace balance whatever a member's cap says. Visitors spend the owner's credits. External collaborators do too. At zero, published pages keep loading, AI features return errors, and apps on Lovable Cloud pause until someone adds credits and clicks Wake up; a top-up alone does not restart them. The full plan arithmetic sits in our Lovable pricing guide.

• Set a workspace usage limit and a default per-member limit (Pro and up) before inviting anyone; a limit that blocks Cloud usage pauses the live app's backend when reached, and adding credits does not lift the block.
• Look at auto top-up: its defaults buy 100 credits whenever the balance drops below 25, up to 400 credits a month.
• Name one person who watches Run credits, since no member cap covers them.
• Treat a check-in alert as a reason to stop and read the plan, not as a ceiling.

Code leaves with you; the backend stays put

Ownership is clear on paper: you own the code, apps and data, subject to third-party rights. Git sync runs two-way with GitHub, GitLab and Bitbucket Cloud on every plan, Free included, while the whole-codebase zip download needs a paid plan. Each project syncs to a new private repository that Lovable creates, and you cannot connect or import one you already have. Either route hands you source, config and database migration files, not the records in the database or the files in storage ^[2].

Everything else is a separate, manual job. The database export carries user accounts with password hashes but leaves out storage files, edge-function code, secrets and sign-in provider settings. An app on Lovable Cloud moves its backend to managed or self-hosted Supabase (any other backend means rewriting integration code), after which you rebuild auth providers, secrets, connectors and AI calls yourself; Supabase back to Cloud is not supported. Newer apps on TanStack Start also need a host that runs server code. Once the app runs on its own hosting and stops calling Lovable services, no subscription is needed. Plan the exit now.

• Connect Git sync on day one, so every change lands in a repository you control.
• Export the database regularly; the limit is one export every 24 hours, and none while the backend is paused.
• Keep secrets and sign-in settings documented outside Lovable, because they never export.
• If an exit is likely, consider your own Supabase from the start: Supabase then bills the backend, and credits pay only for building.

Who gets to read your prompts and code

Since September 2026, Lovable may use Free and Pro customer data, meaning prompts, code, files, outputs and usage, to train its models unless the user opts out ^[3]. The opt-out is per account, costs nothing, lives in Account settings and works only going forward, so whatever you sent earlier is already covered. Business and Enterprise workspaces are excluded by default and governed by a Data Processing Agreement, with no switch to flip. End-user data from your app and billing details are not used. Base44, by contrast, excludes data from training only on its sales-led Enterprise plan.

Two quieter settings deserve a look. The DPA still lets Lovable train its own models on aggregated or de-identified service data, with no opt-out. Separately, an AI app context setting keeps the full requests and responses of your app's AI features for 90 days, secrets removed; it is on by default on Free and Pro and off on Business and Enterprise. Prompts also travel to outside model providers whose contracts restrict training. In March 2026 Lovable named Anthropic Claude, Google Vertex AI, OpenAI and AWS Bedrock. Enterprise runs zero-data-retention models by default.

• Opt out before the first prompt, and ask every Free or Pro member to do the same.
• Review the AI app context setting if your app stores user conversations.
• Read the sub-processor list in Lovable's Trust Center before signing anything.

The April 2026 exposure and the scans that guard your app

Lovable's own account of the lapse is specific. Between February and April 2026, the chat history and source code of public projects could potentially be opened by any Lovable user holding a project link ^[4]. Lovable blames backend regressions from February and HackerOne reports closed without escalation. It says private projects and Lovable Cloud were never affected, that the fix landed within two hours of the public report, and that historically public projects were made private, official templates aside. Public projects can no longer be created.

That history argues for using the free tooling. A Quick scan of about 10 seconds runs whenever the Publish dialog opens, checking database access rules, dependencies and MCP exposure. A Deep scan on demand takes 3 to 13 minutes and hunts exposed secrets, unsafe input handling and authorization gaps. Both are free on every plan, and each account gets 10 free fixes. Lovable sets basic row-level security itself but tells you to review it, and says scans cannot guarantee complete security. Your app inherits none of Lovable's certifications, and Lovable signs no Business Associate Agreements.

• Turn on Block publishing with critical issues; it starts on only in new Enterprise workspaces.
• Run a Deep scan before any launch that touches user data.
• Keep health data, card data, government IDs and biometrics out; the Terms bar them without a written agreement.

Team gates: SSO on Business, SCIM on Enterprise

Lovable charges per workspace, not per seat, so headcount does not move the bill; the credit option does. Controls climb in steps. Business with 100 credits, at $50 per month billed monthly or $42 per month billed annually, adds workspace SSO over SAML 2.0 or OIDC, the Security center, workspace publishing and the default training exclusion. Directory sync, audit logs and forced two-factor sign-in wait for Enterprise, which is quote-only. Mind the unit, too: a Business credit costs twice a Pro credit at most menu options, so the same credit count costs twice as much on Business; for Cloud and AI the dollar cost is the same, which is why a Business workspace shows fewer credits used for them.

Three ownership details catch teams out. External collaborators build on the project owner's credits, so a contractor's busy week lands on your balance rather than theirs. Several workspaces each need their own subscription, because a plan belongs to one workspace only. And if finished apps will go to clients, moving a project to an account outside the workspace needs Business. Price those facts into the plan choice now, not after the first handoff.

Refunds, cancellation and what survives a downgrade

There is no free trial of Pro or Business. Subscription fees and credits are non-refundable except where the law requires, so the billing term is a real decision. The annual charge equals ten monthly payments: $250 a year against $300 billed monthly for 100 Pro credits a month, by our arithmetic. Paying monthly therefore costs less only if you stop within nine months. Canceling needs no ticket: Workspace settings, Plans and credit usage, Manage, then Downgrade to free. The mobile app offers no cancel option.

After the paid period ends, the workspace drops to Free. Unexpired plan credits freeze and return if you re-subscribe before they expire, while top-up credits stay usable. Published apps stay live, yet those on the built-in backend or with AI features still need credits to answer requests. A hidden badge comes back on the next publish. Addresses on lovable.app belong to Lovable and can be reclaimed, generally with 7 days' notice, so a custom domain on a paid plan is the safer home. Support is by email from Pro, first reply within 24 hours, mainly on weekdays; there is no phone or live chat.

• Choose monthly billing for any pilot of nine months or less.
• Export the database before downgrading an app you plan to retire.
• Cancel the subscription separately, since deleting your account does not stop billing.

The bottom line

Lovable passes a buyer's check for ordinary web apps once three settings are in place: a usage limit, the training opt-out on Free or Pro, and Git sync from the first day. Read the April 2026 exposure as a reason to keep projects private and run the free scans, not as a verdict on the platform. It is the wrong home for regulated health data, and for any plan that assumes the backend leaves as easily as the code. Teams that need SSO start at Business; SCIM and audit logs mean an Enterprise quote. If Base44 or Replit is still on your list, our Base44 vs Lovable comparison and Lovable vs Replit comparison set them side by side.