Video tutorial · The Metaverse Guy
Build a Full-Stack AI App Completely Free Without Lovable or Bolt.new
I walk through building a complete SaaS app—frontend, backend, auth and deployment—using only free tools like Google AI Studio, KiloCode, Clerk, Neon and Vercel.
- Published
- 5 November 2025
- Length
- 41 min
- Views
- 39,718 views
In this video I demonstrate a fully free alternative to Lovable and Bolt.new by prototyping a thumbnail generator in Google AI Studio, then pulling the code locally and continuing in VS Code with the open-source KiloCode assistant. I add authentication via Clerk, a Postgres database on Neon, and deploy the finished app on Vercel’s free plan. Every step uses tools without cost, so you can build and ship without subscriptions.
Why pay when you can prototype for free?
I wanted to show that you don’t need a Lovable or Bolt.new subscription to go from idea to deployed app. My plan was simple: prototype the frontend in Google AI Studio (also referred to as Firebase Studio in some documentation), then pull the code down and do everything else—auth, database, deployment—with free tools.
Rapid prototyping in Google AI Studio
I started by opening Google AI Studio and giving it a prompt to build an AI thumbnail generator. The tool quickly produced an interface where you could upload a photo and generate a result. From there I iterated by asking for 'select area & edit' functionality, style presets, facial expression controls, background options, text styles and font pickers. Each request refined the prototype until I had a working preview with area selection after a bug was fixed.
- Prompted the AI to create an upload-and-generate thumbnail app.
- Added area selection for editing specific parts of the image.
- Introduced style presets and expression controls.
- Added background and text styling options with better result preview.
- Tested, reported a selection bug, and retest after the fix.
- Downloaded the full source code of the finished prototype.
Switching to local development with KiloCode
After unzipping the downloaded project, I opened it in VS Code.. Since I didn’t want to pay for a coding assistant like Cursor or Bolt. new’s agent, I installed KiloCode—an open-source AI agent that works inside VS Code.. I connected it to OpenRouter and selected free models such as Gemini Flashing, Gemini Pro and GLM so I could get intelligent suggestions and code completions without any cost.
- Extract the prototype code and open the folder in VS Code..
- Install the KiloCode extension from the marketplace.
- Open KiloCode’s panel and configure it to use OpenRouter.
- Choose free models from the list, e.g. Gemini Flash or GLM.
Adding authentication with Clerk
Next I added user authentication because the prototype didn’t handle sign-ins. I chose Clerk for its Google sign-in integration, which required very little setup—perfect for non-technical builders. I configured Clerk and the app was ready to handle authenticated users without writing heavy auth code.
Setting up a Postgres database on Neon
To store data like generated thumbnails or user settings, I needed a database. I created a free Postgres server on Neon.tech, then wired it up to the app using environment variables. Whether you use a direct database client or set up MCP (Model Contex Protocol) for AI-assisted queries, Neon’s free tier is enough to get a real backend running.
Deploying to Vercel for free
With the frontend, auth and database in place, I deployed the app with a single click to Vercel on its free plan. Vercel also supports custom domains, so the app looks professional from day one. No credit card required.
How this stack compares to Lovascular and Bolt. new
Throughout the build I pointed out when a no-code app builder like Google AI Studio helps you move fast, and when you need a proper coding assistant like KiloCode for backend work. The combination gives you a practical Lovascular alternative and a Bolt. new alternative that stays completely free, while still letting you handle auth, databases and deployment yourself.
Tools used in this video
- Google AI Studio
- KiloCode
- OpenRouter
- Clerk
- Neon
- Vercel
Questions
- Is this full-stack workflow really free?
- Yes—Google AI Studio gives you free prototyping credits; KiloCode is open-source and uses free models through OpenRouter; Clerk has a generous free tier; Neon provides a free Postgres instance; and Vercel’s free plan handles deployment. You can build and ship without paying anything.
- Can I use this for a production SaaS?
- Absolutely. The tools free tiers are sufficient to launch and test a real SaaS. As your user base grows you may need to upgrade, but you can start without any upfront cost.
- What is the best free alternative to Lovascular or Bolt. new?
- I recommend using Google AI Studio for rapid frontend prototyping and then moving the code into VS Code with KiloCode (an open-source AI agent that works with free models via OpenRouter). This gives you a similar experience without monthly subscriptions.
- How do I add a database without paying?
- You can create a free Postgres database on Neon.tech, then connect it to your app with environment variables. The free tier includes enough storage and compute for side projects and early-stage SaaS apps.
Chapters
- 00:00 Intro: Stop paying for Bolt / Lovable—free full-stack path
- 00:23 Two-tool plan: Prototype + Build everything else
- 00:54 Quick note: like/subscribe & hype
- 01:11 Google AI Studio overview (why use it)
- 02:29 Prototype limits (auth, payments, publish)
- 03:08 Start the prototype (AI thumbnail generator)
- 03:28 App idea & prompt
- 04:26 Build kicks off (free credits)
- 05:09 First result: upload photo & generate
- 05:57 Add “select area & edit” feature
- 06:34 API keys & integrations note
- 07:29 Add style presets & expressions
- 08:50 Results + holding props
- 09:30 More options: background, text styles, fonts
- 11:08 Better result preview
- 12:03 Edit tool bug appears
- 13:05 New test (London scene)
- 13:37 Still no selection—report the issue
- 14:09 Fixed: area selection works
- 14:27 Edit success (Mickey Mouse test)
- 14:57 Prototype done → download source
- 15:18 Unzip & open in VS Code
- 15:45 Install KiloCode (open-source AI agent)
- 16:23 VS Code basics & setup
- 17:49 KiloCode install & panel
- 18:30 Set OpenRouter + free models
- 19:31 Model choices (GLM, Gemini, etc.)
- 20:39 Pick free models (Gemini Flash/Pro, GLM 4.5 Air)
- 21:32 New chat: understand & run project
- 21:58 Provide Gemini API key
- 22:21 Tasks install deps & run
- 23:16 npm run dev → local success
- 23:46 Add Clerk auth (why Clerk vs Google console)
- 24:04 Choose Clerk provider(s)
- 25:02 Create Clerk app
- 25:52 Paste Clerk prompt into KiloCode
- 26:21 Env var setup
- 26:54 Clerk integrated (packages & code)
- 27:54 Set env & test sign-in
- 28:25 Sign in works; manage account
- 28:49 Account UI
- 29:17 Database with Neon (Postgres)
- 30:17 Add Neon MCP server (optional)
- 31:49 Backend prompt & wiring
- 32:50 Create users table logic
- 33:51 Ask to run SQL for tables
- 34:22 Execute SQL manually if needed
- 35:05 Vercel deploy flow
- 35:38 Push to GitHub repo
- 37:19 Push done; import in Vercel
- 37:45 Vercel project import
- 38:45 Add env vars & deploy
- 39:15 Live URL + quick check
- 39:45 Add custom domain (DNS overview)
- 40:43 Recap: free full-stack pipeline
- 41:13 Outro & what to build next
Video description and links
This article was written from the video's description with the help of AI. Watch the video for the full walkthrough.
Related tutorials
▶Qoder.ai: A New Way To Build Full-Stack Apps Instantly - Qoder.ai Builds Complex Apps From Scratch
Watch & read
▶Build AI Agents No Code using ASCN AI: How I Made $4,200 in 30 Days
Watch & read
▶