Week 14 of my “I Used It for a Week” series. After autonomous agents and terminal tools, I wanted to test something simpler: an AI coding assistant that lives in VS Code, does autocomplete and chat, but lets you choose your own model. That’s Continue.dev.

Continue is the open-source answer to “I want Copilot but with Claude” or “I want Copilot but running locally.” It’s a VS Code (and JetBrains) extension that provides autocomplete, chat, and edit capabilities — powered by whatever LLM you choose. Claude, GPT, Gemini, Ollama, LM Studio, or any OpenAI-compatible endpoint.

No subscription. No vendor lock-in. Just an extension and your API key.

The Setup

Install the extension from the VS Code marketplace. Open the config file. Point it at your model:

{
  "models": [{
    "provider": "anthropic",
    "model": "claude-sonnet-4.6",
    "apiKey": "your-key"
  }],
  "tabAutocompleteModel": {
    "provider": "ollama",
    "model": "qwen2.5-coder:7b"
  }
}

This is Continue’s superpower: use a powerful cloud model for chat and a fast local model for autocomplete. Best of both worlds, and the autocomplete is completely free if you run it locally.

What Blew Me Away

The hybrid model setup

Running Qwen 2.5 Coder locally via Ollama for autocomplete and Claude Sonnet for chat is the best cost-performance ratio I’ve found in any AI coding tool. Autocomplete is instant (local inference, no network latency) and free. Chat is high-quality (Claude) and costs only what you use.

Copilot charges $10/month for both. With Continue, autocomplete is $0 and chat costs $1-3/day depending on usage.

Context awareness

Continue reads your open files, your workspace structure, and your terminal output. You can @-mention specific files, folders, or documentation URLs in chat. The context system is more flexible than Copilot’s and comparable to Cursor’s.

Custom slash commands

You can define reusable commands like /review (review selected code), /test (generate tests), or /explain (explain selected code). These are customizable — you write the prompt template once and reuse it. Teams can share command libraries.

What Frustrated Me

Autocomplete quality with local models

The local autocomplete with Qwen 2.5 Coder 7B is good but not Copilot-level. It’s faster (no network latency) but less accurate on complex completions. The 32B model is better but requires 20GB+ VRAM. There’s a clear quality-cost tradeoff.

Configuration is manual

There’s no “just works” setup. You need to choose your models, configure providers, set up API keys, and tune settings. For developers who want to open an editor and start coding with AI, Copilot or Cursor are easier.

Extension ecosystem conflicts

Continue occasionally conflicts with other VS Code extensions — particularly other AI extensions. If you have Copilot installed alongside Continue, you’ll get duplicate suggestions. The fix is simple (disable one), but it’s a friction point.

The Cost Comparison

Setup	Monthly cost	Autocomplete	Chat quality
Continue + Ollama + Claude	$30-60	Free (local)	Excellent
Continue + Ollama only	$0	Free	Good (local)
Continue + DeepSeek	$5-15	Good	Good
GitHub Copilot	$10	Good	Good
Cursor Pro	$20	Excellent	Excellent

Continue wins on flexibility and potentially on cost. Copilot wins on simplicity. Cursor wins on polish.

Continue vs Copilot vs Cursor

• Continue: Open-source, any model, most flexible, requires setup. Best for developers who want control over their AI stack.
• Copilot: Easiest setup, good quality, $10/month, GitHub integration. Best for developers who want AI coding without thinking about it.
• Cursor: Most polished, best autocomplete, $20/month, Composer for multi-file edits. Best for developers who want the best AI coding experience regardless of cost.

My Verdict After 7 Days

Continue.dev is the AI coding tool for developers who care about how their tools work. It’s not the easiest, not the most polished, and not the most powerful. But it’s the most transparent and the most flexible.

The hybrid local+cloud setup is genuinely clever and something no commercial tool offers. If you have a decent GPU and an Anthropic API key, Continue gives you 90% of Cursor’s functionality at a fraction of the cost with zero vendor lock-in.

Best for: Privacy-conscious developers, teams with specific model requirements, developers with local GPU hardware, anyone who wants Copilot-style features without a subscription.

Not for: Developers who want zero configuration, anyone without a GPU for local autocomplete, teams that need managed support.

Rating: 7.5/10 — The best open-source IDE AI extension. The setup cost is real, but the flexibility is unmatched.

Who Should Switch From Copilot

After using both for a week each, here’s my honest take on who should consider switching from Copilot to Continue:

• You have a GPU (even a modest one like an RTX 3060) → local autocomplete is free and fast
• You want Claude for chat → Copilot uses GPT, Continue lets you use any model
• You care about privacy → local models mean your code never leaves your machine
• You’re on a team with specific compliance requirements → you control where data goes

Who should stay on Copilot:

• You want zero configuration → Copilot just works
• You use GitHub heavily → Copilot’s GitHub integration is unmatched
• You don’t have a GPU → cloud-only Continue loses the cost advantage

This concludes my “I Used It for a Week” series — 15 tools tested over 15 weeks. For the full comparison, see Best AI Coding Tools in 2026.

FAQ

Is Continue.dev worth it?

Continue.dev itself is free and open-source — you only pay for the LLM API you choose. With a local model via Ollama for autocomplete and a cloud model for chat, you can get 90% of Cursor’s functionality at a fraction of the cost. It’s worth it if you’re willing to spend 30 minutes on initial setup.

Is Continue better than Copilot?

Continue offers more flexibility — you can use any model (Claude, GPT, local models via Ollama) and customize slash commands for your workflow. Copilot is easier to set up and has better GitHub integration. If you have a GPU for local autocomplete and want model choice, Continue wins on value. If you want zero-config simplicity, Copilot is the better pick.

Does Continue work with Ollama?

Yes — Ollama integration is one of Continue’s standout features. You can run models like Qwen 2.5 Coder locally for free autocomplete with zero network latency, while using a cloud model like Claude for chat. This hybrid local+cloud setup is something no commercial AI coding tool offers.

Related: Continue.dev Complete Guide · How to Choose an AI Coding Agent · AI Coding Tools Pricing