Stan Cosmin
All resources
GuideEngineering·6 min read

Free Local Claude Code Alternative

Continue + a local Qwen model on Ollama, wired into VS Code. Offline, free, no API key.

3 September 2026

Claude Code costs $20/month. Worth it once you can afford it, but if you can't yet, or you just want something offline with no API key, Continue + a local Qwen model gets you most of the way there. Free, and your code never leaves the machine. Works the same on macOS, Windows and Linux, pick yours below.

You'll also need VS Code, about 15GB of free disk, and 8GB+ RAM depending on which model tier you pick further down.

What's inside

  • Pick your OS and your RAM, get the exact commands for both
  • The full config.json for Continue, already filled in for your pick
  • How to actually use it: chat, inline edits, slash commands
  • What to do when it's slow, out of RAM, or ignoring your codebase

Your OS

Install Ollama (macOS)

brew install ollama
brew services start ollama

Then VS Code -> Extensions (Cmd+Shift+X), search "Continue", install it. Reload its config later with Cmd+Shift+P -> "Continue: Reload Config".

Pick your setup

Once Ollama and the Continue extension are installed (above), click the tier that matches your RAM.

Start here if you're unsure. A proven coding model, handles real refactors without dragging.

Pull this model

ollama pull qwen2.5-coder:14b

Continue config

Cmd/Ctrl+Shift+P -> "Continue: Open Config File"

{
  "models": [{
    "title": "Local AI Coding",
    "provider": "ollama",
    "model": "qwen2.5-coder:14b",
    "apiBase": "http://localhost:11434",
    "temperature": 0.3,
    "topP": 0.9,
    "maxTokens": 4096,
    "systemMessage": "You are an expert coding assistant. Provide production-grade code with error handling. Explain trade-offs. Be concise but thorough."
  }],
  "slashCommands": [
    {"name": "edit", "description": "Edit selected code"},
    {"name": "explain", "description": "Explain what code does"},
    {"name": "refactor", "description": "Improve code quality"},
    {"name": "fix", "description": "Find and fix bugs"},
    {"name": "test", "description": "Write tests"}
  ],
  "contextProviders": [
    {"name": "codebase"},
    {"name": "diff"},
    {"name": "open-files"}
  ]
}

Reload & test

Use the reload shortcut from the OS picker above -> "Continue: Reload Config". Then open any file, click the Continue icon in the sidebar, type "explain this function". A response in a few seconds means it's working.

How to use it

Chat

Click the Continue sidebar and ask anything about the code.

  • "Refactor this for performance"
  • "What does this function do?"
  • "Write tests for this module"
  • "@codebase show me the authentication flow"

Inline editing

Select code, press Cmd+K (Ctrl+K on Windows/Linux), type what you want, hit Enter. The edit lands in place.

Slash commands

Type / in the chat: /edit, /explain, /refactor, /fix, /test.

Project context

Continue reads every open file plus your project structure by default. Ask @codebase [question] to force a full-project pass instead of just what's open.

Customize

Everything lives in the same config file the picker filled in above. Open it anytime from the Command Palette with "Continue: Open Config File", it finds the right file for your OS automatically.

System prompt: change what it optimizes for.

"systemMessage": "You are an expert in [your language]. Always use [your preference]. Warn about [edge case]."

Temperature: creative vs. precise.

"temperature": 0.2   // coding, precise
"temperature": 0.5   // balanced
"temperature": 0.8   // brainstorming

Response length:

"maxTokens": 2048   // shorter, faster
"maxTokens": 4096   // longer, thorough

Troubleshooting

Continue doesn't respond. Check Ollama is running, then reload the config with the shortcut from the OS picker above:

ollama ps

The 24GB+ tier takes forever. That's expected, not broken. This model reasons out loud before it answers, sometimes for a while, especially on a simple request. It's built for hard problems. For quick edits, switch to the balanced tier instead of waiting it out.

Out of RAM, or it crashes. Close other heavy apps. If it still crashes, drop a tier: pick the 8GB option in the picker above and pull that model instead.

It doesn't know your codebase. Continue only sees what's open or selected. Open the relevant files first, or force a full pass with @codebase.

vs. Claude Code

  • Cost: free vs. $20/month
  • Offline: yes vs. requires internet
  • Speed: seconds, longer on the top tier vs. near-instant
  • Customization: full control over model, prompt, temperature vs. none
  • Ceiling: good for most day-to-day work, but the top hosted models still lead on the hardest problems

Nothing stops you from running both: local for the daily grind, Claude Code for the calls that actually matter.

All resources