Continue.dev

Continue.dev is what you install when you want AI-in-editor without committing to a single vendor. Cursor is faster out of the box but locks you into their stack. GitHub Copilot ships well but ties you to one model family. Continue is the open-source middle path: bring your own model, your own endpoint, your own data policy.

What it does

Chat with the model in a sidebar, get autocomplete inline, run edit-this-block commands, and configure rules per-project that keep the model on-style. The configuration is plain JSON, version-controlled with your repo. Models can be Claude (via Anthropic API or Bedrock), OpenAI, Gemini, local Ollama, or any OpenAI-compatible endpoint your team runs.

Who it’s for

• MS and PhD holders in DS / ML engineering who want a cheaper alternative to Cursor without losing the productivity bump
• Industry research scientists at companies with strict data residency rules — point Continue at a private endpoint and your code never leaves your network
• Anyone running local models for offline work, sensitive code, or air-gapped environments
• Teams that want one shared .continue/config.json so reviewers, juniors, and senior engineers all get the same model behavior

What to watch for

• Setup friction is real. First-run config is more involved than Cursor’s “sign in and go.” Budget 30 minutes
• Autocomplete latency depends on your model choice. Local 7B models are noticeably slower than hosted Claude Haiku for inline completion. Pick your tradeoff deliberately
• Plugin ecosystem is smaller than the editor extensions of bigger vendors. Most people don’t need this; some will

Verdict

If you care about avoiding vendor lock-in for AI-in-editor — and you should, because the model landscape is shifting quarterly — Continue is the right primary tool. Pair with Aider for terminal-first work and you have the open-source coverage of an entire commercial AI-coding stack.