The short answer: the AI stops, and on some builders your live app can stop too. On Bolt and v0 the docs say building or generation stops (they do not say what happens to a published app), and in Cursor, Copilot and Claude Code the tool is an editor, so only the AI stops. On Base44 the docs say actions that need integration credits (email, AI calls) fail with an error, and do not say whether the app itself stays up. On Lovable, Lovable’s docs say apps that use the built-in Cloud backend “can pause” at zero credits while the published pages keep serving. If you have real users, that last one is the difference that matters.
The table
Everything below is from each vendor’s own documentation, read on 29 September 2026. “Not stated” means the docs we could read did not say, not that the answer is no.
| Tool | The AI when you hit zero | Your published app | Top-up mid-cycle | Unused credits |
|---|---|---|---|---|
| Lovable | Blocking dialog; building stops until credits return | Pages keep serving; Cloud-backed apps “can pause”; runtime AI features fail; data stays safe | Yes on paid plans (Pro top-up $15 per 50, $0.30 each, per the pricing index); auto top-up | Monthly plan credits expire 2 months after issue on monthly billing |
| Bolt.new | AI stops; direct edits in Code view stay free | Not stated | Token reloads only on the highest monthly Pro tier or annual Pro plans; otherwise upgrade | Paid tokens roll over one extra month; Free tokens do not roll over |
| v0 | “Generation pauses” | Not stated | Yes on Premium, Plus and Business | Monthly credits expire after 65 days; purchased credits after one year |
| Base44 | Cannot send new chat messages until credits return (monthly reset or plan change) | Not stated for the app itself; any action that needs integration credits fails with an error, and end users see a generic message | No top-up in the docs; upgrade the plan | Expire at the end of the cycle, no carry-over |
| Replit Agent | Controlled by budget limits and usage alerts you set | Deployments billed separately (data transfer, compute); failed payments can suspend published apps | Credit packs from the billing page | Pro credits roll over for two months |
| Atoms.dev | Not stated in the help centre we could read | Not stated | Atoms’ pricing FAQ says to upgrade the plan for more credits; no separate top-up is listed | Some unused credits roll over one month (for example after an upgrade), per the pricing FAQ; not permanent |
| Cursor | On-demand usage at the same API rates, or upgrade; Cursor says requests are not downgraded | Not applicable (editor) | On-demand billing | Not stated |
| GitHub Copilot | Wait for reset, enable paid usage with a dollar budget, or switch to a cheaper model | Not applicable | Yes, budgeted; 1 AI credit is $0.01 | Not stated |
| Claude Code (Pro or Max) | Shared five-hour and weekly limits; wait, upgrade, or turn on pay-as-you-go usage credits | Not applicable | Usage credits at standard API rates, prepaid | Not applicable |
The Lovable case, with a real report
Lovable’s docs state the rule plainly: at zero, “apps that rely on the built-in backend (Cloud) can pause”, database and storage data remains safe but is inaccessible until credits return, and AI features in deployed apps stop. The Prompt-Led Product newsletter published a first-person account on 17 September 2026 of a Pro-plan app whose login and database went dark on a Saturday morning. Its author says it looked identical to a server outage to users, and attributes it to Lovable having merged the build and run credit balances into one pool in June 2026. We could not confirm the June change against Lovable’s docs, so treat that part as the author’s account. The docs page above confirms the behaviour itself.
That single-pool design has a cost angle. A top-up costs $0.30 per credit against $0.25 for a Pro plan credit, so an emergency purchase carries a 20% premium. The Lovable pricing guide has the full credit maths.
Which failure is worst for you
| Your app | Most exposed to | What to do |
|---|---|---|
| Internal tool, a few users | Little; the AI stops, you wait | Nothing urgent |
| Public app with logins on Lovable Cloud | Backend pause at zero credits | Keep a credit buffer, set low-balance alerts, or move the backend to your own Supabase |
| Base44 app that sends email or calls an LLM | Integration credits running out | Watch the integration pool separately from message credits (Base44 vs Lovable) |
| App with AI features for end users on any builder | Runtime AI calls drawing on your credits | Price the per-user cost before launch |
| Editor or agent user (Cursor, Copilot, Claude Code) | A surprise on-demand bill | Set a budget or spending cap; see Cursor pricing and Claude Code pricing |
The exit route matters here as much as the price: an app whose backend is portable is not held hostage by a credit balance. Our lock-in and export matrix compares that across builders.
Method
Researched from vendor documentation and pricing pages on 29 September 2026; the Lovable outage is one third-party report, not a measurement. We have not deliberately run any account to zero, so behaviour not in the docs is marked “not stated”. Vendors change credit rules often; this page is due for re-check every 30 days.
Common questions
Does my Lovable site go down when credits run out?
The published pages keep serving, according to Lovable’s docs. If the app uses Lovable Cloud for login or data, that backend can pause until credits return, so a logged-in app can appear down.
Can I still edit my code when a builder runs out of credits?
Bolt’s docs say direct edits in Code view are free and use no tokens. Lovable’s docs do not say whether manual code edits remain possible.
Do credits carry over?
Sometimes. Bolt paid tokens and Lovable monthly credits last about two months, v0 monthly credits 65 days, Replit Pro credits two months. Base44 credits expire each cycle.