diff --git a/ConfluentTranslator/public/index.html b/ConfluentTranslator/public/index.html
index 4357336..5cca0e2 100644
--- a/ConfluentTranslator/public/index.html
+++ b/ConfluentTranslator/public/index.html
@@ -884,9 +884,8 @@
         } else {
           // User with limited requests
           counter.style.display = 'block';
-          const used = data.used || 0;
           const limit = data.limit || 20;
-          const remaining = limit - used;
+          const remaining = data.remaining !== undefined ? data.remaining : (limit - (data.used || 0));
           text.textContent = `Requêtes LLM restantes: ${remaining}/${limit}`;
         }
       } catch (error) {
@@ -953,23 +952,8 @@
         }
       });
 
-      // Refresh LLM limit counter every 3 seconds if logged in
-      let limitErrorCount = 0;
-      const limitInterval = setInterval(() => {
-        if (getApiKey()) {
-          updateLLMLimit().catch(err => {
-            limitErrorCount++;
-            // Stop polling after 5 consecutive errors
-            if (limitErrorCount >= 5) {
-              console.warn('Too many errors loading LLM limit, stopping auto-refresh');
-              clearInterval(limitInterval);
-            }
-          });
-        } else {
-          // Reset error count if not logged in
-          limitErrorCount = 0;
-        }
-      }, 3000);
+      // Note: LLM limit counter is updated after each translation
+      // No need for automatic polling every few seconds
     });
 
     // Authenticated fetch wrapper with auto-logout on 401/403
@@ -1421,6 +1405,10 @@
         } else {
           document.getElementById('layer1-content').innerHTML = `<span class="error">Erreur: ${data.error}</span>`;
         }
+
+        // Update LLM limit counter after translation
+        updateLLMLimit().catch(err => console.error('Error updating LLM limit:', err));
+
       } catch (error) {
         document.getElementById('layer1-content').innerHTML = `<span class="error">Erreur: ${error.message}</span>`;
       } finally {
@@ -1533,6 +1521,10 @@
         } else {
           document.getElementById('cf2fr-layer1-content').innerHTML = `<span class="error">Erreur LLM: ${llmData.error}</span>`;
         }
+
+        // Update LLM limit counter after translation
+        updateLLMLimit().catch(err => console.error('Error updating LLM limit:', err));
+
       } catch (error) {
         document.getElementById('cf2fr-layer1-content').innerHTML = `<span class="error">Erreur: ${error.message}</span>`;
       } finally {
diff --git a/ConfluentTranslator/rateLimiter.js b/ConfluentTranslator/rateLimiter.js
index 5dfc2f7..83861c7 100644
--- a/ConfluentTranslator/rateLimiter.js
+++ b/ConfluentTranslator/rateLimiter.js
@@ -1,32 +1,8 @@
 const rateLimit = require('express-rate-limit');
 
-// Rate limiter global par IP
-const globalLimiter = rateLimit({
-  windowMs: 15 * 60 * 1000, // 15 minutes
-  max: 200, // max 200 requêtes par IP
-  standardHeaders: true,
-  legacyHeaders: false,
-  skip: (req) => {
-    // Skip pour les endpoints qui doivent être appelés très fréquemment
-    return req.path === '/api/llm/limit';
-  },
-  message: { error: 'Too many requests from this IP, please try again later.' }
-});
-
-// Rate limiter pour les traductions (plus strict)
-const translationLimiter = rateLimit({
-  windowMs: 60 * 1000, // 1 minute
-  max: 10, // max 10 traductions par minute
-  standardHeaders: true,
-  legacyHeaders: false,
-  skip: (req) => {
-    // Skip si l'utilisateur est admin
-    return req.user && req.user.role === 'admin';
-  },
-  message: { error: 'Too many translation requests. Please wait a moment.' }
-});
-
 // Rate limiter pour les endpoints sensibles (admin)
+// Note: Pour les traductions et requêtes LLM, on utilise checkLLMLimit() dans auth.js
+// qui gère les limites par API key (plus flexible et précis que les rate limiters par IP)
 const adminLimiter = rateLimit({
   windowMs: 5 * 60 * 1000, // 5 minutes
   max: 50,
@@ -36,7 +12,5 @@ const adminLimiter = rateLimit({
 });
 
 module.exports = {
-  globalLimiter,
-  translationLimiter,
   adminLimiter
 };
diff --git a/ConfluentTranslator/server.js b/ConfluentTranslator/server.js
index 3352e5f..4e2b854 100644
--- a/ConfluentTranslator/server.js
+++ b/ConfluentTranslator/server.js
@@ -17,7 +17,7 @@ const { translateConfluentToFrench, translateConfluentDetailed } = require('./co
 
 // Security modules
 const { authenticate, requireAdmin, createToken, listTokens, disableToken, enableToken, deleteToken, getGlobalStats, trackLLMUsage, checkLLMLimit } = require('./auth');
-const { globalLimiter, translationLimiter, adminLimiter } = require('./rateLimiter');
+const { adminLimiter } = require('./rateLimiter');
 const { requestLogger, getLogs, getLogStats } = require('./logger');
 
 const app = express();
@@ -26,7 +26,7 @@ const PORT = process.env.PORT || 3000;
 // Middlewares
 app.use(express.json());
 app.use(requestLogger);      // Log toutes les requêtes
-app.use(globalLimiter);       // Rate limiting global
+// Rate limiting: on utilise uniquement checkLLMLimit() par API key, pas de rate limit global par IP
 
 // Route protégée pour admin.html (AVANT express.static)
 // Vérifie l'auth seulement si API key présente, sinon laisse passer (le JS client vérifiera)
@@ -378,7 +378,7 @@ app.post('/api/analyze/coverage', authenticate, (req, res) => {
 });
 
 // Translation endpoint (NOUVEAU SYSTÈME CONTEXTUEL)
-app.post('/translate', authenticate, translationLimiter, async (req, res) => {
+app.post('/translate', authenticate, async (req, res) => {
   const { text, target, provider, model, temperature = 1.0, useLexique = true, customAnthropicKey, customOpenAIKey } = req.body;
 
   if (!text || !target || !provider || !model) {
@@ -586,7 +586,7 @@ function parseTranslationResponse(response) {
 }
 
 // Raw translation endpoint (for debugging - returns unprocessed LLM output) - SECURED
-app.post('/api/translate/raw', authenticate, translationLimiter, async (req, res) => {
+app.post('/api/translate/raw', authenticate, async (req, res) => {
   const { text, target, provider, model, useLexique = true, customAnthropicKey, customOpenAIKey } = req.body;
 
   if (!text || !target || !provider || !model) {
@@ -700,7 +700,7 @@ app.post('/api/translate/raw', authenticate, translationLimiter, async (req, res
 });
 
 // Batch translation endpoint - SECURED
-app.post('/api/translate/batch', authenticate, translationLimiter, async (req, res) => {
+app.post('/api/translate/batch', authenticate, async (req, res) => {
   const { words, target = 'ancien' } = req.body;
 
   if (!words || !Array.isArray(words)) {
@@ -727,7 +727,7 @@ app.post('/api/translate/batch', authenticate, translationLimiter, async (req, r
 });
 
 // Confluent → French translation endpoint (traduction brute) - SECURED
-app.post('/api/translate/conf2fr', authenticate, translationLimiter, (req, res) => {
+app.post('/api/translate/conf2fr', authenticate, (req, res) => {
   const { text, variant = 'ancien', detailed = false } = req.body;
 
   if (!text) {
@@ -755,7 +755,7 @@ app.post('/api/translate/conf2fr', authenticate, translationLimiter, (req, res)
 });
 
 // NEW: Confluent → French with LLM refinement
-app.post('/api/translate/conf2fr/llm', authenticate, translationLimiter, async (req, res) => {
+app.post('/api/translate/conf2fr/llm', authenticate, async (req, res) => {
   const { text, variant = 'ancien', provider = 'anthropic', model = 'claude-sonnet-4-20250514', customAnthropicKey, customOpenAIKey } = req.body;
 
   if (!text) {