Continue — Open-Source AI Code Assistant

What it is

Continue is an open-source AI code assistant that integrates into VS Code and JetBrains IDEs. It provides tab autocomplete, chat, and inline editing powered by any LLM provider -- OpenAI, Anthropic, Ollama, or your own self-hosted endpoint.

Continue is for developers who want AI coding assistance with full control over which model they use and where their code is sent. Unlike closed-source alternatives, Continue lets you swap providers, run local models for privacy, or use your company's internal API endpoint.

How it saves time or tokens

Continue reduces context-switching between your IDE and external chat interfaces. Instead of copying code into a browser-based LLM, you get inline suggestions and edits directly in your editor.

With model flexibility, you can route different tasks to different models. Use a fast local model for tab autocomplete (low latency matters) and a larger cloud model for complex refactoring or code generation. This optimizes both cost and speed.

How to use

1. Install Continue from the VS Code Marketplace or JetBrains Marketplace -- search for 'Continue'.

1. Configure your model provider in ~/.continue/config.json:

{
  "models": [
    {
      "title": "Claude Sonnet",
      "provider": "anthropic",
      "model": "claude-sonnet-4-20250514",
      "apiKey": "sk-ant-..."
    },
    {
      "title": "Ollama Local",
      "provider": "ollama",
      "model": "codellama:13b"
    }
  ],
  "tabAutocompleteModel": {
    "title": "Starcoder",
    "provider": "ollama",
    "model": "starcoder2:3b"
  }
}

1. Use Cmd+I (or Ctrl+I) for inline editing, Cmd+L for chat, and Tab for autocomplete.

Example

Using Continue's inline edit to refactor a function:

# Select this function, press Cmd+I, type 'add error handling and type hints'
def fetch_user(user_id):
    response = requests.get(f'/api/users/{user_id}')
    return response.json()

# Continue generates:
def fetch_user(user_id: int) -> dict:
    try:
        response = requests.get(f'/api/users/{user_id}')
        response.raise_for_status()
        return response.json()
    except requests.RequestException as e:
        raise UserFetchError(f'Failed to fetch user {user_id}') from e

Related on TokRepo

• Coding AI tools -- more AI-powered code assistants
• Local LLM tools -- run models locally for privacy

Common pitfalls

• Tab autocomplete latency depends on your model choice. Cloud models add network round-trip time. For the best autocomplete experience, use a local model (Ollama + Starcoder or CodeLlama).
• Continue's context window is limited by the model you choose. Large files may exceed the context, producing incomplete suggestions. Use the @file and @folder context providers to scope what Continue sees.
• Configuration is per-user in ~/.continue/config.json. For team-wide settings, use the .continue/ directory in your project root.