DEV Community

Free AI
Free AI

Posted on

How I Built a Free AI Chatbot With 12+ Models, Voice, Coding, and Image Generation — All for $0

#webdev #javascript #ai #opensource

How I Built a Free AI Chatbot With 12+ Models, Voice, Coding, and Image Generation — All for $0

I spent the last few months building Free AI — a completely free AI assistant that does what paid tools charge $20/month for. Here's what it can do and how I built it.

What It Does:

  • Chat — unlimited messages, no daily caps
  • Web search — answers with live source citations
  • Code execution — runs JavaScript, Python, TypeScript, HTML/CSS, and more in the browser
  • Image generation — free AI image creation
  • Voice conversation — real-time voice chat with the AI
  • File analysis — upload PDFs, images, documents and ask questions about them
  • Multi-agent bots — a team of AI agents that collaborate on complex tasks
  • Cross-chat memory — the AI remembers facts across conversations
  • Think mode — toggle deeper reasoning for complex problems

The Tech Stack:

  • Frontend: React 19, TypeScript, Vite, Tailwind CSS, Framer Motion
  • Backend: Convex (real-time database + serverless functions)
  • AI Models: 12+ free models via NVIDIA NIM, Groq, Google Gemini, and OpenRouter
  • Auth: Convex Auth with email OTP
  • Code Execution: Pyodide (Python in the browser), sandboxed JavaScript
  • Voice: Edge TTS + Web Speech API
  • Search: Exa API for web search with source citations

How the Model Routing Works

The system automatically routes requests to the best available model:

  1. Tries NVIDIA models first (DeepSeek, Nemotron)
  2. Falls back to Groq (Llama 3.3 70B) if overloaded
  3. Falls back to Gemini Flash
  4. Falls back to OpenRouter free models
  5. Each provider has a 12-second timeout — fails fast, moves on

Users never see the routing. They just get an answer.

The Bot System (My Favorite Part)

When users enable "Bots," a planner agent analyzes the request, creates a task graph, and assigns roles to multiple AI agents that work in parallel:

  • Project Manager
  • Frontend Developer
  • Backend Engineer
  • QA Tester
  • Security Reviewer

They share a workspace, review each other's work, and produce one final result. The user sees a live activity feed and a neural brain visualization that grows as the bots work.

What I Learned:

  1. Free models are surprisingly good — DeepSeek and Llama 3.3 rival GPT-4 for most tasks
  2. Fallback chains are essential — any single provider will go down. Having 4+ providers means 99.9% uptime
  3. Simplicity wins — the chat interface is intentionally minimal. No plugins, no menus, no complexity
  4. Real-time is addictive — Convex reactive queries make the UI feel alive. Messages stream in, the bot activity panel updates live, the brain visualization pulses

Try It:

👉 getfree-ai.com — no signup required, completely free

I'd love feedback from the dev community. What features would you add? What's missing?

Top comments (0)