What's new.

Use your Claude subscription instead of an API key

If you already pay for Claude, you can now connect that subscription directly rather than buying API credits separately. Add it in My Profile → Build. As with any key you bring yourself, those messages don't count against your daily allowance.

Kimi joins the model lineup

Kimi is now selectable in the chat, bringing the list to six: Claude, GPT, Gemini, DeepSeek, GLM and Kimi. You can switch between them mid-project without losing your work — the new model picks up the conversation and the code where the last one left off.

Interrupted tasks now pick themselves back up

When we restart the console for a deployment, any task running at that moment used to leave the page sitting with a blinking cursor forever. Now the interruption is detected and the task is re-run automatically, once, with a clear message explaining what happened.

We also stopped charging your quota for that re-run. It was our restart, not your request.

Live streaming, without opening a cloud account

Apps can now publish and play a live video stream using shared streaming infrastructure that's already running. Ask for it and the AI wires it up.

A deliberate limit worth stating plainly: this is sized for demos and a handful of viewers. Live video is bandwidth-hungry, and if your stream finds a real audience you'll need to move to a dedicated provider. The point is that you can prove the idea works before spending a cent on that.

Your Android signing key and GitHub key are now downloadable

We generate an Android upload key for every app that needs one, because creating it by hand involves command-line tooling most people never touch. Until now there was nowhere to actually get it.

My Profile now has Android and GitHub tabs. The Android tab gives you the keystore as a zip including its passwords — you need both halves if you ever publish outside Appzentic. Keep it somewhere safe: an upload key can't be replaced on an existing Play listing.

GLM added, and Codex is now called GPT

GLM is available as a coding model. Separately, we renamed Codex to GPT in the interface — more people recognise the name, and nothing changed underneath.

Password reset

Forgetting your password used to be a dead end. There's now a reset link on the login page.

Apps can send email

Verification codes, notifications, scheduled reports — your app can now send email through shared infrastructure, with no provider account or domain setup on your side.

Tighter isolation for scheduled jobs and stored credentials

Scheduled jobs now run inside the same locked-down sandbox as everything else rather than directly on the host. The code browser also no longer displays real credential files — you see example configs instead, so secrets can't leak through a shared screen or an exported project.

Usage credits, and email verification at signup

Invite someone and you both earn usage credits, which are spent before your plan's own quota. Signup now confirms your email with a code first — which is also what makes password reset possible.

Voice and video calling

Apps that need real-time calls now get the relay infrastructure that makes them actually connect. Without it, calls fail for a large share of users on home and mobile networks — it's the unglamorous piece that decides whether a call app works at all.

See how your app is actually doing

A metrics panel now shows what's happening with a deployed app — traffic, response codes, memory and disk. Mobile apps can opt in to usage analytics covering opens, active users by platform, and next-day return rate. For web apps, there's a performance check that scores real page-load quality.

Analytics is off unless you turn it on. Your app, your call.

Notifications and a feedback channel

Announcements now arrive in My Profile → Notifications, with anything urgent shown directly. You can also report problems from your profile and attach a screenshot — we reply on the same thread, so you're not hunting through email for an answer.

Every app runs in its own sandbox

Apps built on the platform run isolated from each other, with their own memory and CPU ceilings. One app misbehaving can't reach another app's files or drag down the platform for everyone else.

React and Vue for web apps

Web apps were limited to plain HTML, CSS and JavaScript, which capped how much you could build. React and Vue are now supported, so more ambitious interfaces are possible — and the resulting code is standard enough that any web developer can pick it up later.

More backend languages, and a third database

• Backends can now be built in Go or Java (Spring Boot), alongside Python — the AI picks what suits your app, or uses whatever you ask for
• Redis is available for caching, sessions and counters, joining MySQL and MongoDB
• MySQL and MongoDB were upgraded to current versions, unlocking modern query features

Build together in one conversation

You can now invite someone into a project — a colleague, your boss, a client. Everyone sees the same chat, code and build results as they happen, and can talk to the AI directly.

Actions with consequences stay with you: only the owner can build, deploy, connect databases, export code or delete the project. Use @ to direct a message at a specific model, or at a person when you want to talk among yourselves without the AI joining in.

Export your code, or push it straight to GitHub

Download the complete source as a zip, or connect a GitHub account and sync your project to a repository you own. Projects also get a README and CHANGELOG generated automatically, so the repo looks presentable instead of like machine output.

Credentials are stripped from exports. You get your code, not our secrets.

iOS testing moved to direct install

iOS test builds now install directly on devices you've registered, via a QR code, instead of going through TestFlight.

The reason: TestFlight makes every build wait for Apple's review and requires you to have already created the app record in App Store Connect. For someone trying to see their idea running on their own phone, that's a discouraging first experience. Direct install is ready in minutes. If you'd rather sign under your own Apple developer account, you can now add your credentials and builds will use your team instead of ours.

Building and fixing without leaving the chat

Builds used to happen in a separate panel you had to go watch. Now the whole cycle runs in the conversation: you ask, the build starts, progress appears as a message, and the result comes back with install links.

If a build fails, the log is read and classified automatically and the problem is handed back to the AI as if you'd reported it yourself — so in many cases the fix and the rebuild happen before you've finished reading what went wrong.

Install an iOS build by scanning a QR code

Getting a test build onto an iPhone was the roughest edge on the platform. Apple requires every device to be registered before it can run an app from outside the App Store, and collecting that device ID normally means plugging into a Mac.

Now you scan a QR code, tap once, and your device registers itself. From then on every iOS build produces a code you scan to install — no cable, no Mac, no store.

Apps that run on a schedule

Not everything needs a screen. You can now build something that just runs — scrape a page every morning, total yesterday's orders, email a weekly summary — with no interface at all. Describe the job and when it should happen.

Shared storage and CDN — the start of "no cloud account required"

Apps could now store uploaded files and serve them fast worldwide, using infrastructure already provisioned and waiting. This was the first piece of what became the platform's shared services: the answer to "I want to build this, but first I have to go open three cloud accounts."

Everything that followed — databases, email, calling, live streaming — is the same idea extended.

Real databases for your app, and a way to look inside them

Apps got their own dedicated database — created, credentialed and wired up automatically when the AI decides your app needs one. MySQL and MongoDB to begin with.

A browser landed alongside it, so you could actually see what your app had stored. Being able to look at your own data turns out to matter enormously for trusting that the thing works.

The first real mobile builds

The moment the whole thing was pointed at. Android and iOS build pipelines came together within a day of each other, and for the first time a conversation produced an actual installable app — signed, packaged, and ready to put on a phone.

Everything before this was a chat interface that wrote code. This is where it became a way to make apps.

One-click deploy, with versions that track themselves

Publishing your backend and web app became a single action instead of a sequence of server steps. Every deploy and build bumps a version number automatically, so you can always tell which build is on your phone and which one is live.

Console v2 — conversations became projects

The biggest structural change in the platform's life. Until now a chat was just a chat. From here, the moment the AI writes code, the conversation becomes a project: it has a name, a file tree you can browse, a build history and a deployment target.

A code browser arrived with it, so you could read everything the AI had written — not because you need to, but because being able to look is the difference between owning something and renting it.

"AI Builds. You Own."

The line that ended up defining the product showed up on the login screen. It has since become AI Builds, Your Own. — but the commitment behind it hasn't moved: whatever gets made here belongs to the person who asked for it.

Start on your laptop, keep going on your phone

If the AI was mid-answer and you switched devices, the second device used to show nothing at all. Now the work in progress follows you — open the same project anywhere and you see the response still arriving, live, rather than a blank screen and a guess about whether anything is happening.

Show, don't describe

File attachments and screenshot paste. Instead of explaining what looked wrong, you could paste a picture of it — which remains the single fastest way to get a precise fix out of the AI.

iOS and Android apps for the console itself

Two days after the first working version, the console got native mobile apps — so you could start an app from a phone, on a sofa, without a laptop anywhere nearby.

They were built the same way everything here is built, which made them the first honest test of whether the idea held up.

Day one

The first working version: a chat window, three AI models wired up and answering, accounts, and a workspace per person that nobody else could reach.

No builds, no deploys, no databases. Just the bet that describing what you want should be enough — and a lot of unglamorous work getting three different command-line tools to stream their output into the same box.