Limita o uso da chave de IA compartilhada

src/app/api/ai/ask/route.ts

@@ -6,6 +6,7 @@ import { getApiUser } from "@/lib/session";
 import { getUserSettings } from "@/lib/app-settings";
 import { resolveModel } from "@/lib/ai/providers";
 import { articleText, errorMessage, languageInstruction } from "@/lib/ai/content";
+import { aiQuotaResponse, allowSharedAiCall } from "@/lib/ai/quota";
 import { accessibleArticles } from "@/lib/article-state";
 
 const MAX_HISTORY = 10;
@@ -39,6 +40,7 @@ export async function POST(request: Request) {
   } catch (err) {
     return Response.json({ error: errorMessage(err, await getLocale()) }, { status: 400 });
   }
+  if (!allowSharedAiCall(user.id, resolved.provider.userId == null)) return aiQuotaResponse();
 
   const result = streamText({
     model: resolved.model,

src/app/api/ai/chat/route.ts

@@ -11,12 +11,16 @@ import { articleText, errorMessage, languageInstruction } from "@/lib/ai/content
 import { parseJsonArray, truncate } from "@/lib/utils";
 import { acquireLock } from "@/lib/jobs/lock";
 import { semanticSearch } from "@/lib/ai/embeddings";
+import { aiQuotaResponse, allowSharedAiCall } from "@/lib/ai/quota";
 
 export const maxDuration = 120;
 
+const MAX_CHAT_MESSAGES = 40;
+const MAX_CHAT_CHARS = 100_000;
+
 const body = z.object({
   threadId: z.string(),
-  messages: z.array(z.custom<UIMessage>((m) => typeof m === "object" && m !== null && "role" in m)),
+  messages: z.array(z.custom<UIMessage>((m) => typeof m === "object" && m !== null && "role" in m)).max(500),
   // Escolhas do compositor (opcionais): provedor, modelo e nível de raciocínio.
   providerId: z.string().optional(),
   model: z.string().max(200).optional(),
@@ -29,7 +33,11 @@ export async function POST(request: Request) {
 
   const parsed = body.safeParse(await request.json().catch(() => null));
   if (!parsed.success) return new Response("Bad request", { status: 400 });
-  const { threadId, messages, providerId, model, reasoning } = parsed.data;
+  const { threadId, providerId, model, reasoning } = parsed.data;
+  const messages = parsed.data.messages
+    .filter((message) => message.role === "user" || message.role === "assistant" || message.role === "system")
+    .slice(-MAX_CHAT_MESSAGES);
+  if (JSON.stringify(messages).length > MAX_CHAT_CHARS) return new Response("Payload too large", { status: 413 });
 
   const thread = await db.chatThread.findFirst({ where: { id: threadId, userId: user.id } });
   if (!thread) return new Response("Not found", { status: 404 });
@@ -43,11 +51,18 @@ export async function POST(request: Request) {
     return Response.json({ error: errorMessage(err, locale) }, { status: 400 });
   }
 
+  if (!allowSharedAiCall(user.id, resolved.provider.userId == null)) return aiQuotaResponse();
+
   const subscribed = { feed: { subscriptions: { some: { userId: user.id } } } };
-  // Artigos da conversa: assinados ou salvos (links avulsos salvos não têm assinatura).
-  const contextArticles = await db.article.findMany({
-    where: { id: { in: parseJsonArray(thread.articleIds) }, ...accessibleArticles(user.id) },
-    take: 5,
+  // Artigos da conversa, na ordem em que foram anexados (no máximo cinco).
+  const contextIds = parseJsonArray(thread.articleIds).slice(0, 5);
+  const contextRows = contextIds.length
+    ? await db.article.findMany({ where: { id: { in: contextIds }, ...accessibleArticles(user.id) } })
+    : [];
+  const contextById = new Map(contextRows.map((article) => [article.id, article]));
+  const contextArticles = contextIds.flatMap((id) => {
+    const article = contextById.get(id);
+    return article ? [article] : [];
   });
 
   const tools = {

src/app/api/ai/speech/route.ts

@@ -7,6 +7,7 @@ import { getApiUser } from "@/lib/session";
 import { articleText, errorMessage } from "@/lib/ai/content";
 import { localizeError, LocalizedError } from "@/lib/localized-error";
 import { mergeAudioParts, requestSpeech, resolveTtsProvider, ttsMimeType } from "@/lib/ai/tts";
+import { aiQuotaResponse, allowSharedAiCall } from "@/lib/ai/quota";
 
 const body = z.object({
   articleId: z.string(),
@@ -28,6 +29,7 @@ export async function POST(request: Request) {
 
   try {
     const provider = await resolveTtsProvider(user.id);
+    if (!allowSharedAiCall(user.id, provider.userId == null)) return aiQuotaResponse();
     const speechText = parsed.data.source === "summary"
       ? parsed.data.text!
           .replace(/\[([^\]]+)]\([^)]+\)/g, "$1")

src/app/api/ai/summarize/route.ts

@@ -7,6 +7,7 @@ import { getApiUser } from "@/lib/session";
 import { getUserSettings } from "@/lib/app-settings";
 import { resolveModel } from "@/lib/ai/providers";
 import { answerInLanguage, articleText, errorMessage, languageInstruction } from "@/lib/ai/content";
+import { aiQuotaResponse, allowSharedAiCall } from "@/lib/ai/quota";
 
 const body = z.object({ articleId: z.string(), force: z.boolean().optional() });
 
@@ -30,6 +31,7 @@ export async function POST(request: Request) {
   } catch (err) {
     return Response.json({ error: errorMessage(err, await getLocale()) }, { status: 400 });
   }
+  if (!allowSharedAiCall(user.id, resolved.provider.userId == null)) return aiQuotaResponse();
 
   const cacheKey = {
     articleId,

src/app/api/ai/translate/route.ts

@@ -8,6 +8,7 @@ import { getUserSettings } from "@/lib/app-settings";
 import { resolveModel } from "@/lib/ai/providers";
 import { sanitizeArticleHtml } from "@/lib/feeds/sanitize";
 import { articleHtml, errorMessage, languageInstruction, languageName } from "@/lib/ai/content";
+import { aiQuotaResponse, allowSharedAiCall } from "@/lib/ai/quota";
 
 const body = z.object({ articleId: z.string(), force: z.boolean().optional() });
 
@@ -31,6 +32,7 @@ export async function POST(request: Request) {
   } catch (err) {
     return Response.json({ error: errorMessage(err, await getLocale()) }, { status: 400 });
   }
+  if (!allowSharedAiCall(user.id, resolved.provider.userId == null)) return aiQuotaResponse();
 
   const cacheKey = {
     articleId,

src/lib/ai/quota.ts

@@ -0,0 +1,27 @@
+/** Cota por usuário quando a chamada usa a chave da instância, não a chave dele. */
+const WINDOW_MS = 10 * 60_000;
+const MAX_CALLS = 30;
+
+const calls = new Map<string, number[]>();
+
+export function allowSharedAiCall(userId: string, usesSharedKey: boolean): boolean {
+  if (!usesSharedKey) return true;
+  const now = Date.now();
+  const recent = (calls.get(userId) ?? []).filter((at) => now - at < WINDOW_MS);
+  if (recent.length >= MAX_CALLS) {
+    calls.set(userId, recent);
+    return false;
+  }
+  recent.push(now);
+  calls.set(userId, recent);
+  if (calls.size > 5000) {
+    for (const [key, stamps] of calls) {
+      if (stamps.every((at) => now - at >= WINDOW_MS)) calls.delete(key);
+    }
+  }
+  return true;
+}
+
+export function aiQuotaResponse() {
+  return new Response("Too many requests", { status: 429, headers: { "retry-after": "60" } });
+}