How to build your own AI chatbot app (no code)

“Build me an AI assistant like ChatGPT” is one of the first things people try on App Builder Agent — and a chat app is a great first project, because the hard part (the model) is an API call away and the interesting part is the experience around it: personas, memory, voice, a focused use case.

This guide walks through a chatbot app that you can actually ship: a clean chat screen, conversation history, a few personas, and a settings screen for the API key. Along the way it covers the one thing most tutorials skip — where the key lives and why that matters.

What an AI chat app really needs

A chat screen alone is a demo. These are the pieces that turn it into something people keep on their phone:

• A message list that feels right — Bubbles, timestamps, streaming replies token by token, and a composer that grows with the text.
• Conversation history — Saved on the device: a list of past chats with titles, delete and rename. Users come back to old answers.
• Personas or modes — “Study tutor”, “Email writer”, “Travel planner” — each with its own system prompt. This is where your app stops being a generic clone.
• A clear place for the API key — A settings screen where the user pastes their own key (OpenAI, OpenRouter, Gemini…). Keys typed into an app are stored on that phone only.
• Error states — Out of credits, no network, key rejected — each needs a friendly message, not a spinner.
• Copy, share, regenerate — Small buttons under each answer that people use constantly.

Build it with App Builder Agent, step by step

You will describe the app, watch it run, and refine it. The agent handles the networking code and the streaming.

1. Start a mobile app project. Pick “Mobile app” so you can export an APK; a website version works the same if you prefer a link.
2. Describe the assistant and its personas. Name the model provider you intend to use, and be explicit that the user supplies their own key.
   Build an AI assistant app called Sable. Chat screen with streaming replies, a sidebar with saved conversations, and 4 personas: General, Study tutor, Email writer, Recipe helper. Settings screen where the user pastes their own OpenAI API key (stored on the device). Dark theme, purple accent.
3. Test with your own key. Open Settings in the simulator, paste a key, and chat. The agent will fix anything that fails; tell it exactly what you saw. Try a wrong key and a switched-off network in the Simulate panel as well — the error messages you see there are the ones your users will see.
4. Add the features that make it yours. Voice input, image upload, a daily tip, a “favourite answers” list — one message each.
   Add a microphone button that transcribes speech into the composer, and a long-press on any answer to save it to a Favourites tab.
5. Export. Build the APK and install it. Because the key is entered by each user on their phone, you can share the app freely.

Prompts that work well

These prompts produce noticeably different apps — pick the one closest to your idea and change the details.

Common mistakes (and the fix)

• The key in the prompt. Anything pasted into the chat can end up in the app’s source, and an APK can be unpacked in minutes. Enter keys only in the app’s settings screen (stored on that phone) or keep them on a hosted backend.
• Sending the whole history every time. A 200-message conversation costs more per turn and eventually exceeds the model’s context. Ask for a rolling window (last 20 messages) plus a short summary of older ones, and a visible “New chat” button.
• No streaming. Waiting 15 seconds for a full answer feels broken. Say “stream the reply token by token” and ask for a stop button; the agent wires the streaming API for you.
• One giant persona. A single system prompt that tries to be tutor, lawyer and chef gives mushy answers. Give each persona a real, short system prompt and a starter question, and let users create their own with a name and instructions.
• Ignoring failure states. Out of credits, rate limited, provider down — each should be a plain sentence with the fix (“Add credit at your provider” / “Try again in a minute”), not a spinner that never ends. Cache the last answers on the device so history still opens offline.
• Unlimited use on your own key. If the app ships with a shared key and no quota, one bored user can spend your month’s budget in an evening. Put the key on a hosted backend with a per-user daily limit and a simple sign-in.

Make it feel finished

• Stream the replies. Waiting 20 seconds for a wall of text feels broken; words appearing one by one feels alive.
• Give every persona a real system prompt — the agent will write them if you describe the tone you want.
• Keep the free tier honest: if the app uses your key, add a daily limit or you will pay for strangers’ chats.
• Add a one-tap “New chat” button; users otherwise keep polluting one giant conversation.

Installing, publishing and next steps

The APK export gives you an installable app in minutes: Download APK in the Preview tab produces a release-signed test build, and a signed store build (APK and AAB) with your own package id and version is ready when you want the Play Store. If each user brings their own key, there is nothing to run on a server and nothing to pay for; if you want a shared key so users don’t need one, ask the agent to deploy a small backend with the key kept server-side and a daily quota per user — the platform hosts that backend with the project. The AI app builder page lists what the free first app includes and what the $1.49 Member plan and plans add. For a web version, publish the project to a link; the same chat UI works in any browser.

Browse the ready-made AI assistant apps to see how others handled personas and settings. If you want the assistant to listen and talk back, the voice assistant app guide adds speech in and out; if your users are students, the AI study helper guide turns the same chat screen into flashcards and explanations. Open the builder and name your four personas.

Frequently asked