Use explicit tool selection in compare

[zahlekhan]

Summary

• have Markdown and OpenUI OSS comparisons request an empty tool selection
• validate requested tool names against the server-owned tool registry
• omit provider tools and use plain streaming when the resolved selection is empty
• change the OpenUI Cloud default model from Claude Sonnet 4.6 to GPT-5.4

Why

The compare experience describes built-in tools as a Cloud-only capability, but the Markdown and OpenUI OSS endpoint was still exposing three mock tools. Passing toolNames: [] keeps the generic chat route independent from the compare UI while preserving the existing tools for callers that omit a selection. Using GPT-5.4 as the Cloud default also lets the surfaces be evaluated with the requested model.

Impact

Compare Markdown and OpenUI OSS no longer invoke local function tools. Standalone OSS chat retains web search, weather, and stock. OpenUI Cloud keeps its existing built-in tools and now starts with openai/gpt-5.4.

Validation

• pnpm exec eslint app/api/chat/route.ts lib/openui-cloud/models.ts
• pnpm exec prettier --check app/api/chat/route.ts lib/openui-cloud/models.ts
• pnpm types:check
• live A/B test: compare Markdown and OSS emitted zero tool calls while standalone OSS invoked get_weather
• invalid tool names return 400 before reaching the provider
• direct Cloud stream metadata confirmed openai/gpt-5.4