freegate 0.6.6 → 0.6.7

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "freegate",
-  "version": "0.6.6",
+  "version": "0.6.7",
   "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

@@ -4,7 +4,7 @@
     "model": "openai/gpt-oss-120b",
     "priority": 1,
     "dailyLimit": 1000,
-    "keyHint": "console.groq.com \u2192 API Keys",
+    "keyHint": "console.groq.com → API Keys",
     "envVar": "PROVIDER_GROQ_APIKEY",
     "free": true,
     "category": "general"
@@ -14,7 +14,7 @@
     "model": "qwen/qwen3.6-27b",
     "priority": 2,
     "dailyLimit": 1000,
-    "keyHint": "console.groq.com \u2192 API Keys",
+    "keyHint": "console.groq.com → API Keys",
     "envVar": "PROVIDER_GROQ_APIKEY",
     "free": true,
     "category": "general"
@@ -24,7 +24,7 @@
     "model": "allam-2-7b",
     "priority": 3,
     "dailyLimit": 1000,
-    "keyHint": "console.groq.com \u2192 API Keys (allam-2-7b, \u043e\u0447\u0435\u043d\u044c \u0431\u044b\u0441\u0442\u0440\u0430\u044f 229ms)",
+    "keyHint": "console.groq.com → API Keys (allam-2-7b, очень быстрая 229ms)",
     "envVar": "PROVIDER_GROQ_APIKEY",
     "free": true,
     "category": "general"
@@ -34,7 +34,7 @@
     "model": "groq/compound",
     "priority": 12,
     "dailyLimit": 1000,
-    "keyHint": "console.groq.com \u2192 API Keys (compound)",
+    "keyHint": "console.groq.com → API Keys (compound)",
     "envVar": "PROVIDER_GROQ_APIKEY",
     "free": true,
     "category": "general"
@@ -44,7 +44,7 @@
     "model": "codestral-latest",
     "priority": 3,
     "dailyLimit": 500000,
-    "keyHint": "console.mistral.ai \u2192 API Keys",
+    "keyHint": "console.mistral.ai → API Keys",
     "envVar": "PROVIDER_MISTRAL_APIKEY",
     "free": true,
     "category": "coding"
@@ -54,7 +54,7 @@
     "model": "mistral-small-latest",
     "priority": 4,
     "dailyLimit": 500000,
-    "keyHint": "console.mistral.ai \u2192 API Keys",
+    "keyHint": "console.mistral.ai → API Keys",
     "envVar": "PROVIDER_MISTRAL_APIKEY",
     "free": true,
     "category": "general"
@@ -64,7 +64,7 @@
     "model": "gemini-3.6-flash",
     "priority": 5,
     "dailyLimit": 1500,
-    "keyHint": "aistudio.google.com \u2192 Get API key",
+    "keyHint": "aistudio.google.com → Get API key",
     "envVar": "PROVIDER_GEMINI_APIKEY",
     "free": true,
     "category": "general"
@@ -74,7 +74,7 @@
     "model": "deepseek-ai/deepseek-v4-flash-0731",
     "priority": 6,
     "dailyLimit": 40,
-    "keyHint": "build.nvidia.com \u2192 Get API key",
+    "keyHint": "build.nvidia.com → Get API key",
     "envVar": "PROVIDER_NIM_APIKEY",
     "free": true,
     "category": "general"
@@ -84,7 +84,7 @@
     "model": "meta/llama-3.1-8b-instruct",
     "priority": 7,
     "dailyLimit": 40,
-    "keyHint": "build.nvidia.com \u2192 Get API key",
+    "keyHint": "build.nvidia.com → Get API key",
     "envVar": "PROVIDER_NIM_APIKEY",
     "free": true,
     "category": "general"
@@ -94,7 +94,7 @@
     "model": "glm-4.7-flash",
     "priority": 8,
     "dailyLimit": 1000,
-    "keyHint": "open.bigmodel.cn \u2192 API Keys",
+    "keyHint": "open.bigmodel.cn → API Keys",
     "envVar": "PROVIDER_ZAI_APIKEY",
     "free": true,
     "category": "general"
@@ -104,7 +104,7 @@
     "model": "cohere/north-mini-code:free",
     "priority": 17,
     "dailyLimit": 50,
-    "keyHint": "openrouter.ai \u2192 Keys (\u043c\u0435\u0434\u043b\u0435\u043d\u043d\u044b\u0439, \u0437\u0430\u043f\u0430\u0441\u043d\u043e\u0439)",
+    "keyHint": "openrouter.ai → Keys (медленный, запасной)",
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "category": "coding"
@@ -114,7 +114,7 @@
     "model": "z-ai/glm-5.2:free",
     "priority": 10,
     "dailyLimit": 50,
-    "keyHint": "openrouter.ai \u2192 Keys",
+    "keyHint": "openrouter.ai → Keys",
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "category": "general"
@@ -124,7 +124,7 @@
     "model": "nvidia/nemotron-3-ultra-550b-a55b:free",
     "priority": 11,
     "dailyLimit": 50,
-    "keyHint": "openrouter.ai \u2192 Keys",
+    "keyHint": "openrouter.ai → Keys",
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "category": "reasoning"
@@ -134,7 +134,7 @@
     "model": "nvidia/nemotron-3-super-120b-a12b:free",
     "priority": 12,
     "dailyLimit": 50,
-    "keyHint": "openrouter.ai \u2192 Keys",
+    "keyHint": "openrouter.ai → Keys",
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "category": "reasoning"
@@ -144,7 +144,7 @@
     "model": "openai/gpt-oss-20b:free",
     "priority": 13,
     "dailyLimit": 50,
-    "keyHint": "openrouter.ai \u2192 Keys",
+    "keyHint": "openrouter.ai → Keys",
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "category": "general"
@@ -154,7 +154,7 @@
     "model": "stealth/ox-alpha",
     "priority": 3,
     "dailyLimit": 100,
-    "keyHint": "openrouter.ai \u2192 Keys (Ox Alpha: 1M \u043a\u043e\u043d\u0442\u0435\u043a\u0441\u0442, reasoning, \u0431\u0435\u0441\u043f\u043b\u0430\u0442\u043d\u043e)",
+    "keyHint": "openrouter.ai → Keys (Ox Alpha: 1M контекст, reasoning, бесплатно)",
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "reasoning": true,
@@ -165,7 +165,7 @@
     "model": "nvidia/nemotron-3.5-lightning:free",
     "priority": 14,
     "dailyLimit": 100,
-    "keyHint": "openrouter.ai \u2192 Keys (Nemotron 3.5 Lightning: 977K \u043a\u043e\u043d\u0442\u0435\u043a\u0441\u0442)",
+    "keyHint": "openrouter.ai → Keys (Nemotron 3.5 Lightning: 977K контекст)",
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "category": "reasoning"
@@ -175,7 +175,7 @@
     "model": "poolside/laguna-s-2.1:free",
     "priority": 15,
     "dailyLimit": 100,
-    "keyHint": "openrouter.ai \u2192 Keys (Laguna S: \u043a\u043e\u0434\u0438\u043d\u0433, 118B)",
+    "keyHint": "openrouter.ai → Keys (Laguna S: кодинг, 118B)",
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "category": "coding"
@@ -185,7 +185,7 @@
     "model": "gemini-3.6-flash",
     "priority": 5,
     "dailyLimit": 1500,
-    "keyHint": "aistudio.google.com \u2192 Get API key (vision: \u0440\u0430\u0431\u043e\u0442\u0430\u0435\u0442 \u0441\u043e \u0441\u043a\u0440\u0438\u043d\u0448\u043e\u0442\u0430\u043c\u0438)",
+    "keyHint": "aistudio.google.com → Get API key (vision: работает со скриншотами)",
     "envVar": "PROVIDER_GEMINI_APIKEY",
     "free": true,
     "vision": true,
@@ -196,7 +196,7 @@
     "model": "meta/llama-3.2-11b-vision-instruct",
     "priority": 18,
     "dailyLimit": 40,
-    "keyHint": "build.nvidia.com \u2192 Get API key (vision, \u043c\u0435\u0434\u043b\u0435\u043d\u043d\u044b\u0439, \u0437\u0430\u043f\u0430\u0441\u043d\u043e\u0439)",
+    "keyHint": "build.nvidia.com → Get API key (vision, медленный, запасной)",
     "envVar": "PROVIDER_NIM_APIKEY",
     "free": true,
     "vision": true,
@@ -207,7 +207,7 @@
     "model": "qwen2.5:0.5b",
     "priority": 17,
     "dailyLimit": 100000,
-    "keyHint": "\u041b\u043e\u043a\u0430\u043b\u044c\u043d\u0430\u044f \u043c\u043e\u0434\u0435\u043b\u044c. \u0423\u0441\u0442\u0430\u043d\u043e\u0432\u0438 Ollama \u0438 \u0441\u043a\u0430\u0447\u0430\u0439 \u043c\u043e\u0434\u0435\u043b\u044c: ollama pull qwen2.5:0.5b (\u0431\u0435\u0437 \u043a\u043b\u044e\u0447\u0430, \u0431\u0435\u0437\u043b\u0438\u043c\u0438\u0442, \u043f\u0440\u0438\u0432\u0430\u0442\u043d\u043e)",
+    "keyHint": "Локальная модель. Установи Ollama и скачай модель: ollama pull qwen2.5:0.5b (без ключа, безлимит, приватно)",
     "envVar": "",
     "free": true,
     "local": true,
@@ -218,7 +218,7 @@
     "model": "local-model",
     "priority": 18,
     "dailyLimit": 100000,
-    "keyHint": "\u041b\u043e\u043a\u0430\u043b\u044c\u043d\u0430\u044f \u043c\u043e\u0434\u0435\u043b\u044c. \u0423\u0441\u0442\u0430\u043d\u043e\u0432\u0438 LM Studio \u0438 \u0437\u0430\u0433\u0440\u0443\u0437\u0438 \u043c\u043e\u0434\u0435\u043b\u044c (\u0431\u0435\u0437 \u043a\u043b\u044e\u0447\u0430, \u0431\u0435\u0437\u043b\u0438\u043c\u0438\u0442)",
+    "keyHint": "Локальная модель. Установи LM Studio и загрузи модель (без ключа, безлимит)",
     "envVar": "",
     "free": true,
     "local": true,
@@ -229,7 +229,7 @@
     "model": "gpt-oss-120b",
     "priority": 2,
     "dailyLimit": 1000,
-    "keyHint": "cloud.cerebras.ai \u2192 API Keys (gpt-oss-120b, \u043e\u0447\u0435\u043d\u044c \u0431\u044b\u0441\u0442\u0440\u0430\u044f)",
+    "keyHint": "cloud.cerebras.ai → API Keys (gpt-oss-120b, очень быстрая)",
     "envVar": "PROVIDER_CEREBRAS_APIKEY",
     "free": true,
     "category": "general"
@@ -239,7 +239,7 @@
     "model": "gemma-4-31b",
     "priority": 8,
     "dailyLimit": 1000,
-    "keyHint": "cloud.cerebras.ai \u2192 API Keys (gemma-4-31b)",
+    "keyHint": "cloud.cerebras.ai → API Keys (gemma-4-31b)",
     "envVar": "PROVIDER_CEREBRAS_APIKEY",
     "free": true,
     "category": "general"
@@ -249,7 +249,7 @@
     "model": "deepseek-v4-flash",
     "priority": 4,
     "dailyLimit": 1000,
-    "keyHint": "platform.deepseek.com \u2192 API Keys (deepseek-v4-flash)",
+    "keyHint": "platform.deepseek.com → API Keys (deepseek-v4-flash)",
     "envVar": "PROVIDER_DEEPSEEK_APIKEY",
     "free": true,
     "category": "general"
@@ -259,7 +259,7 @@
     "model": "deepseek-v4-flash-vision-exp",
     "priority": 14,
     "dailyLimit": 1000,
-    "keyHint": "platform.deepseek.com \u2192 API Keys (vision)",
+    "keyHint": "platform.deepseek.com → API Keys (vision)",
     "envVar": "PROVIDER_DEEPSEEK_APIKEY",
     "free": true,
     "vision": true,
@@ -270,7 +270,7 @@
     "model": "dots-studio/dots-3-note-preview:free",
     "priority": 19,
     "dailyLimit": 50,
-    "keyHint": "openrouter.ai \u2192 Keys (Dots 3: 280B MoE, \u043c\u043e\u0449\u043d\u0430\u044f)",
+    "keyHint": "openrouter.ai → Keys (Dots 3: 280B MoE, мощная)",
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "category": "reasoning"
@@ -280,7 +280,7 @@
     "model": "liquid/lfm-2.5-2.6b:free",
     "priority": 20,
     "dailyLimit": 50,
-    "keyHint": "openrouter.ai \u2192 Keys (LFM 2.5: \u0431\u044b\u0441\u0442\u0440\u0430\u044f reasoning)",
+    "keyHint": "openrouter.ai → Keys (LFM 2.5: быстрая reasoning)",
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "category": "reasoning"
@@ -290,9 +290,19 @@
     "model": "poolside/laguna-xs-2.1:free",
     "priority": 21,
     "dailyLimit": 50,
-    "keyHint": "openrouter.ai \u2192 Keys (Laguna XS: \u043a\u043e\u0434\u0438\u043d\u0433)",
+    "keyHint": "openrouter.ai → Keys (Laguna XS: кодинг)",
     "envVar": "PROVIDER_OPENROUTER_APIKEY",
     "free": true,
     "category": "coding"
+  },
+  "or-nemotron-3-nano-omni-30b-a3b-r": {
+    "endpoint": "https://openrouter.ai/api/v1/chat/completions",
+    "model": "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free",
+    "priority": 20,
+    "dailyLimit": 50,
+    "keyHint": "openrouter.ai → Keys (автодобавлено, vision)",
+    "envVar": "PROVIDER_OPENROUTER_APIKEY",
+    "free": true,
+    "category": "vision"
   }
-}
+}

package/server.js

@@ -257,7 +257,10 @@ async function handleChatCompletion(req, res, body) {
       let weight = score / lat;
       // Rate-limited providers are last resort — heavy penalty
       if (h.status === 'ratelimited') weight *= 0.05;
-      if (key === targetProviderKey) weight *= 2;
+      // Mapped (target) provider gets a SLIGHT preference, but the pool must
+      // still be able to pick faster/healthier providers for tier aliases —
+      // otherwise Freegate always routes tier-s to Codestral and never varies.
+      if (key === targetProviderKey) weight *= 1.15;
       const dailyLimit = provider.dailyLimit || 1000;
       const usedToday = getStats().providerUsage[key] || 0;
       if (usedToday >= dailyLimit * 0.9) weight *= 0.5;
@@ -279,31 +282,35 @@ async function handleChatCompletion(req, res, body) {
     }).sort((a, b) => b.weight - a.weight);
 
     const totalWeight = scored.reduce((s, p) => s + p.weight, 0);
+    // Weighted random: pick ONE provider as the starting candidate. The rest
+    // are kept as fallbacks below. (Bugfix: was assigning the whole `scored`
+    // list, which made the first (heaviest) provider win every time.)
     let r = Math.random() * totalWeight;
     for (const p of scored) {
       r -= p.weight;
-      if (r <= 0) { selected = scored; break; }
+      if (r <= 0) { selected = [p]; break; }
     }
-    if (selected.length === 0) selected = scored;
+    if (selected.length === 0) selected = [scored[0]];
   }
 
-  const enabledProviders = selected.length > 0 ? selected.map(s => [s.key, s.provider]) :
-    Object.entries(PROVIDERS).filter(([_, p]) => p.enabled)
+  // Weighted-random picked ONE provider as the primary; append the rest of the
+  // healthy pool (by weight) as fallbacks so a failing pick still recovers.
+  const restOfPool = selected.length > 0 && pool.length > 0
+    ? pool.map(([k, p]) => ({ key: k, provider: p })).filter(s => s.key !== selected[0].key)
+      .sort((a, b) => (getHealth()[b.key]?.score || 0) - (getHealth()[a.key]?.score || 0))
+    : [];
+  const enabledProviders = selected.length > 0
+    ? [selected[0]].concat(restOfPool).map(s => [s.key, s.provider])
+    : Object.entries(PROVIDERS).filter(([_, p]) => p.enabled)
       .sort((a, b) => (getHealth()[b[0]]?.score || 50) - (getHealth()[a[0]]?.score || 50));
 
-  // The requested model's mapped provider MUST be tried first — even for
-  // tier aliases. Without this, weighted selection may route the agent to a
-  // slow/empty-streaming provider (e.g. Nemotron-120b) instead of the fast one.
+  // Ensure the requested model's mapped provider is at least IN the candidate
+  // list (it may have been filtered out), but DON'T force it to the front —
+  // the weighted selection above should pick the fastest/healthiest provider.
   if (MODEL_MAP[requestedModel] && PROVIDERS[targetProviderKey]) {
-    // Ensure the target provider is in the candidate list at all
     if (!enabledProviders.some(([k]) => k === targetProviderKey)) {
       enabledProviders.unshift([targetProviderKey, PROVIDERS[targetProviderKey]]);
     }
-    const targetIdx = enabledProviders.findIndex(([k]) => k === targetProviderKey);
-    if (targetIdx > 0) {
-      const [t] = enabledProviders.splice(targetIdx, 1);
-      enabledProviders.unshift(t);
-    }
   }
 
   if (enabledProviders.length === 0) {