Add SaladCloud AI Gateway provider

[mgorkii-nlplogix]

Summary

Adds SaladCloud AI Gateway as a first-class OpenAI-compatible model provider.

Changes

• Add saladcloud provider wrapper with default API base https://ai.salad.cloud/v1/
• Disable Qwen thinking output by default using chat_template_kwargs.enable_thinking: false
• Preserve explicit user-provided chat_template_kwargs
• Register SaladCloud in the provider list
• Add SaladCloud model presets for the currently available Qwen models
• Mark SaladCloud Qwen models as tool-capable
• Add SaladCloud provider documentation
• Add unit coverage for the provider wrapper and tool support

Validation

• npm run vitest -- llm/llms/SaladCloud.vitest.ts llm/llms/OpenAI-compatible.vitest.ts
• npm test -- llm/toolSupport.test.ts
• npm run tsc:check in core
• npm run tsc:check in gui
• Live SaladCloud smoke test for model listing, streaming chat, and forced tool calling

---

Summary by cubic

Adds SaladCloud AI Gateway as an OpenAI-compatible provider with Qwen presets and UI/docs integration. Disables Qwen “thinking” by default and marks Qwen models as tool-capable.

• New Features
  • New saladcloud provider with default API base https://ai.salad.cloud/v1/ and non-legacy completions.
  • Sets chat_template_kwargs.enable_thinking to false by default; preserves user overrides.
  • Registers provider and GUI presets for qwen3.6-35b-a3b, qwen3.6-27b, and qwen3.5-9b; updates tool support checks for these models.
  • Adds provider docs and unit tests (wrapper behavior, OpenAI subclass tests, tool support).

^{Written for commit 9d114a9. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 9 files

[mgorkii-nlplogix]

I have read the CLA Document and I hereby sign the CLA

[mgorkii-nlplogix]

Before rebuilding this contribution against current main, I’m narrowing the proposal to the currently advertised SaladCloud AI Gateway model only: qwen3.6-35b-a3b, using https://ai.salad.cloud/v1, bearer authentication, and the current $0.09/M input and $0.60/M output prices. The replacement will remain a draft for review, preserve the existing branch, and include current tests/docs. If maintainers prefer this to use only Continue’s generic OpenAI-compatible provider instead of a first-class provider entry, feedback here is welcome.

[mgorkii-nlplogix]

Superseded by #13088, which is rebuilt from the current main branch and narrows the integration to qwen3.6-35b-a3b with current per-token pricing and capability metadata.