Contents
An app built with Bolt or Cursor can be recovered: Bolt lets you download the project or sync it with GitHub, and with Cursor the code already sits in your own folder. A takeover starts with an audit (from $660 (€589)excl. VAT); a rebuild is priced like a business app, from $6,800 to $22,400 (€6,000–€20,000)excl. VAT.
Bolt and Cursor share one feature that changes everything for a takeover: they produce real code. Unlike a closed no-code tool, nothing forces you to rebuild from scratch. But they work very differently, and the risks are not in the same place. Here is what each tool lets you recover, the weaknesses I often find in AI-generated apps, and how to decide between consolidating and rebuilding.
What each tool lets you recover
The information below comes from each tool’s official documentation, checked in October 2026.
| Item | Bolt | Cursor |
|---|---|---|
| What the tool is | In-browser app generator that hosts the project | Code editor installed on a computer, working in a local folder |
| Code | Archive download or GitHub sync | Already on your machine: you open a project folder |
| Data | Built-in database: export tables as CSV or JSON; Supabase database: in your Supabase account | Depends on what the project uses (Supabase, another database) |
| Moving the database | Into your own Supabase account, on Pro or Teams plans (advanced settings) | Not applicable: the database is already with a provider you chose |
| Main point to watch | Getting code and data off the platform without missing anything | Finding the access: database, keys, hosting, often in one person’s account |
Two details. For Bolt, the documentation says that opening projects in StackBlitz has been phased out since April 2026 for projects moved to the new storage format: download and GitHub remain the ways out. For Cursor, its data use page explains that file contents may be sent to its servers for AI features and describes a privacy mode (data use): worth knowing if the code contains sensitive information.
Common weaknesses in generated code
AI-generated code often works well on screen. The problems lie in what you cannot see, as I explain in the guide on building an app yourself with AI:
- Checks done only in the browser. A curious user can call the API directly and bypass the screen.
- Secret keys in the code, or even in the Git repository, where they stay in the history even after deletion.
- Missing or overly broad database access rules.
- No tests, so every fix risks breaking something else.
- Outdated or unnecessary dependencies, added prompt after prompt.
- No tested backup, and no separation between test and production.
The page on the most common security flaws covers these risks in detail.
Consolidate or rebuild
With real code, consolidation is often possible. The audit tells you whether it is reasonable.
- Consolidate if the structure is coherent: clean up, secure access, add tests on critical flows, set up two environments (test and production) and backups.
- Rebuild one part if a core module (payments, permissions, calculations) is too fragile: redo it properly and keep the rest.
- Rebuild everything if every fix breaks something else: the prototype becomes the requirements.
My usual stack (React, TypeScript, Node.js, Supabase) often matches what these tools generate, which makes the takeover easier.
What a takeover costs
| Step | Price | Timeline |
|---|---|---|
| Express audit | $660 (€589)excl. VAT | A few days |
| Full audit | $2,300 (€1,990)excl. VAT | 5 to 7 days |
| Consolidation or partial rebuild | Quoted after the audit | Depends on the action plan |
| Rebuild as a business app | $6,800 to $22,400 (€6,000–€20,000)excl. VAT, 0.85 factor in the estimator with a prototype | 1 to 3 months |
| You keep using Bolt or Cursor, with support | Coaching at $170 (€150)excl. VAT per hour | At your own pace |
For a range that fits, the estimator takes the existing prototype into account. AI subscriptions and credits remain your cost, as does hosting.
When not to take it over
I will be straightforward: an internal tool built with Bolt or Cursor, used by a small team, with no sensitive data or payments, does not need a full takeover. Put the code in a company Git repository, get the keys out of the code, set up backups and carry on. A takeover makes sense when the tool opens up to customers, handles personal data or money, or when nobody but its creator can develop it further.
Where to start
- Gather the access: code repository, database, hosting, domain name, third-party services. Check that they belong to the company.
- Export the code and data and keep them somewhere safe.
- Ask for an audit: it is the starting point of my AI project takeover offer, with a costed action plan.
- Then choose between consolidating, rebuilding one part, or carrying on yourself with support.
Frequently asked questions
What is the difference between Bolt and Cursor for a takeover?
Bolt is an in-browser generator that hosts the project and often the database, so code and data must be taken off the platform. Cursor is a code editor installed on a computer: the code is already in a folder, sometimes in a Git repository. A Cursor takeover starts with finding where the database, keys and hosting live.
Can data in a Bolt database be recovered?
According to Bolt’s documentation, table data in the built-in database can be exported as CSV or JSON, and the database can be claimed into your own Supabase account on Pro or Teams plans. If the project was already connected to Supabase, the data is in your Supabase account.
Can a developer take over AI-generated code?
Yes, it is standard code, often React and TypeScript. The difficulty is not the technology but the build-up: duplicate functions, abandoned pieces, incomplete security rules. The audit measures that gap before you commit.
Should we stop using AI after the takeover?
No. I work with AI tools every day myself. The difference is a clean Git repository, reviews before each production release, and someone who understands what is being shipped. Hourly coaching is there for that if you want to keep building yourself.