Taking a Bolt.new app to production means moving it from a project that lives in Bolt's browser workspace to a codebase with an owned database, server-side secrets, a Git history, automated tests and a staging environment that every change passes through. Keep the screens and flows that users already like, harden the data layer and deploy path in place, and rewrite only the pieces that cannot be fixed. For Bolt, success is decided by one question: what can an anonymous visitor do when they talk to your database directly, because Bolt's own docs say they can.
Bolt is built for speed of iteration. The development environment runs on StackBlitz WebContainers, technology that runs the whole dev environment inside your browser, so there is no laptop setup and no server to provision while you build. Around that, Bolt Cloud bundles hosting, domains, a database, auth, storage and payments into the same product.
The pieces that matter for a production review:
That is a strong setup for a demo, a pilot or a first paying cohort. Production asks different questions. A customer's security questionnaire wants to know who can read which rows, where keys live and who can rotate them. Finance wants to know a failed charge never grants access. Your on-call engineer wants to know what changed in the last deploy and how to roll it back without opening a chat with an agent. None of those answers come from the builder; they come from the repository and the pipeline around it.
Bolt is unusually direct about this: its security page says "Anyone who visits your site can send requests to your database directly". The public key in the bundle is meant to be there, so the only thing standing between a stranger and your rows is row-level security. The common gaps are a table created during a late refactor with RLS never switched on, or a policy that checks that a user is signed in but not that the row is theirs.
Detect: run Bolt's built-in database security audit, which is available on all plans and flags missing RLS and over-open policies. Then go further than the audit: sign up two test accounts, open the browser devtools on account A, copy a request, swap in an id that belongs to account B and replay it. If it returns data, you have broken object level authorization (OWASP API1:2023).
Fix: every policy should compare the row's owner or organization column to the caller's identity, for reads and for writes separately. Write the policies as migration files in the repo so they are reviewed like code.
Because the .env file is part of the project, a payments key or an LLM provider key added there during a fast session can end up in the JavaScript that ships to visitors. Bolt's docs say a browser-read value "ends up in what's sent to anyone visiting your published site." If GitHub sync was on at the time, the same value is also in commit history.
Detect: search the built bundle and the full git log for key prefixes from each provider you use. Turn on GitHub secret scanning for the repo and read every alert.
Fix: rotate any exposed key first, then move the call that uses it into a server function and store the value as a Bolt secret. Enable push protection so the next one is blocked at push time.
When the project has never been connected to GitHub, the real code lives in a browser-run environment. There is no pull request to review, no diff to blame and no pipeline that refuses a broken build. Each publish goes from the builder to users.
Detect: ask who can tell you what changed between yesterday's site and today's. If the answer is "scroll the chat," this is your issue.
Fix: connect GitHub, protect the main branch, and from that point treat Bolt as one contributor whose changes arrive on a branch and merge after review and green tests.
A Bolt Database is convenient, but a security review will ask about backups, point-in-time restore, access logs and who holds admin credentials. Claiming the database into your own Supabase account puts those controls in your hands and opens the Supabase tooling (Security, Performance and Health Advisors). Bolt also notes that free-plan Supabase databases pause after a week of low activity, even when the site is published, which is fine for a prototype and wrong for a customer-facing app.
Detect: list which account owns the database, who has admin access and when you last tested a restore.
Fix: claim the database on a paid plan, put it in a team-owned organization, and run a restore drill before you need one.
Generated apps tend to fetch a list and then fetch each item's details in a loop (the N+1 pattern), and RLS adds a filter to every query. With twenty rows nobody notices. With twenty thousand, the dashboard times out.
Detect: after claiming into Supabase, open the Performance Advisor and read every finding. Check every column referenced in a policy for an index.
Fix: Supabase recommends you index every column a policy filters on and wrap auth.uid() in a select so it is evaluated once per statement. Collapse the per-row fetches into one query with a join or a view (remembering that, per Supabase, views bypass RLS by default).
Bolt Cloud includes payments, which makes it easy to ship a paywall. The production risk is a paywall that only exists in the UI: a flag on the user's profile row that the client sets, or a webhook handler that does not verify signatures or handle the same event twice.
Detect: as a free user, try to update your own plan column through the API. Replay a webhook payload twice and see whether you get two grants.
Fix: entitlements are written only by a server function that verifies the webhook signature and records each event id before acting on it. Users cannot write the columns that decide what they paid for.
The builder shows you errors while you are in it. Once real users arrive, the errors happen while you are asleep.
Detect: break something on staging on purpose and see how long it takes for a human to find out.
Fix: client and server error tracking, uptime checks on the main routes and the server functions, and an alert that reaches a named person.
Bolt keeps working throughout. What changes is the order of operations: lock the data first, then make it fast, then put a pipeline between the builder and your users.
Stop prompting new features for a few days. If the project isn't on GitHub yet, connect it now so the browser workspace becomes a commit history, and grab a desktop download as a spare copy. Export schema plus data. List each value in .env, each Bolt secret by name and each outside API the app talks to.
Ready to move on when a repo you own holds a tag for the live version and the key inventory is written down.
Close every finding from Bolt's audit, then run the two-account replay against all data calls. Rewrite policies around ownership, adding an organization column where accounts share data. Route sensitive provider calls through server functions with Bolt secrets, rotate whatever leaked, enable scanning. On auth: verification emails, expiring reset links, sane session length, OAuth redirects limited to your own domains, and no profile column a user can flip to become an admin.
Ready to move on when replays fail closed and no live credential remains in the bundle or the log.
Claim the database into a Supabase account your company controls. Record schema and policies as migration files. Work through the Advisors, index every policy column, rewrite the per-row fetch loops on the busiest views, and check that your client library works with the pooler's transaction mode, which Supabase says lacks prepared statements.
Ready to move on when a blank database can be rebuilt from migrations and a timed restore has succeeded.
Pin current behavior with tests on account creation, login, the main create and edit flows, checkout and plan changes. GitHub Actions (or your CI of choice) runs them for each pull request. A staging copy with a separate database sits between Bolt's branch and users. Pick the publish target here too: Bolt hosting suits an app still using Bolt's backend; a team running its backend in its own accounts can choose freely.
Ready to move on when merging needs review plus green checks, and reverting a bad release is one written step.
Instrument the client and server functions for errors, ping the key routes for uptime, log auth and payment events in a structured format, and write short playbooks for a paused or unreachable database, a payment provider incident and a leaked key. Put a name on the pager.
Ready to move on when a drill failure wakes the right person and the playbook got them through it.
Long-running jobs, heavy reporting or an integration that needs a persistent process may not fit server functions. Lift just those into services in your own cloud; the Bolt-built app stays.
Finished when each lifted service has an owner, tests and an independent deploy.
The AI pod leans on coding assistants for scaffolding tests, first drafts of RLS policies, bulk refactors and sifting logs. A person reviews each diff, and nothing merges until the suite passes.
An investor demo or an internal tool for five people does not need engineers brought in. Paying customers, a security questionnaire or money moving through the app are the signals that it does.
Bolt work tends to call for a tech lead to set the plan and review every merge, one to three senior full-stack people at home in Postgres, row-level security and the generated front end, a part-time security engineer and DevOps engineer, plus QA automation if billing is intricate. Many teams begin with a pod of 3 to 5 senior engineers (a tech lead plus engineers), and a full pod is about seven. Smaller rescues can start with a tech lead plus one or two engineers.
Ryz engineers on your team start from $7,000 per month (our published range is $7,000–$15,000 per engineer per month, mostly mid-level to senior), quoted per team. Budget it as headcount × months × monthly rate. Once you say go, engineers can typically start in under two weeks. Ryz engineers work on your team, reporting to your leads. Scope your Bolt team.
Yes, provided the database policies, key handling, query performance and release process are brought up to standard. The generated interface is rarely the bottleneck; an unindexed policy or an untested deploy usually is.
Seldom. Typical Bolt fixes are ownership-based RLS, keys moved behind server functions, a claimed database, tests and CI. Starting over is justified only when the tables themselves model the business wrongly, and the current app still serves as the spec.
That depends on table and role count, whether payments or multiple organizations are involved, whether the code already syncs to GitHub, how many audit findings appear and how quickly your reviewers approve changes. We scope after the audit rather than guessing first.
Engineers times months times each engineer's monthly rate. Quotes are per team, and you meet the named people and see the plan and price before anything begins.
Yes. With GitHub connected, Bolt's edits land on a branch and pass the same review, tests and staging as anyone else's. Plenty of teams keep prototyping screens in Bolt while engineers own data, security and deploys.
Questions we didn't answer? Email info@ryzlabs.com.
Thanks — your message has been sent. We’ll get back to you soon.
Something went wrong while sending your message. Please try again or email info@ryzlabs.com.