AI builder no longer gets stuck mid-generation on tool errors
The AI that builds your app will no longer incorrectly declare itself "done" when it hit an unresolved issue mid-way through — it now catches those situations and handles them properly before finishing.
Fewer half-finished app generations that looked complete but weren't
The builder now correctly finishes or flags issues rather than quietly stopping early
More consistent, complete results from a single generation run
Turn off desktop notifications and they actually turn off everywhere
Disabling desktop notifications from the account settings now correctly turns them off across all your devices — not just the one you used to toggle the setting.
One toggle now silences notifications on your laptop, phone, and any other signed-in device
No more stray notifications pinging on a second device after you thought you'd turned them off
When you preview your own app as a builder or team member, you can now switch to a true end-user view — so permission and access rules are tested accurately, not bypassed.
Catch permission issues that were previously invisible during your own testing
Preview the exact experience a real user sees, not an admin-level override
Build with confidence that what you test is what your users get
Deploying updates to your app now takes roughly 10–12 minutes instead of 26, thanks to smarter behind-the-scenes preparation that kicks in the moment a change is detected.
Ship updates to your users in half the time
Faster deploys with no change to reliability or safety
Failed deploys caused by temporary glitches now self-recover
If a deployment fails in the very first moments due to a temporary network or cache hiccup — not anything wrong with your app — OverSkill now automatically retries it for you instead of leaving it broken.
Transient deploy failures resolve themselves without you needing to do anything
Genuine build problems are still surfaced clearly so nothing is hidden
Fewer surprise "deploy failed" notifications for issues that weren't your fault
Cancelling a queued message now actually cancels it
Removing a message from the build queue in the editor used to make it disappear visually but still get processed — it would run anyway. Now cancelling a queued message truly stops it.
Queued messages you remove won't silently continue running in the background
Your build queue reflects reality — what you see is what will run
Avoids wasted AI credits on builds you already cancelled
Deploy progress bar now survives closing and reopening the publish panel
If you closed the publish panel mid-deploy and reopened it, the progress bar would reset to blank. Now it restores exactly where it left off, and the styling is cleaner too.
Reopen the publish panel mid-deploy and pick up right where you left off
No more wondering if your deploy is still running after closing the panel
Progress bar now matches the rest of the editor's visual style
Published apps refresh automatically after a new deploy
Users with your app already open in their browser will now see a prompt to reload and get the latest version after you publish an update — instead of staying stuck on the old one.
Your users always get the freshest version of your app without a hard refresh
A clear update prompt appears so users know a new version is ready
Works reliably across all browsers without any setup on your part
Publish button dot clears correctly after a successful publish
The 'unpublished changes' indicator on the Publish button could stay lit even after a fully successful publish, making it look like something still needed attention. That's now fixed.
Publish button accurately reflects whether you have unpublished changes
No more false alarms after a clean, successful publish
Editor state stays in sync with what's actually live
The OverSkill AI builder now gives helpful plain-language answers when you ask questions like 'can you build X?' or 'where do I find Y?' — instead of getting stuck or going silent.
Ask capability questions before you start building and get a real answer
Navigation and advice questions now get conversational responses
No more cancelled turns or silent failures when asking non-build questions
Failed build steps shown in gray, not alarming red
Tool steps that fail during a build were showing up in red, which made it look like you needed to take action. They're now shown in a neutral gray — red is reserved for things that actually need your attention.
Easier to tell at a glance what needs your input vs. what the AI is handling
Red and amber colors now only appear for genuine action items
When you upload your own brand logo, the AI editor can now see and use it — so you can ask it to place your logo in the header (or anywhere else) without it getting confused or failing repeatedly.
Ask the AI to add your logo to any part of your app and it will actually find it
No more repeated failed attempts or wasted credits when working with uploaded brand assets
Works with logos you've uploaded yourself, not just AI-generated ones
Your support replies now handle retention conversations automatically
When a customer replies to a personal outreach email about their subscription, your support inbox picks it up and routes it intelligently — no manual triage needed. The whole conversation stays in one place.
Replies to outreach emails automatically reopen the right conversation thread
Your AI support agent responds on the same path as all other tickets — consistent, no gaps
Cancelling or dormant customers who write back get handled without you lifting a finger
Follow-ups panel header looks clean and stays on one line
The follow-ups section now displays its icon, label, and navigation arrows all neatly on a single full-width row — no more icon jumping to its own line above the text.
Follow-ups header no longer wraps awkwardly across multiple lines
Navigation arrows consistently appear on the right, icon and label on the left
Cleaner, more readable layout when reviewing AI-generated follow-up suggestions
AI model costs now fairly reflect what you actually use
Credits are now calculated proportionally to the real cost of each AI model, so you're never overcharged for using a lighter model or surprised by a heavier one. Every model now has a transparent, consistent credit rate.
Cheaper AI models now cost fewer credits — your budget goes further when you don't need the most powerful option
Credit costs are consistent and predictable across all available models
No surprise overages — what you see is what you spend