How to build an AI-powered app for free

Build it with

The stack

Every pick has a free plan you can launch a business on, unless it says otherwise. Free tiers checked 2026-10-27.

Start from open source

Worth studying before you prompt, or self-host one instead of building from scratch.

UI kit and libraries

AI skills and MCP servers

Guides and connections that make your AI builder produce better code.

• Lovable: Prompting best practicesService

  Plan the product and key user action first, build UI one component at a time, use real copy and style keywords, ask questions in Plan mode.

• Bolt: Prompt effectivelyService

  Plan the app and first prompt, use Enhance prompt, and make small one-at-a-time changes.

• v0: How to prompt v0 + Vibecoding GuideService

  Describe the product surface, context of use, and constraints and taste (2025-12-15); the Academy guide adds patterns and starter prompts.

• AGENTS.mdMIT

  Open 'README for agents’ format stewarded by the Agentic AI Foundation (Linux Foundation); used by 60k+ open-source projects.

  Add an AGENTS.md file at the repository root

The plan

1. 1

   Write down the plan

   30 minutes
   • Name the people who use it and the one job it does for them.
   • List the screens: A chat or form UI that streams the answer; A server-side call to the model, with the system prompt kept on the server; Upload documents and answer from them (find the relevant chunks first).
   • Sketch them on paper or in Excalidraw. Rough is fine.
   • Download this kit as AGENTS.md and keep it in your project: it’s the brief your AI tool reads first.
2. 2

   Set up the foundation

   Day 1
   • Start a new Lovable project and paste the prompt below.
   • Connect Supabase from Lovable’s integrations, then turn on email and Google login.
   • Publish it and open it on your phone to check it works.
   • Connect the project to GitHub so you own the code.
   Prompt for Lovable

   Build an AI-powered app: an AI-powered app (chat with your PDFs)
   The main screens: A chat or form UI that streams the answer; A server-side call to the model, with the system prompt kept on the server; Upload documents and answer from them (find the relevant chunks first).
   Use Supabase for the database and login (email and Google). Turn on row-level security on every table so people only see their own data.
   Use shadcn/ui components, a clean light design, and make every screen work on a phone.
   Don’t add payments yet. Start with sample data so I can click through it.

3. 3

   Build the core

   Days 2–7
   • A chat or form UI that streams the answer
   • A server-side call to the model, with the system prompt kept on the server
   • Upload documents and answer from them (find the relevant chunks first)
   • Usage limits per user and a paid plan for more
   • Save and reopen past conversations
   Prompt for Lovable

   Now build these, one at a time, and let me test each before the next:
   1. A chat or form UI that streams the answer
   2. A server-side call to the model, with the system prompt kept on the server
   3. Upload documents and answer from them (find the relevant chunks first)
   4. Usage limits per user and a paid plan for more
   5. Save and reopen past conversations
   For AI features, call Google Gemini API from the server only, never from the browser, and limit how often each user can call it.

4. 4

   Take payments and send emails

   Week 2
   • Add Stripe checkout.
   • Mark orders as paid only from the payment webhook, never from the browser.
   • Send emails through Resend, and add its DNS records (SPF and DKIM) so they don’t land in spam.
   Prompt for Lovable

   Add payments with Stripe. Use test mode first.
   Create a webhook handler on the server that verifies the signature and marks orders paid; never trust the browser for this.
   Send receipts and account emails with Resend from a server function.

5. 5

   Get it ready to launch

   Week 2–3
   • Add PostHog and track the one action that means it’s working (a booking, a sale, a sign-up).
   • Add Sentry so errors reach you before users complain.
   • Go through the launch checklist below.
   • Add a privacy policy and terms, then connect your own domain.
   • Show it to ten people who have the problem, and watch them use it.
   Prompt for Lovable

   Review the whole app for security and launch problems before I go live:
   - every table has row-level security (or server-side checks) and I can’t read another user’s data,
   - no secret keys in the browser bundle or the repo,
   - inputs are validated on the server, and sign-up, login and forms are rate-limited,
   - error pages and empty states exist for every screen,
   - it works on a phone and with the keyboard.
   List what you find, fix the serious ones, and tell me what’s left.

Before you launch

• Row-level security (or server checks) on every Supabase table, tested with two accounts.
• No secret keys in browser code or in git. Rotate any that leaked.
• Server-side validation for every form, plus rate limits on sign-up and login.
• Payments confirmed by webhook only, and tested end to end in test mode.
• A per-user limit and a monthly spending cap on the AI API.
• Your sending domain verified (SPF, DKIM), so emails reach the inbox.
• Backups: know how to restore your data (free database plans often have none).
• Privacy policy, terms, and a way for people to contact you.
• Error alerts on, and an uptime check on the home page.
• Tested on a phone, with a keyboard only, and with a slow connection.

Risks to plan for

• Model costs grow with usage: cap usage per user and price above your cost.
• Wrong answers: show sources and let users report problems.
• Keep API keys on the server and rate-limit to stop abuse.