allow chat responses to use full available space for coherent answers - cap at 4096 tokens max

src/chat.js

@@ -110,8 +110,15 @@ export async function* sendMessage(message) {
 	
 	// Cap to model's max completion tokens limit
 	const maxCompletionTokens = getSelectedModelMaxCompletionTokens();
+	// For chat, prefer using the full calculated available space for coherent responses
+	// Only cap if we're exceeding the model's hard limit
 	const finalAvailable = Math.min(calculatedAvailable, maxCompletionTokens);
-	console.log("Chat using tokens:", finalAvailable, "calculated:", calculatedAvailable, "max completion:", maxCompletionTokens);
+	
+	// However, for very large available token counts, cap at a reasonable max to prevent rambling
+	const reasonableMax = 4096; // Enough for very thorough responses without being excessive
+	const actualFinal = Math.min(finalAvailable, reasonableMax);
+	
+	console.log("Chat using tokens:", actualFinal, "calculated:", calculatedAvailable, "max completion:", maxCompletionTokens);
 	
 	const response = await fetch(apiUrl, {
 		method: "POST",
@@ -120,7 +127,7 @@ export async function* sendMessage(message) {
 			model: model,
 			messages: chatHistory,
 			temperature: 0.5,
-			max_tokens: finalAvailable,
+			max_tokens: actualFinal,
 			stream: true,
 		}),
 	});