Optimized Inference API for Coding · VS Code integration
npx @deepvariance/vscodeEnter your email — no API key to copy, no config files to edit. Then quit and reopen VS Code: Qwen3.6 is already selected in Chat.
| Model | Context |
|---|---|
| Qwen3.6 27B | 131k tokens |
- Reasoning — the model shows its thinking before it answers
- Tool use — works in Chat's agent mode, so it can read and edit your code
- Vision — attach a screenshot, a diagram, or an error dialog (up to 4 images)
| IDE | Min version | What's supported |
|---|---|---|
| VS Code | 1.128 | Everything — what we test on |
| VS Code | 1.104 | Chat only; agent window needs 1.128 |
| Cursor, Windsurf, VSCodium | — | Unverified |
Also needs Node.js 18+ with npm, for npx.
Most people never need these.
npx @deepvariance/vscode --health # is the server up?--email <email> |
Skip the email prompt |
--gateway <url> |
Use a different server |
--invite <token> |
Use a different invite |
--yes |
Don't ask anything |
--health |
Check the server and exit |
Architecture, the gateway contract, and a change recipe per edit live in SPEC.md. Workflow is in CLAUDE.md — every change goes on a branch and lands via reviewed PR.
Beta. Deep Variance's inference API is in active development. Availability, models, and behavior may change without notice, and the service may be unavailable at times. Not for production use.