freegate 0.6.8 → 0.6.10

package/lib/clean.js

@@ -53,4 +53,16 @@ function cleanDelta(delta) {
   return delta;
 }
 
-module.exports = { stripThink, cleanMessage, cleanDelta, fixReasoningMessage };
+// True if a non-streaming completion actually contains answer text.
+// Used before caching — empty responses (provider glitch) must NOT be cached,
+// otherwise every similar request returns the empty answer forever.
+function hasContent(data) {
+  if (!data || !Array.isArray(data.choices)) return false;
+  const msg = data.choices[0]?.message;
+  if (!msg) return false;
+  const content = typeof msg.content === 'string' ? msg.content.trim() : '';
+  const reasoning = typeof msg.reasoning === 'string' ? msg.reasoning.trim() : '';
+  return content.length > 0 || reasoning.length > 0;
+}
+
+module.exports = { stripThink, cleanMessage, cleanDelta, fixReasoningMessage, hasContent };

package/lib/normalize.js

@@ -45,21 +45,36 @@ function looksLikeCode(messages) {
 
 function normalizeMessages(messages) {
   if (!Array.isArray(messages)) return '';
-  const userOnly = messages.filter(m => m && m.role === 'user');
+  const dialog = messages.filter(m => m && (m.role === 'user' || m.role === 'assistant'));
   if (looksLikeCode(messages)) {
     // Код: точное совпадение, чтобы операторы не схлопывались в один ключ.
-    return JSON.stringify(userOnly);
+    return JSON.stringify(dialog);
   }
-  const userTexts = userOnly
+  // Включаем ВЕСЬ диалог (user + assistant), но НЕ system-промпт: короткие
+  // команды («продолжай») из разных диалогов с разным контекстом не должны
+  // схлопываться в один кэш-ключ и возвращать чужой ответ.
+  const texts = dialog
     .map(m => {
-      if (typeof m.content === 'string') return m.content;
-      if (Array.isArray(m.content)) {
-        return m.content.filter(c => c && c.type === 'text').map(c => c.text || '').join(' ');
+      const role = m.role === 'user' ? 'u' : 'a';
+      let body = '';
+      if (typeof m.content === 'string') body = m.content;
+      else if (Array.isArray(m.content)) {
+        body = m.content.filter(c => c && c.type === 'text').map(c => c.text || '').join(' ');
       }
+      return role + ':' + body;
+    })
+    .join('\n');
+  // Нет извлекаемого текста (только картинки/инструменты) → пусто, чтобы cache
+  // упал на точное совпадение (иначе image-only запросы схлопнутся в один ключ).
+  const rawText = dialog
+    .map(m => {
+      if (typeof m.content === 'string') return m.content;
+      if (Array.isArray(m.content)) return m.content.filter(c => c && c.type === 'text').map(c => c.text || '').join(' ');
       return '';
     })
     .join('\n');
-  return normalizeText(userTexts);
+  if (!rawText.trim()) return '';
+  return normalizeText(texts);
 }
 
 function normalizeText(text) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "freegate",
-  "version": "0.6.8",
+  "version": "0.6.10",
   "description": "Free multi-provider LLM gateway with automatic failover. One OpenAI-compatible endpoint routes to 25 free models (Groq, Mistral, Gemini, NIM, OpenRouter, ZAI, Cerebras, DeepSeek). Never pay for LLMs.",
   "license": "MIT",
   "bin": {

package/providers.json

@@ -304,5 +304,75 @@
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "category": "vision"
+  },
+  "hf-Qwen3-8-2-4T-A95B": {
+    "endpoint": "https://router.huggingface.co/v1/chat/completions",
+    "model": "Qwen/Qwen3.8-2.4T-A95B",
+    "priority": 20,
+    "dailyLimit": 200,
+    "keyHint": "huggingface.co → Settings → Tokens (автодобавлено, general)",
+    "envVar": "PROVIDER_HF_APIKEY",
+    "free": true,
+    "category": "general"
+  },
+  "hf-GLM-5-2": {
+    "endpoint": "https://router.huggingface.co/v1/chat/completions",
+    "model": "zai-org/GLM-5.2",
+    "priority": 20,
+    "dailyLimit": 200,
+    "keyHint": "huggingface.co → Settings → Tokens (автодобавлено, general)",
+    "envVar": "PROVIDER_HF_APIKEY",
+    "free": true,
+    "category": "general"
+  },
+  "hf-gemma-4-31B-it": {
+    "endpoint": "https://router.huggingface.co/v1/chat/completions",
+    "model": "google/gemma-4-31B-it",
+    "priority": 20,
+    "dailyLimit": 200,
+    "keyHint": "huggingface.co → Settings → Tokens (автодобавлено, general)",
+    "envVar": "PROVIDER_HF_APIKEY",
+    "free": true,
+    "category": "general"
+  },
+  "hf-DeepSeek-V4-Pro": {
+    "endpoint": "https://router.huggingface.co/v1/chat/completions",
+    "model": "deepseek-ai/DeepSeek-V4-Pro",
+    "priority": 20,
+    "dailyLimit": 200,
+    "keyHint": "huggingface.co → Settings → Tokens (автодобавлено, general)",
+    "envVar": "PROVIDER_HF_APIKEY",
+    "free": true,
+    "category": "general"
+  },
+  "hf-gemma-4-26B-A4B-it": {
+    "endpoint": "https://router.huggingface.co/v1/chat/completions",
+    "model": "google/gemma-4-26B-A4B-it",
+    "priority": 20,
+    "dailyLimit": 200,
+    "keyHint": "huggingface.co → Settings → Tokens (автодобавлено, general)",
+    "envVar": "PROVIDER_HF_APIKEY",
+    "free": true,
+    "category": "general"
+  },
+  "hf-Llama-3-1-8B-Instruct": {
+    "endpoint": "https://router.huggingface.co/v1/chat/completions",
+    "model": "meta-llama/Llama-3.1-8B-Instruct",
+    "priority": 20,
+    "dailyLimit": 200,
+    "keyHint": "huggingface.co → Settings → Tokens (автодобавлено, general)",
+    "envVar": "PROVIDER_HF_APIKEY",
+    "free": true,
+    "category": "general"
+  },
+  "hf-DeepSeek-V4-Flash": {
+    "endpoint": "https://router.huggingface.co/v1/chat/completions",
+    "model": "deepseek-ai/DeepSeek-V4-Flash",
+    "priority": 20,
+    "dailyLimit": 200,
+    "keyHint": "huggingface.co → Settings → Tokens (автодобавлено, general)",
+    "envVar": "PROVIDER_HF_APIKEY",
+    "free": true,
+    "category": "general"
   }
 }

package/server.js

@@ -8,7 +8,7 @@ const { loadState, initHealth, isCircuitOpen, recordSuccess, recordFailure, reco
 const { checkRateLimit } = require('./lib/rateLimit');
 const { handleDashboard } = require('./lib/dashboard');
 const { acquire, stats: poolStats } = require('./lib/pool');
-const { stripThink, cleanDelta, cleanMessage, fixReasoningMessage } = require('./lib/clean');
+const { stripThink, cleanDelta, cleanMessage, fixReasoningMessage, hasContent } = require('./lib/clean');
 const { classifyComplexity, maybeUpgradeTier } = require('./lib/routing');
 const logger = require('./lib/logger');
 
@@ -388,9 +388,11 @@ async function handleChatCompletion(req, res, body) {
           }
         });
         result.stream.on('end', () => {
-          // Cache the assembled answer for repeat prompts (only if complete)
+          // Cache the assembled answer for repeat prompts (only if complete
+          // AND non-empty — empty answers must not be cached).
           if (chunks.length > 0) {
             const full = chunks.join('');
+            if (full.trim().length > 0) {
             cache.set(effectiveModel, body.messages, body.temperature, {
               id: 'chatcmpl-cached',
               object: 'chat.completion',
@@ -399,6 +401,7 @@ async function handleChatCompletion(req, res, body) {
               choices: [{ index: 0, message: { role: 'assistant', content: full }, finish_reason: 'stop' }],
               usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 },
             });
+            }
           }
         });
         result.stream.on('error', (err) => {
@@ -415,7 +418,7 @@ async function handleChatCompletion(req, res, body) {
             fixReasoningMessage(result.data.choices[0].message);
             cleanMessage(result.data.choices[0].message);
           }
-        cache.set(effectiveModel, body.messages, body.temperature, result.data);
+        if (hasContent(result.data)) cache.set(effectiveModel, body.messages, body.temperature, result.data);
         recordTokens(key, result.usage);
         res.writeHead(200, { 'Content-Type': 'application/json' });
         res.end(JSON.stringify(result.data));
@@ -476,7 +479,7 @@ async function handleChatCompletion(req, res, body) {
           fixReasoningMessage(result.data.choices[0].message);
           cleanMessage(result.data.choices[0].message);
         }
-          cache.set(effectiveModel, body.messages, body.temperature, result.data);
+          if (hasContent(result.data)) cache.set(effectiveModel, body.messages, body.temperature, result.data);
           recordTokens(key, result.usage);
           res.writeHead(200, { 'Content-Type': 'application/json' });
           res.end(JSON.stringify(result.data));