← All posts

Editor

Continue: Open-Source AI Autocomplete for Your IDE

Continue is an open-source AI assistant for VS Code and JetBrains - autocomplete, chat, and edits on any model you pick.

C Editor

GitHub Copilot proved that in-editor AI is worth having, but it's a closed box wired to one vendor's models. Continue is the open-source answer: a VS Code and JetBrains extension that gives you autocomplete, chat, and inline edits, but lets you choose the model behind each of those features - a hosted frontier model for chat, a fast local one for autocomplete, whatever fits. It's Apache-2.0, and your configuration lives in a file you own.

Bring Your Own Model

Continue separates the feature from the model. You can point tab-autocomplete at a small local model running in Ollama for speed and privacy, and route the chat sidebar to a stronger hosted model for harder questions. Nothing is hardcoded to a single provider.

# ~/.continue/config.yaml (excerpt) models: - name: Local Autocomplete provider: ollama model: qwen2.5-coder:1.5b roles: [autocomplete] - name: Chat provider: anthropic model: claude-sonnet-4 roles: [chat, edit]

More Than Autocomplete

The chat panel understands your codebase through context providers - you can @-mention a file, the whole repo, the terminal output, a git diff, or documentation you've indexed, and Continue feeds exactly that into the prompt. Inline edit lets you highlight code, describe a change, and review the diff before it applies.

Config You Can Commit

Because the setup is a plain config file, you can commit a project's Continue configuration alongside the code, so everyone on the team gets the same models, context providers, and custom prompts. It turns "how I use AI here" into something version-controlled rather than tribal knowledge.

Rough Edges and Limitations

  • Autocomplete quality tracks the model. A tiny local model is fast but will suggest less useful completions than a larger one; there's a real speed-versus-quality dial to tune.
  • Setup is more involved than Copilot. The flexibility costs you a config step. Sensible defaults exist, but the power is in customizing.
  • You manage the keys. Bring-your-own-model means bring-your-own-API-keys and their costs.
  • Indexing large repos takes time. The first codebase index on a big project isn't instant.

If you like the shape of Copilot but want to own the model choice, keep sensitive code local, and version-control how your team uses AI, Continue is the most mature open option.

Internal link suggestion: link "choose the model" to the Cody or Cline post for a comparison of in-editor AI approaches.