The short answer
Continue.dev and Onyx Code want the same thing: a coding assistant that runs on your own hardware. Continue is the extension you wire up to a runtime you provide. Onyx Code is the editor that arrives with the runtime inside it.
Pick Continue when
You want open source, you already run Ollama or LM Studio, you like configuring your own stack, and you want to stay in real VS Code or a JetBrains IDE.
Pick Onyx Code when
You want it to work in one download, and you want privacy controls that are enforced by the product rather than achieved by configuring it correctly.
Either is fine when
You are curious about local AI and want to try it cheaply. Both cost nothing to find out.
The setup difference, stated plainly
A working Continue setup is roughly: install the extension, install a runtime such as Ollama, decide which model your GPU can hold, pull several gigabytes, learn what a quantisation is, write a configuration file that names a chat model and a separate fill-in-the-middle model, then keep all of that current as models improve. People who enjoy this genuinely enjoy it, and they end up with a stack that is exactly theirs.
Onyx Code does that work at install time. The llama.cpp engine is bundled, the first-run wizard looks at your hardware and picks a model that fits, and a dedicated autocomplete model and next-edit model are set up alongside the chat model rather than being your job. You can still drop in any GGUF you like afterwards, and you can point it at an external Ollama or LM Studio server on the Pro tier.
Feature comparison
| Question | Onyx Code | Continue.dev |
|---|---|---|
| Inference engine | Bundled, with GPU acceleration configured for you | You install and run one, usually Ollama or LM Studio |
| Model choice | Recommended for your hardware, or any GGUF you drop in. One chat model on Free, unlimited on Pro | Anything your runtime supports, which is a real strength |
| Time to a first local completion | About two minutes from download | Fifteen to thirty minutes if you have not used Ollama before |
| Runs inside real VS Code | No. Standalone editor, Open VSX extensions | Yes, and JetBrains too. You keep your marketplace |
| Source licence | VS Code OSS base is MIT. The AI layer is proprietary | Apache 2.0, and actively developed |
| Hard switch that blocks all outbound traffic | Air-Gap Mode, enforced in three places, Free | Privacy comes from pointing it at a local model, not from an enforced boundary |
| Exclude paths from every AI feature | .onyxnoai, honoured by chat, index, autocomplete and agents | No equivalent AI-wide exclusion file |
| Audit trail of what the AI read and wrote | AI Activity Log plus audit-PDF export, Free | Not provided |
| Secret redaction before a model sees the prompt | Prompt sanitizer on every send, Free | Not provided |
| Telemetry | None by default | Anonymous usage telemetry, opt-out |
| Agent mode | Ask and Edit-auto on Free. Plan, Auto and worktree sandboxing are Pro | Yes, an agent is core to the project |
| Signed, notarised installer with auto-update | macOS, Windows and Linux | Not applicable. It updates with your editor’s extensions |
| Price | Free tier forever, Pro $39 once for three machines | Free and open source |
What Continue.dev does better
- It is open source, Apache 2.0. You can read every line, fork it, and depend on it without depending on us. That is a real answer to a real question, and we cannot match it.
- You keep your editor. It installs into VS Code or a JetBrains IDE, so your extensions, keybindings and marketplace are untouched. Switching to a fork always costs something; here it costs nothing.
- Total control of the stack. Any runtime, any model, any endpoint, configured exactly how you want it. If you want to run one model for chat and a different one for completions on a machine down the hall, that is a config file, not a feature request.
- It is free, with no paid tier to bump into. Nothing is held back for a licence.
- JetBrains support. Onyx Code is a VS Code-family editor only.
Who should switch
- You set up a local stack, it worked, and then a model update or a runtime upgrade broke it and you did not want to spend the evening on it.
- You are recommending local AI to colleagues, and the honest instruction is longer than the number of steps they will follow.
- You need to prove the boundary, not just describe it. Air-Gap Mode, the Activity Log and the audit PDF exist because “it points at localhost” is not evidence.
- You want secrets stripped before a prompt is built, rather than trusting yourself never to attach the wrong file.
- You want autocomplete, next-edit prediction and codebase search configured as a set, rather than as three separate decisions.
Questions people ask
Why would I pay when Continue is free?
For a lot of people the answer is that you would not, and the Onyx Code Free tier is the intended path anyway: local chat, autocomplete, next-edit prediction, codebase indexing and every privacy control, with no time limit. Pro is the agent workstation upgrade. What you are weighing is setup burden and enforcement, not price.
Can I use my existing Ollama models?
Yes. Connecting to an external Ollama, LM Studio or OpenAI-compatible server is supported on the Pro tier, and you can drop any GGUF file into the bundled engine on either tier.
Is Onyx Code open source?
The editor is VS Code OSS, which is MIT licensed. The AI layer is proprietary, and Pro is a paid licence. If open source is a requirement rather than a preference, Continue is the honest recommendation.
Do I have to give up my VS Code extensions?
You move to Open VSX, which carries most popular extensions but not all. This is the real cost of any fork, and it is the strongest argument for staying with an extension like Continue. The first-run wizard tells you by name which of yours are missing before you commit.
The same idea, with the assembly already done
One download, engine included, a model picked for your hardware, and no configuration file to write before the first completion.
Free tier, no account, no sign-up, no telemetry. macOS, Windows and Linux.