Coderay.

Docs

Models and effort modes

The default mode is fast and inexpensive enough to feel unlimited. Think harder reasons longer before answering; use it for debugging and design questions, not everyday ones. New here? The getting-started guide covers switching it on.

How usage works

You get generous usage — enough for a full workday of questions and edits. Unused usage carries over, so an easy week leaves you more room the next one. Your dashboard always shows where you stand. See pricing for how the plan itself is priced.

Troubleshooting

The extension says I'm signed out, or answers stopped coming.
Run “Coderay: Sign in” from the Command Palette and press Connect VS Code again. The link is one-use and expires in 5 minutes, so start from VS Code each time.
Everything's blocked, even short questions.
You've hit your usage limit. Your dashboard shows when it resets.
Think harder feels slow.
That's expected. It reasons longer before answering, which is also why it uses more of your budget. Switch it off once you're done.

Still stuck? Use the contact form. We read every message.

FAQ

Which models does Coderay use?

A leading open-source coding model by default, with a slower "Think harder" mode for tougher problems. We tune and host these ourselves — we don't train them, and we don't claim they're ours.

Is my code stored anywhere outside this request?

Requests are processed to generate a response and logged for billing (which model, how many tokens, what it cost). We don't make claims about where that infrastructure physically sits.

Can I use this with an editor other than VS Code?

The extension also installs in VS Code forks like Cursor and Windsurf (from Open VSX). Prefer another client? See the bottom of this page.

Does unused budget carry over?

Yes — unused usage carries over, so an easy week leaves you more room the next one.

Prefer your own client?

Any client that can talk to an OpenAI-compatible endpoint works, for example Cline. It is not the recommended way: the Coderay extension is built for this service.

  1. Copy your API key from the dashboard.
  2. In the client, set the provider to OpenAI Compatible and the base URL to https://klaversof-production.up.railway.app/v1.
  3. Paste the key, and type deepseek-v4-flash into the Model ID field exactly. For a harder question, use deepseek-v4-flash-think-max.

An error mentioning gpt-4o or rejected credentials means the Model ID field didn't save: Cline falls back to gpt-4o when it's empty. Re-type the model ID and save.