Continue — Open-Source AI Code Assistant for IDEs

What it is

Continue is an open-source AI code assistant that runs inside VS Code and JetBrains IDEs. Unlike closed-source alternatives, Continue lets you bring your own model — OpenAI, Anthropic, local Ollama, or any OpenAI-compatible endpoint. It provides three core features: tab autocomplete, chat sidebar, and inline edit.

The tool targets developers who want AI assistance without vendor lock-in. You control which model runs, what context it sees, and how it behaves through a JSON configuration file.

How it saves time or tokens

Continue uses context-aware autocomplete that indexes your codebase locally. This means suggestions are grounded in your actual code, not generic training data. The chat sidebar lets you ask questions about your codebase without leaving the IDE or copying code into a browser. Token usage stays efficient because Continue sends only relevant file context, not your entire project.

How to use

1. Install the Continue extension from the VS Code or JetBrains marketplace.
2. Open the Continue config file and add your model provider credentials.
3. Start coding — autocomplete activates on tab, chat opens in the sidebar.

// ~/.continue/config.json
{
  "models": [
    {
      "title": "Claude Sonnet",
      "provider": "anthropic",
      "model": "claude-sonnet-4-20250514",
      "apiKey": "sk-ant-..."
    }
  ],
  "tabAutocompleteModel": {
    "title": "Local Ollama",
    "provider": "ollama",
    "model": "starcoder2:3b"
  }
}

Example

// In the chat sidebar, you can ask:
'Explain what the handleAuth middleware does in this project'

// Continue finds the relevant file, reads the function,
// and explains it in context of your codebase.

// For inline edits, select code and press Cmd+I:
'Refactor this function to use async/await instead of callbacks'

Related on TokRepo

• AI coding tools — Compare Continue with other AI code assistants
• Local LLM with Ollama — Run the autocomplete model locally for zero API cost

Common pitfalls

• Using a large model for tab autocomplete creates noticeable latency; use a small, fast model (3B parameters or less) for autocomplete and reserve larger models for chat.
• The config.json file is per-user, not per-project; use workspace-level overrides if different projects need different models.
• Context window limits apply — if you select too many files for the chat context, the model may truncate or ignore earlier files.