Use explicit tool selection in compare (#830)

* Update compare defaults and disable local tools

* Scope tool removal to compare mode

* Use explicit tool selection for compare

* Align tool access across chat surfaces

* Adjust comparison starter prompts

* Trim OSS and Cloud comparison features

Author: zahlekhan

docs/app/api/chat/route.ts

@@ -11,15 +11,19 @@ const openUiSystemPrompt = readFileSync(
 
 const markdownSystemPrompt = `You are a helpful assistant. Respond using clear, well-structured GitHub-Flavored Markdown.
 
-Use headings, lists, tables, links, block quotes, and fenced code blocks when they make the response easier to understand. Use the available tools when they are relevant, and incorporate their results into the answer.
+Use headings, lists, tables, links, block quotes, and fenced code blocks when they make the response easier to understand.
 
 Return only Markdown content. Do not emit OpenUI Lang, component syntax, JSON UI descriptions, or instructions for a renderer.`;
 
 type ResponseMode = "markdown" | "openui";
+const TOOL_NAMES = ["get_weather", "get_stock_price", "search_web"] as const;
+type ToolName = (typeof TOOL_NAMES)[number];
+const TOOL_NAME_SET = new Set<string>(TOOL_NAMES);
 
 interface ChatRequestBody {
   messages: unknown[];
   responseMode?: ResponseMode;
+  toolNames?: ToolName[];
 }
 
 function invalidRequest(message: string) {
@@ -31,7 +35,7 @@ function parseRequestBody(body: unknown): ChatRequestBody | Response {
     return invalidRequest("Request body must be a JSON object");
   }
 
-  const { messages, responseMode } = body as Record<string, unknown>;
+  const { messages, responseMode, toolNames } = body as Record<string, unknown>;
 
   if (!Array.isArray(messages)) {
     return invalidRequest("messages must be an array");
@@ -41,9 +45,18 @@ function parseRequestBody(body: unknown): ChatRequestBody | Response {
     return invalidRequest('responseMode must be either "markdown" or "openui"');
   }
 
+  if (
+    toolNames !== undefined &&
+    (!Array.isArray(toolNames) ||
+      !toolNames.every((toolName) => typeof toolName === "string" && TOOL_NAME_SET.has(toolName)))
+  ) {
+    return invalidRequest(`toolNames must contain only: ${TOOL_NAMES.join(", ")}`);
+  }
+
   return {
     messages,
     responseMode: responseMode as ResponseMode | undefined,
+    toolNames: toolNames as ToolName[] | undefined,
   };
 }
 
@@ -262,7 +275,11 @@ export async function POST(req: NextRequest) {
     return parsedBody;
   }
 
-  const { messages, responseMode = "openui" } = parsedBody;
+  const { messages, responseMode = "openui", toolNames } = parsedBody;
+  const selectedTools =
+    toolNames === undefined
+      ? tools
+      : tools.filter((tool) => toolNames.includes(tool.function.name as ToolName));
 
   const apiKey = process.env.OPENROUTER_API_KEY;
   if (!apiKey) {
@@ -328,15 +345,24 @@ export async function POST(req: NextRequest) {
       let callIdx = 0;
       let resultIdx = 0;
 
-      const runner = (client.chat.completions as any).runTools(
-        {
-          model: MODEL,
-          messages: chatMessages,
-          tools,
-          stream: true,
-        },
-        { signal: req.signal },
-      );
+      const runner: any =
+        selectedTools.length === 0
+          ? client.chat.completions.stream(
+              {
+                model: MODEL,
+                messages: chatMessages,
+              },
+              { signal: req.signal },
+            )
+          : (client.chat.completions as any).runTools(
+              {
+                model: MODEL,
+                messages: chatMessages,
+                tools: selectedTools,
+                stream: true,
+              },
+              { signal: req.signal },
+            );
       activeRunner = runner;
 
       const handleAbort = () => {
@@ -351,27 +377,29 @@ export async function POST(req: NextRequest) {
         close();
       };
 
-      runner.on("functionToolCall", (fc: any) => {
-        const id = `tc-${callIdx}`;
-        pendingCalls.push({ id, name: fc.name, arguments: fc.arguments });
-        enqueue(sseToolCallStart(encoder, { id, function: { name: fc.name } }, callIdx));
-        callIdx++;
-      });
-
-      runner.on("functionToolCallResult", (result: string) => {
-        const tc = pendingCalls[resultIdx];
-        if (tc) {
-          enqueue(
-            sseToolCallArgs(
-              encoder,
-              { id: tc.id, function: { arguments: tc.arguments } },
-              result,
-              resultIdx,
-            ),
-          );
-        }
-        resultIdx++;
-      });
+      if (selectedTools.length > 0) {
+        runner.on("functionToolCall", (fc: any) => {
+          const id = `tc-${callIdx}`;
+          pendingCalls.push({ id, name: fc.name, arguments: fc.arguments });
+          enqueue(sseToolCallStart(encoder, { id, function: { name: fc.name } }, callIdx));
+          callIdx++;
+        });
+
+        runner.on("functionToolCallResult", (result: string) => {
+          const tc = pendingCalls[resultIdx];
+          if (tc) {
+            enqueue(
+              sseToolCallArgs(
+                encoder,
+                { id: tc.id, function: { arguments: tc.arguments } },
+                result,
+                resultIdx,
+              ),
+            );
+          }
+          resultIdx++;
+        });
+      }
 
       runner.on("chunk", (chunk: any) => {
         // Keep credit handling to non-2xx responses. Provider-specific mid-stream

docs/app/chat/_components/agent-surfaces/oss-agent-surface.tsx

@@ -47,6 +47,7 @@ export function OssAgentSurface({ themeMode, onCreditsExhausted }: OssAgentSurfa
           headers: { "Content-Type": "application/json" },
           body: JSON.stringify({
             messages: openAIMessageFormat.toApi(messages),
+            toolNames: [],
           }),
           signal,
         });

docs/app/compare/_components/agent-surfaces/use-comparison-chat-llm.ts

@@ -23,6 +23,7 @@ export function useComparisonChatLLM(
           body: JSON.stringify({
             messages: openAIMessageFormat.toApi(messages),
             responseMode,
+            toolNames: [],
           }),
           signal,
         });

docs/app/compare/_components/chat-page-header.tsx

@@ -60,7 +60,6 @@ interface ComparisonPairMetadata {
 const ALL_MODES_SUPPORTED: FeatureSupport = { markdown: true, oss: true, cloud: true };
 const OSS_AND_CLOUD_SUPPORTED: FeatureSupport = { markdown: false, oss: true, cloud: true };
 const CLOUD_ONLY_SUPPORTED: FeatureSupport = { markdown: false, oss: false, cloud: true };
-const MARKDOWN_AND_OSS_SUPPORTED: FeatureSupport = { markdown: true, oss: true, cloud: false };
 
 function comparisonFeature(label: string, support: FeatureSupport): ComparisonFeature {
   return { label, support };
@@ -101,7 +100,6 @@ const PAIR_METADATA: Record<ComparisonPair, ComparisonPairMetadata> = {
       { label: "Built-in tools", support: CLOUD_ONLY_SUPPORTED },
       { label: "Responsive output by default", support: CLOUD_ONLY_SUPPORTED },
       { label: "Automatic UI error correction", support: CLOUD_ONLY_SUPPORTED },
-      { label: "Self-hostable and open source", support: MARKDOWN_AND_OSS_SUPPORTED },
     ],
   },
   "markdown-cloud": {

docs/app/compare/_components/comparison-controls.tsx

@@ -4,7 +4,6 @@ import {
   ArrowRight,
   ArrowUp,
   FileText,
-  LayoutDashboard,
   ListChecks,
   Moon,
   Plane,
@@ -27,6 +26,7 @@ const COMPARISON_SUGGESTIONS = [
       "Show me a chart of the top 5 US stocks outperforming the market in 2025 with key trendlines.",
     icon: TrendingUp,
     color: "#067647",
+    hiddenForPair: "oss-cloud",
   },
   {
     label: "Hidden travel gems to explore",
@@ -35,13 +35,6 @@ const COMPARISON_SUGGESTIONS = [
     icon: Plane,
     color: "#dd517b",
   },
-  {
-    label: "Create an executive dashboard",
-    prompt:
-      "Visualize following SaaS metrics: MRR $1.28M, growth 8.4%, NRR 112%, churn 2.1%, CAC $740, and pipeline $3.6M. Highlight trends, risks, and the three actions leadership should take.",
-    icon: LayoutDashboard,
-    color: "#b54708",
-  },
   {
     label: "Create an editable launch plan",
     prompt:
@@ -148,7 +141,9 @@ export function ComparisonControls({
         <div className={styles.suggestionScroller} aria-label="Try a comparison prompt">
           <div className={styles.suggestionRow}>
             {COMPARISON_SUGGESTIONS.filter(
-              (suggestion) => !("pairOnly" in suggestion) || suggestion.pairOnly === comparisonPair,
+              (suggestion) =>
+                (!("pairOnly" in suggestion) || suggestion.pairOnly === comparisonPair) &&
+                (!("hiddenForPair" in suggestion) || suggestion.hiddenForPair !== comparisonPair),
             ).map((suggestion) => {
               const SuggestionIcon = suggestion.icon;
               const needsCloud = "cloudOnly" in suggestion && suggestion.cloudOnly;

docs/lib/openui-cloud/models.ts

@@ -1,4 +1,4 @@
-export const DEFAULT_MODEL = "anthropic/claude-sonnet-4.6";
+export const DEFAULT_MODEL = "openai/gpt-5.4";
 
 export interface ModelOption {
   id: string;