Skip to main content
AppBlips turns a plain-English description into a working, interactive project in one step. Before writing your prompt, you choose a studio: App (single-file interactive tools), Website (multi-page websites with navigation, plus carousels, photo galleries, animations or maps when the site needs them), or Game (browser games built on plain Canvas, the Phaser engine, or Three.js for 3D, with keyboard and touch controls). Type what you want in the prompt box, press build, and watch your project appear in the live preview alongside it — no project setup, no files to juggle, no build step to run. Once it’s built, you can keep refining it with follow-up prompts until it’s exactly what you had in mind.

​
Writing Your First Prompt

Start by describing what you want in ordinary language. You don’t need to mention React, HTML, or any technical details — just describe the app as you would to a friend. A few examples to get you started:
The more specific you are, the better the result. Name the features you want, describe how they should behave, and mention any constraints (like “keep it to a single screen” or “use a dark background”). Vague prompts produce generic apps; detailed prompts produce apps that match your vision.

​
Prompt Techniques

List the individual things your app needs to do rather than describing it at a high level. Instead of “a budget app,” try “a monthly budget tracker where I can add income and expense items with a category and amount, and see a running total at the bottom.” The AI builds exactly what you describe, so covering the details up front saves revision rounds.
If you have a look in mind, say so. Mention colors (“a dark navy background with white text”), layout (“two columns side by side”), or a general aesthetic (“minimal and clean, like a notes app”). You can also reference familiar design patterns: “a card-based layout” or “a sidebar navigation.”
If your app works with a list, a table, or structured data, describe that structure in your prompt. For example: “a task list where each item has a title, a due date, a priority (low / medium / high), and a done checkbox.” The more you define the shape of the data, the more accurate the generated forms and displays will be.
Don’t assume AppBlips will add interactive behavior unless you ask for it. Be explicit: “clicking a card should expand it to show the full description,” or “the list should be sortable by clicking the column headers.” Describing the user actions you expect produces apps that actually behave the way you want.

​
Starter ideas

Not sure what to build? A new project shows ready-made ideas for the studio you’re in, under the prompt box (in the build pane they’re headed Or start with one of these). Click one to put its prompt in the prompt box, then change it however you like before you send it. Each studio has dozens of ideas, and they’re shuffled every time you open the app. To see more:
  • Click the arrow keys beside the ideas (in the build pane they sit next to a page number, like 1 / 11), or
  • Drag the ideas sideways with your mouse, or swipe them on a touchscreen or trackpad.

​
Attaching Images

You can drop a screenshot, mockup, or reference image directly into the prompt area alongside your description. AppBlips will use the image as a visual reference when generating your app — useful when you want to recreate an existing layout, match a design you’ve sketched, or base your app on something you’ve seen elsewhere. To attach an image, drag and drop it onto the build pane or use the image attachment button next to the prompt input. You can also paste a copied image or screenshot straight into the prompt box with Ctrl+V (Cmd+V on a Mac). Once your app is built, the camera key next to it attaches a screenshot of the current preview, which is handy for pointing out what looks wrong. You can combine an image with a written prompt: for example, attach a screenshot of a form and write “build this, but replace the file upload field with a text area.”

​
Using your own images (beta)

My images is a beta feature, so it may change and some things may not work smoothly yet. To put your own photos, logo or pictures into what you build, add them to the project’s image gallery. You can do this before the first build or at any time while editing.
  1. Open Settings from the header, go to Tools and click the My images card. Your project’s images are listed on that page.
  2. Click Upload images and pick one or more pictures. You can also drag them onto the upload box or paste them with Ctrl+V (Cmd+V on a Mac). JPEG, PNG, WebP, GIF and SVG work, up to 25 images per project. Big photos are made smaller (2000 pixels on the longest side) to keep your project light.
  3. Optionally, change an image’s name or type a short note about what it shows, such as “our team at the summer party”.
  4. Close Settings and ask for what you want. To use a specific image, mention its name, the one shown in the Name box under it. Click Copy name above the box to copy it, then paste it into your request, for example “put team-photo in the hero section”. To use them all, say so: “use all my images in a gallery on the about page”. The exact name is safest, but a close one also works: “the team picture” finds team-photo.jpg. If no image matches, or two match equally well, the AI doesn’t guess: its reply says which image it couldn’t find and lists the names you can use.
Before it builds anything with your images, the AI shows which image it matched to your words, for example “the mountain lake photo” → mountain-lake-sunset.jpg. Click Yes, use it (or Yes, use them for several) to start the build. If it picked the wrong one, type the right name instead, such as “no, use tropical-beach”, and it checks again. Nothing changes in your project until you approve. Changing a picture that’s already on the page, for example asking the AI from click-to-edit to round its corners or make it bigger, doesn’t ask first; only swapping in a different image from My images does. Don’t have your own photos? Set up the Pixabay connector, then click Find free photos on Pixabay on the My images page to search millions of free photos and add the ones you pick. The AI doesn’t see the pictures themselves. It gets each image’s name, size and your note, so a clear name or note helps it pick the right one. An image only appears where you ask for it. Images marked In use are on one of your pages; if you remove one, that spot shows a broken image until you ask the AI to change it. Renaming an image updates your pages automatically. To keep the AI from using an image without deleting it, click the eye on it: it is dimmed and marked Hidden from the AI. The Enable My images switch on its page in Settings → Tools does the same for all images at once. Hidden images still show on pages that already use them. The images are saved with the project and go with it when you export or publish. They aren’t included in a JSON backup, so a backup restored on another computer lists the images without the pictures. My images isn’t available for imported sites, which keep their own image files.

​
Speaking your prompt

On the start screen of a new project, you can say your idea instead of typing it:
  1. Click the microphone key next to the prompt box (its tooltip says Speak your prompt).
  2. If AppBlips asks to use your microphone, click Allow. See Microphone permission below.
  3. While the key pulses red and Listening… shows, describe what you want to build.
  4. Click the microphone key again to stop (its tooltip now says Stop voice input). Transcribing… shows for a few seconds, then your words appear in the prompt box.
  5. Check the text, fix anything it misheard, and send it as usual.
If you’d already typed something, your spoken words are added after it. Typing in the prompt box while AppBlips is listening or transcribing cancels the recording. A recording stops by itself after 2 minutes. Voice input understands most languages. It expects your computer’s language first, and works out the language itself if that one isn’t supported.

​
How your voice is turned into text

AppBlips uses Whisper, a speech-recognition model, which comes built into AppBlips. A model is the trained “brain” that turns sound into words. AppBlips includes the base size of Whisper (about 76 MB), made by OpenAI and released as open source. It runs on your own computer:
  • Your recording is never uploaded, saved or sent to your AI provider. Only the text you send becomes part of your prompt.
  • Voice input works without an internet connection. AppBlips never downloads a speech model while it runs.
  • The first time you use it after opening AppBlips, it takes a few extra seconds to load the model. After that it’s quicker.
  • Transcribing uses your computer’s processor, so a long recording on an older computer can take a little while.
  • The base model is small enough to run on most computers, but it can mishear unusual names, brand names and technical words. Check the text before you send it.
Whisper’s license is listed in Settings → About → Licenses, under Other credits.

​
Microphone permission

  • Desktop app (installer from GitHub, or Linux): AppBlips doesn’t ask. Clicking the microphone key is your permission. Windows still needs its microphone privacy switches turned on, which they are by default.
  • Microsoft Store version: the first time, Windows asks whether AppBlips may use your microphone. Click Yes.
  • Browser version: your browser asks the first time. Click Allow.
Only AppBlips’ own screen can use the microphone. The apps you build, running in the preview, can’t. If you see Microphone access is blocked, see Troubleshooting.

​
Refining with Follow-up Prompts

After your app is built, you don’t need to start over to make changes. Type a follow-up prompt describing what you want to adjust and AppBlips will update the app accordingly. Each refinement creates a new version in your history, so you can always go back if a change doesn’t land the way you expected. Follow-up prompts work the same way as your first prompt — plain English, no code required:
You can chain as many refinements as you like. Each one picks up from the current version of the app. To copy a message, yours or the AI’s, click the copy button in the bottom-right corner of its bubble. It turns into a tick once the text is on your clipboard. The AI’s replies are copied as plain text with their Markdown formatting (like **bold**). To take your project back to how it was at one of your earlier messages, click the rewind button (the curved arrow) just left of the copy button in that message’s bubble. It only appears on messages that came before the current version. To start a fresh conversation without losing anything, click New chat in the header. Your versions stay in the history.

​
Finishing Setup

When a build is done, the AI tells you in the chat what it built or changed. Sometimes part of your app can’t work until you add something yourself. For example, a weather app may need an API key (a password-like code from the weather service that lets your app use it). When that happens, the AI’s message ends with To finish setting it up: and numbered steps that say where to get what’s needed and how to add it. Usually the last step is to paste the value into the chat and ask the AI to add it.
A key you put into your app is visible to anyone who opens the app or the exported file. Use a free key, or one the service lets you restrict, never one tied to a paid account you can’t limit.

​
Asking Questions (Ask Mode)

The key next to the prompt box switches between Build and Ask. In Build mode, your message changes the app. In Ask mode, the AI only answers: ask how something works, what a piece of code does, or what to try next, and your app stays as it is. Click the key again to go back to Build. You don’t have to switch modes for a quick question. If you ask something in Build mode that only needs an answer, like “how does the score get saved?”, the AI answers it without changing your app. Requests phrased as questions, like “can you add a dark mode?”, still get built.

​
Reasoning

AppBlips can let the AI think before it writes code. In Settings → AI you can set Reasoning to any level your provider and model support, from Off up to Max (see the full list): It defaults to Low. Choose the lowest level for the fastest builds, or a higher one for apps with intricate logic or multi-step flows. Higher reasoning takes longer and uses more tokens, on every build and edit.

​
How much did that use?

AI providers measure and bill your use in tokens. A token is a small piece of text, roughly three-quarters of a word. One build is often several AI requests (planning, editing, fixing errors, writing the reply), and each one sends your message, your app’s code, the conversation so far and AppBlips’ instructions, then counts everything the AI writes back, including any thinking. AppBlips doesn’t show a count in the chat. To see your usage:
  • Your own API key: open the usage or billing page on your AI provider’s website. It shows exactly what you were charged.
  • LM Studio or Ollama: these run on your computer and cost nothing.

​
Game engines

In the Game studio, AppBlips picks an engine for each new game before it starts building. A quick extra request asks the AI which engine suits your idea: To choose yourself, name the engine in your prompt, for example “make it in Three.js” or “plain canvas, no engine”. Your choice always wins.

​
See what was picked

The choice appears in the chat, in the AI’s reply for that build. It comes right after the AI’s first sentence and before the game is written:
I’ll build a side-scrolling platformer with coins and enemies. Engine: Phaser 3, a game engine for 2D games with sprites, levels and physics. It adds about 1 MB to download.
The message is saved with that version, so you can still see it later in the chat history. You may also see:
  • “Engine: the one you asked for in your request.” You named an engine yourself.
  • “Engine: the engine picker didn’t answer, so the builder chooses as it writes.” The extra request failed. Your game is still built.
The engine is picked once, for the first build only. Later edits keep the engine the game already uses, and no message appears. To let the builder decide as it writes instead, turn off Smart game engine in Settings → AI. No message appears then either.

​
Website add-ons

In the Website studio, AppBlips picks ready-made libraries (“add-ons”) for each new website before it starts building. A quick extra request asks the AI which add-ons suit your site: A site gets at most three, and only what it clearly needs. A photography portfolio might get GLightbox and AOS. A simple one-page site gets none, because every add-on is an extra download. To choose yourself, name the library in your prompt, for example “use Swiper for the reviews” or “no libraries”. Your choice always wins.

​
See what was picked

The choice appears in the chat, in the AI’s reply for that build. It comes right after the AI’s first sentence and before the site is written:
I’ll build a pet care website with services, a photo gallery and booking. Add-ons: Swiper (carousels), GLightbox (full-screen photos).
The message is saved with that version, so you can still see it later in the chat history. You may also see:
  • “Add-ons: none. This site doesn’t need any, so it loads faster.”
  • “Add-ons: none. The add-on picker didn’t answer, so the site uses plain JavaScript.” The extra request failed. Your site is still built.
Add-ons are picked once, for the first build only, and no message appears for later edits. When an edit asks for something an add-on is made for, such as “add a photo carousel”, the AI can add that add-on then. Extra pages use the same add-ons as the landing page when they need them. To build new websites with plain JavaScript only, turn off Smart website add-ons in Settings → AI. No message appears then.

​
Clarifying Questions

AppBlips can ask you a short round of clarifying questions before generating your app, helping it nail the details that matter most to you. Toggle this on or off in Settings → AI with the Clarifying questions switch. When the toggle is on and your prompt leaves room for interpretation, AppBlips will pause before building and ask a few targeted questions — things like what data the app should work with, how a particular interaction should behave, or what the layout should prioritize. Answer them and AppBlips will use your answers to produce a more accurate first build. If you prefer to just hit go and refine afterward, turn the toggle off.

​
Reviewing a Build

Code Review is a beta feature and is off by default. To try it, turn on Code Review in Settings → AI. Once it’s on, after each build or edit in Build mode, AppBlips asks you to try it in the preview first and say whether anything looks wrong on screen. Choose Looks wrong and the AI opens your project in the built-in browser, checks each page’s layout at mobile, tablet and desktop sizes, clicks through it, and reads the code. It doesn’t change anything while it looks. Screenshots use a lot of tokens, so it takes one only when it has found or suspects a specific problem and a picture would confirm it or show it to you, scrolled to the part of the page in question. The preview blinks like a camera shutter each time it takes a screenshot. When it’s done, it writes a report in the chat listing each problem it found, what it saw and what should change. Problems it photographed come with their screenshot under the report (click one to see it full size); problems found by testing or in the code, like a button that does nothing, a code error or a broken link, are listed without one. The report ends by saying how many screenshots it took, if any. Screenshots need an AI model that can read images: with a text-only model, the report says so, and the AI checks the page’s text and layout without seeing it. Choose Looks fine if it looks right but you’d like the AI to check that it works. The AI clicks through your project in the built-in browser and reads the code, but takes no screenshots, which uses far fewer tokens. It then writes the same kind of report. If the report lists problems, AppBlips asks whether you want them fixed. Choose Fix problems and AppBlips clears the chat and starts a new one with the report as its first message, and the AI fixes those problems as a normal edit. Choose Not now to keep things as they are; you can still ask for changes yourself. Starting fresh keeps the AI focused on the report instead of everything said before. The earlier chat, with the report, stays in your version history. The fixes are saved as a new version, so you can undo them. If the report finds nothing, nothing else happens. A fix that comes from a review isn’t offered another review, so reviews don’t repeat endlessly. Choose Skip if you’re happy with the result. A review can take many AI requests, so it uses extra tokens. To stop being asked, turn off Code Review in Settings → AI. AppBlips also checks every build for code errors and fixes them on its own. If those quick fixes don’t work and Code Review is on, it runs one full review automatically, once per build. Code errors aren’t something you can see, so this review tests your project without taking screenshots. If that review finds problems, it asks whether to fix them as described above. With Code Review off, it doesn’t review or take screenshots: it tells you the problem couldn’t be fixed automatically.

​
Long Apps

Each AI reply has a length limit, set in Settings → AI → Output tokens: building (64,000 tokens by default, enough for a large app). If a reply is cut off at the limit, AppBlips asks the AI to carry on where it stopped, up to three times, and tells you if the app is still incomplete afterwards. See The app was cut off if that happens.
Once your app is built, you can also edit it visually by pointing and clicking on any element. See Editing Apps to learn how.