fix: web/app/api/chat/route.ts

web/app/api/chat/route.ts

@@ -1,22 +1,23 @@
 import { NextRequest, NextResponse } from "next/server";
 import { createClient } from "@/lib/supabase/server";
 
-const GRADIO_URL = process.env.CLAUSEGUARD_GRADIO_URL || "https://gaurv007-clauseguard.hf.space";
-
 /**
- * FIX v4.1: The chat endpoint now properly documents its limitations.
+ * FIX v4.3: Chat route completely rewritten.
  *
- * ARCHITECTURE NOTE:
- * The Gradio ChatInterface uses gr.State to store RAG embeddings per-session.
- * This state is NOT accessible via the Gradio API from an external caller —
- * each API call creates a new session with empty state.
+ * ARCHITECTURE:
+ * The Gradio ChatInterface uses gr.State for RAG embeddings — these are
+ * per-browser-session and NOT accessible via the Gradio REST API. Every API
+ * call creates a new session with empty state, so chat via Gradio API will
+ * NEVER have contract context.
  *
- * For the Next.js web app, chat should either:
- * 1. Use the FastAPI backend (/api/chat) which manages its own RAG sessions, OR
- * 2. Embed the Gradio Space in an iframe for direct interaction
+ * The correct approach:
+ * 1. PRIMARY: Use the FastAPI backend (/api/chat) which manages RAG sessions
+ *    with proper TTL-based expiry. The session_id comes from /api/analyze.
+ * 2. FALLBACK: If FastAPI is unavailable, return a clear error directing
+ *    the user to use the Gradio Space directly.
  *
- * This route attempts to use the Gradio API as a best-effort fallback,
- * but will clearly communicate to the user if the session is unavailable.
+ * The old code tried to call a non-existent Gradio "chat" endpoint which
+ * always failed. Removed the broken Gradio fallback entirely.
  */
 export async function POST(req: NextRequest) {
   try {
@@ -37,7 +38,6 @@ export async function POST(req: NextRequest) {
       );
     }
 
-    // FIX v4.1: Input validation
     if (message.length > 2000) {
       return NextResponse.json(
         { error: "Message too long (max 2000 characters)" },
@@ -45,7 +45,7 @@ export async function POST(req: NextRequest) {
       );
     }
 
-    // Try the FastAPI backend first (it has proper RAG session management)
+    // Try the FastAPI backend (it has proper RAG session management)
     const apiUrl = process.env.CLAUSEGUARD_API_URL || "";
     if (apiUrl && session_id) {
       try {
@@ -58,67 +58,40 @@ export async function POST(req: NextRequest) {
           const data = await apiRes.json();
           return NextResponse.json({ response: data.response });
         }
+        // If 404, session expired
+        if (apiRes.status === 404) {
+          return NextResponse.json({
+            response: "⚠️ Your chat session has expired (sessions last 1 hour). " +
+              "Please analyze the contract again to start a new chat session."
+          });
+        }
       } catch {
-        // Fall through to Gradio attempt
+        // FastAPI backend unreachable — fall through to error message
       }
     }
 
-    // Fallback: Try the Gradio API
-    const submitRes = await fetch(`${GRADIO_URL}/gradio_api/call/chat`, {
-      method: "POST",
-      headers: { "Content-Type": "application/json" },
-      body: JSON.stringify({ data: [message] }),
-    });
-
-    if (!submitRes.ok) {
-      const errText = await submitRes.text().catch(() => "");
-      throw new Error(`Chat submit failed (${submitRes.status}): ${errText}`);
-    }
-
-    const { event_id } = await submitRes.json();
-    if (!event_id) throw new Error("No event_id from Gradio chat");
-
-    // Poll for result with timeout
-    let resultText = "";
-    let attempts = 0;
-    const maxAttempts = 30;
-
-    while (attempts < maxAttempts) {
-      const resultRes = await fetch(
-        `${GRADIO_URL}/gradio_api/call/chat/${event_id}`,
-        { headers: { Accept: "text/event-stream" } }
-      );
-
-      if (!resultRes.ok) {
-        throw new Error(`Chat result failed: ${resultRes.status}`);
-      }
-
-      resultText = await resultRes.text();
-
-      if (resultText.includes("event: complete")) break;
-      if (resultText.includes("event: error")) {
-        const errMatch = resultText.match(/event:\s*error\s*\ndata:\s*(.+)/);
-        if (errMatch) throw new Error(`Chat error: ${errMatch[1]}`);
-        throw new Error("Chat error from backend");
-      }
-
-      await new Promise(r => setTimeout(r, 1000));
-      attempts++;
+    // No FastAPI backend available or no session_id
+    // FIX v4.3: Return a clear, helpful message instead of trying a broken Gradio endpoint
+    if (!apiUrl) {
+      return NextResponse.json({
+        response: "⚠️ Contract Q&A chat requires the FastAPI backend which is not currently deployed. " +
+          "You can use the chat feature directly in the [Gradio Space](https://gaurv007-clauseguard.hf.space) " +
+          "— analyze a contract there, then switch to the Q&A tab."
+      });
     }
 
-    // Find the complete event data
-    const dataMatch = resultText.match(/event:\s*complete\s*\ndata:\s*(.+)/);
-    if (!dataMatch) {
+    if (!session_id) {
       return NextResponse.json({
-        response: "⚠️ Chat is unavailable. The contract needs to be analyzed in the same session. " +
-          "Please analyze a contract first in the Gradio Space, then use the chat tab there directly."
+        response: "⚠️ No active chat session. Please analyze a contract first — " +
+          "the chat session is created when you run analysis."
       });
     }
 
-    const responseData = JSON.parse(dataMatch[1]);
-    const responseText = typeof responseData === "string" ? responseData : responseData[0] || "";
+    return NextResponse.json({
+      response: "⚠️ Chat service is temporarily unavailable. Please try again, or use the " +
+        "[Gradio Space](https://gaurv007-clauseguard.hf.space) directly."
+    });
 
-    return NextResponse.json({ response: responseText });
   } catch (error: any) {
     console.error("Chat error:", error.message);
     return NextResponse.json(