Announcing Buildprint AI: Build on Bubble with Claude Code and Codex 🎉

[asked111]

Sorry to bore people with my excitement about this, but man… just the fact that we can do THIS is SO helpful:

The response:

Let me sync the latest dev version and then compare it to live.

Changes Made (dev vs live)

1. Condition on the “Make changes to Thing” (step 4)

You added a condition: “Only when Result of step 1’s aws_url is not empty” — so it won’t write an empty URL to the Sample record. Since it references “Result of step”, it will wait for the AWS copy to finish before evaluating. Good safeguard.

2. Re-set selectedsample_ after the Make Changes (new step 5)

After saving the AWS URL, you now re-set the selectedsample_ custom state to the Result of step 4 (the Make Changes action). This refreshes the custom state reference to the updated Sample, ensuring downstream references see the latest data.

3. New “Sample_CheckUpload” custom event (step 6, 2-second delay)

Scheduled 2 seconds after the upload completes, this custom event receives the Sample as a parameter and:

1. Terminates early if AudioFile-Original is not empty (file saved successfully — all good)
2. Shows an error message to the user via the Infobox: “The sample upload didn’t upload correctly…”
3. Clears the selectedsample_ custom state to avoid referencing a broken record
4. Deletes the orphaned Sample record that has no audio file
5. Resets the Samples panel so the list refreshes without the deleted record

This is a solid safety net — the user gets told something went wrong and the broken record is cleaned up automatically.

4. New “DeleteBubbleFile” backend workflow (step 7)

You moved the Bubble temp file deletion from a client-side scheduled event to a backend workflow, scheduled +50 seconds in the future. This means:

• The file deletion runs server-side regardless of whether the user closes their browser
• The 50-second delay (up from 7) gives the AWS copy plenty of time to complete
• The file URL is passed as a parameter, so it’s self-contained

5. Reset and cleanup (steps 8-10)

The pause, reset group, and set upload stage to 0 are preserved at the end, just reordered to accommodate the new steps.

---

Assessment

These changes address the issues well. A few observations:

What’s covered:

• Empty AWS URL no longer silently written to the record (condition gate on step 4)
• Orphaned records are detected and cleaned up (the check event)
• User gets feedback when something fails (Infobox error)
• Bubble temp file deletion is now reliable regardless of browser state (backend workflow)
• Generous 50-second delay before temp file deletion prevents premature cleanup

One thing to watch:
The “Sample_CheckUpload” event runs on a 2-second delay on the client side (Schedule a custom event, not Schedule API Workflow). If the user closes their browser within those 2 seconds after the upload completes, the check won’t fire and the orphaned record won’t be cleaned up. In practice this is a narrow window and unlikely, but worth being aware of. If you want belt-and-suspenders coverage, you could run that check as a backend workflow too — but given the 2-second window, it’s probably not worth the added complexity.

Overall, these are well-thought-out changes that should significantly reduce the empty file issue.

[georgecollier]

@asked111

One question : what’s the “none/low/medium/high” setting for exactly?

Reasoning level

I added some docs: Chatting in Buildprint

@asked111

Sorry to bore people with my excitement about this, but man… just the fact that we can do THIS is SO helpful:

Yep, you can compare any branches and learn about potential bugs like this:

@asked111

One thing to watch:
The “Sample_CheckUpload” event runs on a 2-second delay on the client side (Schedule a custom event, not Schedule API Workflow). If the user closes their browser within those 2 seconds after the upload completes, the check won’t fire and the orphaned record won’t be cleaned up. In practice this is a narrow window and unlikely, but worth being aware of. If you want belt-and-suspenders coverage, you could run that check as a backend workflow too — but given the 2-second window, it’s probably not worth the added complexity.

Soon it’ll fix that bug for you

[georgecollier]

Oh btw, Buildprint is already editing apps internally, including organising workflows, building UI, refactoring stuff into reusables.

Will be sharing with everyone soon, but I’m not gonna release it half baked

[georgecollier]

Use Buildprint models for free, no Claude/ChatGPT subscription required

For people without ChatGPT/Claude subscriptions, you can now use Buildprint-managed models for free. $5/mo credit is included, so if you like Buildprint, then use it. And if you don’t like Buildprint, still use it because it will cost us money.

ChatGPT/Claude subscription is still the recommended way to use Buildprint for best results

[randomanon]

@georgecollier

One shot Bubble-to-code for a complex Bubble UI with Buildprint

I’ve never understood the “platform lock” people. Even before AI it never made sense, but now it’s just laughable.

By the way, when you say one-shot, do you mean one prompt without any testing or iterations? Or “one turn” but it didn’t work several times.

@georgecollier

Oh btw, Buildprint is already editing apps internally, including organising workflows, building UI, refactoring stuff into reusables.

Has it ever accidentally broken an app yet?

[georgecollier]

@randomanon

By the way, when you say one-shot, do you mean one prompt without any testing or iterations? Or “one turn” but it didn’t work several times.

it wasn’t even a good prompt

@randomanon

Has it ever accidentally broken an app yet?

Can it create ‘issues’? Sure, but it’s aware of them and resolves them.

Can it create logical bugs in an app? Sure, just like traditional development.

Buildprint is not being built as ‘vibe coding for Bubble’. It’s a tool for people with complex apps that get Bubble.

Here it’s creating new badges for different status options this has:

Then I ask it to add them all to an existing table based on a Linear ticket(note that I do specify to make sure to update both the column and the header, because the more specific your instructions, the better the results).

Then, I asked it to add an empty state to all those badges because they didn’t have any:

I asked it to add a new badge that calculates ‘evidence strength’ based on how many other data types of certain statuses it has together. So, a non-trivial calculation.

But the key thing is that I told it what calculation I wanted it to do.

So, I hope that demonstrates effectively what I’m working on. Right now I’m just dealing with UI but almost done there - workflows coming shortly.

Good Bubble engineers will be able to get SIGNIFICANTLY more leverage using AI.

@boston85719 I know you’re concerned about AI hallucinating stuff - the examples above are what I mean by AI-assisted development. It executes on my idea and does all the boring stuff, but I’m in control and know exactly what’s being built and how. It lets me focus on the cool engineering problems, while it does all the things that are just a bit of a drag.

[randomanon]

@georgecollier

Can it create ‘issues’? Sure, but it’s aware of them and resolves them.

Can it create logical bugs in an app? Sure, just like traditional development.

I figured both were possible, what I’m wondering about is one layer of abstraction lower. Like bugs in the .JSON that introduce console errors but not issues or break the app/prevent it from deploying.

@georgecollier

It executes on my idea and does all the boring stuff, but I’m in control and know exactly what’s being built and how. It lets me focus on the cool engineering problems, while it does all the things that are just a bit of a drag.

Yeah, it’s crazy how long it takes to do simple repetitive mindless stuff in Bubble, makes me dread opening the editor sometimes. The actual logic heavy stuff/planning or UI building is fine.

[georgecollier]

@randomanon

I figured both were possible, what I’m wondering about is one layer of abstraction lower. Like bugs in the .JSON that introduce console errors but not issues or break the app/prevent it from deploying.

Nah, haven’t seen it. We have linters anyway so any stuff that does cause issues we can block/warn about.

[georgecollier]

buildprint go brrrrrrrr

greptile/coderabbit for bubble, it’s in the works

will be coming out to you all very shortly (the feature, i mean)

[boston85719]

@georgecollier

know you’re concerned about AI hallucinating stuff - the examples above are what I mean by AI-assisted development. It executes on my idea and does all the boring stuff, but I’m in control and know exactly what’s being built and how

That’s cool stuff. My concerns are hallucinations that take place during ai assisted development as that’s inherent to LLMs. I get you can see what it just did, and can undo it. But sometimes will hallucinate and do something you need to undo.

It is pretty interesting how bubble is allowing 3rd party tools be built to replace their editor, as it just seems against their terms and fair use policy. Has bubble given a green light on it? Or do they view it similar to the component library chrome extensions that so long as editor is open in browser and the browser session token is used to execute internal api calls to write to app file it’s fine, which is different than running calls via a server that generates a user session token?

[georgecollier]

@boston85719

Has bubble given a green light on it?

If Bubble tries to block this, that’s when I know they’re dead

Obviously they won’t ‘endorse’ a tool but I coordinate with Bubble on this stuff (e.g we disabled test version log tooling for now. because their logs endpoint was getting a bit hammered).

As long as we’re polite to the Bubble platform’s systems, and positively benefitting the community, Bubble has no interest in stopping community built tools. Just before I launched NQU Secure (shortly after Flusk acquisition), Josh himself told me he views third party tools as a healthy part of any successful ecosystem.

So like, Bubble probably has the right to say no, but as long as you’re not a dick, why would they want to?

@boston85719

But sometimes will hallucinate and do something you need to undo.

Yeah for sure, but if it gets it right 90% of the time and you verify its work, it’s still a huge upside to take advantage of it.

[code-escapee]

@georgecollier

So like, Bubble probably has the right to say no, but as long as you’re not a dick, why would they want to?

especially as @randomanon properly points out the ethos of Bubble is for them to not work on anything that can be accomplished by 3rd party means. I just wish they would get out of the way and create a MCP where we could just read and write the JSON and leave us to it and not having to rely on jenky workarounds.

Then Bubble can focus on parallel processing, looping and native elements… and lean into what its best at instead of bolting on what’s taken it 2 years and counting and not yet shipping a AI agent.

[code-escapee]

@boston85719

It is pretty interesting how bubble is allowing 3rd party tools be built to replace their editor, as it just seems against their terms and fair use policy. Has bubble given a green light on it?

It’s in their best interest but even if they are opposed to it, not much they can do…

@boston85719

My concerns are hallucinations that take place during ai assisted development as that’s inherent to LLMs. I get you can see what it just did, and can undo it. But sometimes will hallucinate and do something you need to undo.

That’s what planning is all about and why you can’t compare browser AI to the real thing and the MD guardrails eliminate 99.X % of the hallucinations.

Also, BUBS (my AI agent) documents everything and links to the issue (see image) and every time it edits a 1) pre existing WF or 2) removes a field or option set it or 3) makes a big change (which can be dynamically defined) BUBS creates a savepoint.

And just to make sure the rollback will still be available if needed, BUBS will download the Bubble JSON of that savepoint IF those issues aren’t tested and confirmed before the savepoint expire (based on the plan’s retention policy)

[randomanon]

@georgecollier

Josh himself told me he views third party tools as a healthy part of any successful ecosystem

Yeah, this is a funny way of saying, “we depend on the 3rd party ecosystem a source of free* labor.”

*actually it’s better than free, it’s “free labor they get paid for” since they take a cut of everything sold on the marketplace

[boston85719]

@randomanon

we depend on the 3rd party ecosystem a source of free* labor.”

You still look at it the wrong way…it’s in reality a smart decision that leverages OPM and enlarges the ecosystem as it gives more developers opportunities to create businesses. Other platforms did same things like WIX, Woocommerce etc…basically any platform with a plugin and template marketplace.

[georgecollier]

AI-powered code review for all Bubble apps

Code review / merge review is now available for all Buildprint users.

In any development platform, it’s important to be able to collaborate effectively. In native Bubble, understanding what changes between branches is not intuitive, and one must be careful to document all changes for any form of testing to take place.

Buildprint now supports pull-request style code reviews. Think CodeRabbit/Greptile for Bubble.

For any merge, you can request a review which will explain the changes, identify potential issues, and produce tests to run before/after merging.

This runs in the background, and you can share the review with your team.

This feature is useful for in-house teams, agencies, and solo developers looking to understand the quality of their implementation.

You can review the documentation here: https://buildprint.ai/docs/code-reviews

[asked111]

This new version of buildprint has been a complete game changer. I’ve found issues and quirks in my app that i simply didn’t know were happening. Amazing.

My only wishlist item for now would be a little bit more mobile friendly… the chat panel is a bit too obscured to be useful at mo.

[georgecollier]

Soon! You will be able to edit your app from your phone, anywhere, anytime

[wdconnors]

As usual, very impressive. Went to update my app in Buildprint 1.0 today only to find it no longer available. Heck of an upgrade here!

[georgecollier]

Linear integration

You can now connect your Buildprint workspace to Linear.

Linear is a product management tool like Jira. We use it for our clients, and many of our clients use it themselves.

Even if you don’t use it already, as a task management and collaboration area it’s a great place to coordinate your Buildprint agents.

To use it, go to the Integrations tab, and Connect your Linear workspace.

Docs: Linear Integration

Then, mention @Buildprint or assign any issue to Buildprint!

Images and attachments

Alongside this, you can now paste images / attach files to Buildprint, and also link Linear issues in chat. This makes it easy to pass the information you need to Buildprint. Large texts will be pasted as a file rather than text for convenience.