Skip to main content
Most AppBlips issues fall into a handful of categories — installation, running from source or Docker, preview rendering, and API connectivity. Work through the section that matches your symptom to find a fix.

​
Windows installer is blocked

AppBlips is distributed on Windows only through the Microsoft Store now. The Store version is signed by Microsoft, so Windows doesn’t block it, and the Store installs updates for you.
Older versions of AppBlips offered a .exe installer on the GitHub Releases page. That installer wasn’t code-signed, so Windows 11’s Smart App Control could block it or its updates, and it no longer receives new versions.Install AppBlips from the Microsoft Store to get a signed install that keeps itself up to date. Your existing projects aren’t affected: they’re saved in Documents/AppBlips/Projects, and the new install keeps using that folder automatically.
The Linux AppImage and .deb don’t use Smart App Control and are never affected.

​
Running from source or Docker

Another program is using port 3000.
  • From source: start AppBlips on another port. On macOS or Linux run PORT=3001 npm start; on Windows (PowerShell) run $env:PORT=3001; npm start. Then open http://localhost:3001.
  • Docker: add APPBLIPS_PORT=3001 to a .env file in the AppBlips folder and run docker compose up -d again. See Use another port.
Projects are saved per address, so the new port starts with an empty project list. Export them from Settings → Data at the old address first if you need them.
AppBlips only answers when you open it as localhost or 127.0.0.1. Open http://localhost:3000 instead. If you deliberately use another name (for example behind a reverse proxy), add it to APPBLIPS_ALLOWED_HOSTS in .env. See the Configuration reference.
No AI provider is set up. Open Settings → AI, choose a provider, enter your API key and choose a model, then click Save. (If you run AppBlips from source or Docker, you can set a provider in .env instead; see Configuration.)
The browser version keeps projects in your browser, which has a size limit. Delete apps you no longer need, or use the desktop app, which saves projects as folders and has no such limit.
The browser version saves projects in your browser, separately for each address. Check that you opened the same address as before (same port, localhost rather than 127.0.0.1), in the same browser and not a private window. Clearing your browser’s site data deletes them for good, so back up with Settings → Data → Export now and then.
Check your Node.js version with node --version. AppBlips needs 22.12 or newer; install the LTS version from nodejs.org. Then delete the node_modules folder and run npm install again.If the error mentions Could not download or does not match its pinned sha256, the build couldn’t fetch the voice-input speech model from Hugging Face. Check your internet connection and run npm start again. It only downloads the model once.To get the model on its own, for example before building on a computer that will be offline, run node scripts/fetch-whisper-model.js in the AppBlips folder. It saves the model in public/whisper/, and later builds use that copy.
Ubuntu 24.04 and some other distributions need Electron’s sandbox helper to be owned by the system. In the AppBlips folder, run:
Then run npm run desktop again. Don’t use --no-sandbox instead.
On Linux, the local server has to accept connections from the container. Start Ollama with OLLAMA_HOST=0.0.0.0, or turn on Serve on Local Network in LM Studio. If it still fails, a firewall may block the container. See Local models.

​
Preview shows a blank screen

AppBlips needs your permission to use the microphone for voice input.Windows:
  1. Open Start → Settings → Privacy & security → Microphone.
  2. Turn on Microphone access and Let apps access your microphone.
  3. Turn on Let desktop apps access your microphone (or, for the Microsoft Store version, the switch next to AppBlips).
  4. Close AppBlips and open it again.
Browser version: click the icon to the left of the address bar, set Microphone to Allow, then reload the page.If it says No microphone was found., check that one is plugged in and selected as the default input in your system’s sound settings. If it says The microphone is in use by another app., close the other app that’s recording (such as a video call) and try again.
AppBlips recorded you but couldn’t turn it into text.
  1. Check that the right microphone is the default input in your system’s sound settings.
  2. Speak a little closer to the microphone, and record for at least a second or two.
  3. Click the microphone key and try again.
If it keeps failing, close AppBlips and open it again, then see Find the error details.
A blank preview after a successful generation usually means the generated code contains a syntax or runtime error that prevents it from rendering. AppBlips catches most of these and fixes them on its own, and runs a full review if the quick fixes don’t work.If the preview is still blank:
  1. In the AppBlips prompt input, type a follow-up like The preview is blank. Fix any JavaScript errors and let AppBlips correct the code.
  2. Click the camera key next to the prompt box to attach a screenshot of the preview, so the AI can see what you see.
  3. If the problem persists, step back with Previous version and try rephrasing your prompt with more specific instructions about the behavior you want.
The Code tab in the preview toolbar lets you inspect the raw generated HTML, which can help you spot issues at a glance. If you spot the problem, click Edit to fix it yourself. If you don’t see the tab, turn Code view back on in Settings.

​
API and build errors

A message such as “Anthropic rejected the API key in Settings → AI” means your AI provider didn’t accept the key. Other API errors during a build almost always point to a wrong or invalid provider setting too.Open Settings → AI and check that:
  • An API key is filled in, and it is valid and active for your provider
  • The model is a name your provider recognizes and your key has access to
Click Test connection to check the settings.Also confirm that your API key has remaining quota and has not been revoked in your provider’s dashboard.
AppBlips couldn’t connect to your AI provider at all. Check that:
  • Your computer is online.
  • If you use LM Studio or Ollama, its local server is running. Start it again if you closed it.
  • The Local server URL in Settings → AI is right, or empty to use the usual address.
If the build spinner keeps going without producing output, the request likely timed out or the AI provider is responding slowly. When no data arrives for 3 minutes, AppBlips stops waiting and shows “Stream stalled: no data received for 180s”.Try the following:
  1. Click Cancel next to the prompt box and try again.
  2. Check your AI provider’s status page for any ongoing outages or elevated latency.
  3. Try a lower Reasoning level in Settings → AI (Off, where your model offers it). Reasoning takes longer and is more likely to hit timeouts on slower providers.
Every AI reply has a length limit. When a reply hits it, AppBlips asks the AI to continue where it stopped, up to three times. If the app is still incomplete, the reply says so.Try the following:
  1. Open Settings → AI and look at Output tokens: building. If you typed a number there, clear the field to go back to the default of 64,000.
  2. Ask for the rest in a follow-up prompt, for example Finish the parts that are missing.
  3. For a very large app, build it in smaller steps: start with the main screen, then add features one prompt at a time.
  4. Some models have a low limit of their own. If this keeps happening, choose a different model in Settings → AI.
AppBlips sets no request limit of its own. An error such as “OpenRouter turned the request down: …” comes from your AI provider, and the text after the colon is the provider’s reason. Common causes:
  • No credit left on your provider account. Many providers report this as a rate limit (HTTP 429).
  • The provider’s own rate or concurrency limits. Wait and try again, or check your plan’s limits.
Claude’s safety checks turned the request down. This sometimes happens by mistake with ordinary requests. Reword your prompt, or choose another model in Settings → AI.

​
Find the error details

When something fails without a clear message, AppBlips writes the details to its log. Include those lines when you report a problem.
  1. In AppBlips, press Ctrl+Alt+Shift+I. The developer tools open next to the app.
  2. Click the Console tab.
  3. Try the action that failed again.
  4. Copy the yellow and red lines that start with [AppBlips].
If you started the app with npm run desktop, the same warnings and errors also appear in that terminal.
  1. In the browser tab with AppBlips, press F12 (on a Mac, Cmd+Option+J).
  2. Click the Console tab.
  3. Try the action that failed again.
  4. Copy the yellow and red lines that start with [AppBlips].
Messages from the AI connection appear where AppBlips runs instead: the terminal where you ran npm start or npm run dev, or docker compose logs -f for Docker. Lines starting with [anthropic] show what each Claude request asked for and what came back. In the [anthropic] done: lines, cache_read is how many tokens came from Anthropic’s prompt cache (billed at a reduced price), and cache_write is how many were stored for the next step. During a build, cache_read should be above 0 from the second step on.
The logs never contain your API key. They can include the first few lines of an AI reply, so check what you paste before sharing it publicly.
For general questions about how AppBlips works, visit the FAQ.