How to Build a 100% Free AI Coding Setup in VS Code (No Subscriptions Required)

The Bring Your Own Key (BYOK) model fundamentally changes the economics of AI-assisted coding. Instead of paying $19-39/month for GitHub Copilot or Cursor Pro, you connect your VS Code client directly to free LLM API endpoints. The architecture is simple: your coding tool (VS Code Continue, Cline, or OpenCode) sends requests to an OpenAI-compatible API endpoint at providers like Groq, SambaNova, or Google AI Studio. These providers offer generous free tiers — Groq provides 30 requests/minute and 14,400 requests/day on Llama 3.3 70B, SambaNova offers ~20 RPM on Qwen 2.5 Coder 32B, and Google AI Studio gives 15 RPM on Gemini 2.0 Flash with 1M tokens/day. The key insight: these free tiers are not trials — they are permanent, infrastructure-subsidized offerings designed to build developer ecosystems. You get GitHub Copilot-grade assistance for $0/month with no credit card required.

Continue.dev is the most popular open-source AI coding assistant for VS Code, supporting both inline autocomplete (tab completions) and chat-based code generation. Installation takes 30 seconds: open VS Code Extensions panel (Ctrl+Shift+X), search 'Continue', and install. Then configure two separate models for optimal performance. For tab autocomplete, use a small, fast model like Llama 3.1 8B on Groq (330+ tokens/sec). For chat and code generation, use a larger model like Llama 3.3 70B or Qwen 2.5 Coder 32B. The configuration file lives at ~/.continue/config.json:

{
  "models": [
    {
      "title": "Llama 3.3 70B (Chat)",
      "provider": {
        "name": "openai",
        "baseUrl": "https://api.groq.com/openai/v1",
        "apiKey": "gsk_YOUR_GROQ_KEY"
      }
    }
  ],
  "tabAutocompleteModel": {
    "title": "Llama 3.1 8B (Autocomplete)",
    "provider": {
      "name": "openai",
      "baseUrl": "https://api.groq.com/openai/v1",
      "apiKey": "gsk_YOUR_GROQ_KEY"
    }
  }
}

Cline is an autonomous AI coding agent that can plan, edit multiple files, run terminal commands, and execute complex workflows. Unlike Continue (which requires manual acceptance of each suggestion), Cline operates in Plan Mode (analyzes code, proposes changes) and Act Mode (executes changes). To configure Cline with free providers: 1. Install Cline from VS Code Extensions 2. Open Cline settings (Ctrl+, → Extensions → Cline) 3. Set Provider to 'OpenAI Compatible' 4. Enter Base URL: https://api.groq.com/openai/v1 5. Enter API Key: gsk_YOUR_GROQ_KEY 6. Model ID: qwen/qwen2.5-coder-32b Rate Limit Strategy: Cline consumes tokens aggressively during Act Mode (full file rewrites, test execution). To stay within free tier limits: use Plan Mode for analysis (cheap), then switch to Act Mode only for critical edits. Set MAX_REQUESTS_PER_MINUTE=25 in Cline settings to avoid 429 errors. For complex tasks, break them into smaller steps — each step uses fewer tokens and stays within rate limits.

OpenCode is a terminal-first AI coding agent that runs headless tasks directly in your integrated terminal. It reads your codebase context from the filesystem and executes multi-step coding tasks autonomously. Configuration is via JSON file at ~/.config/opencode/opencode.json: Install: npm install -g opencode Configure the JSON file with your provider:

{
  "providers": [{
    "name": "groq",
    "baseUrl": "https://api.groq.com/openai/v1",
    "apiKey": "gsk_YOUR_GROQ_KEY",
    "models": ["meta-llama/llama-3.3-70b"]
  }]
}

HTTP 429 Rate Limits: Every free provider enforces request limits. Groq: 30 RPM, 14,400 req/day. SambaNova: ~20 RPM with daily quota. Google AI Studio: 15 RPM. When you hit 429 errors, implement a fallback chain: primary provider → secondary provider → local Ollama. Most tools support configuring multiple providers and automatic fallback. Context Window Truncation: Free models have context limits (Llama 3.3 70B: 128k, Qwen 2.5 Coder: 32k). When working with large files (>500 lines), the model may lose context. Fix: break large files into smaller functions, use @file mentions in chat to focus on specific sections, or switch to a model with larger context (Gemini 2.0 Flash: 1M tokens). API Key Management: Never commit API keys to git. Use environment variables (export GROQ_API_KEY=...) or local config files outside your repo. For team setups, consider using OpenRouter as a single API key that routes to multiple free providers.