freegate 0.6.18 → 0.6.20

package/providers.json

@@ -370,5 +370,269 @@
     "envVar": "PROVIDER_MISTRAL_APIKEY",
     "free": true,
     "category": "coding"
+  },
+  "sambanova-gpt-oss-120b": {
+    "endpoint": "https://api.sambanova.ai/v1/chat/completions",
+    "model": "gpt-oss-120b",
+    "priority": 12,
+    "dailyLimit": 100000,
+    "keyHint": "cloud.sambanova.ai → API Keys",
+    "envVar": "PROVIDER_SAMBANOVA_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 131072
+  },
+  "sambanova-minimax-m2": {
+    "endpoint": "https://api.sambanova.ai/v1/chat/completions",
+    "model": "MiniMax-M2.7",
+    "priority": 13,
+    "dailyLimit": 100000,
+    "keyHint": "cloud.sambanova.ai → API Keys",
+    "envVar": "PROVIDER_SAMBANOVA_APIKEY",
+    "free": true,
+    "category": "reasoning",
+    "context_window": 131072
+  },
+  "sambanova-deepseek-v3": {
+    "endpoint": "https://api.sambanova.ai/v1/chat/completions",
+    "model": "DeepSeek-V3.2",
+    "priority": 13,
+    "dailyLimit": 100000,
+    "keyHint": "cloud.sambanova.ai → API Keys",
+    "envVar": "PROVIDER_SAMBANOVA_APIKEY",
+    "free": true,
+    "category": "reasoning",
+    "context_window": 131072
+  },
+  "sambanova-llama4-maverick": {
+    "endpoint": "https://api.sambanova.ai/v1/chat/completions",
+    "model": "Llama-4-Maverick-17B-128E-Instruct",
+    "priority": 13,
+    "dailyLimit": 100000,
+    "keyHint": "cloud.sambanova.ai → API Keys",
+    "envVar": "PROVIDER_SAMBANOVA_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 1000000
+  },
+  "sambanova-llama3.3-70b": {
+    "endpoint": "https://api.sambanova.ai/v1/chat/completions",
+    "model": "Meta-Llama-3.3-70B-Instruct",
+    "priority": 14,
+    "dailyLimit": 100000,
+    "keyHint": "cloud.sambanova.ai → API Keys",
+    "envVar": "PROVIDER_SAMBANOVA_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 131072
+  },
+  "siliconflow-deepseek-r1": {
+    "endpoint": "https://api.siliconflow.com/v1/chat/completions",
+    "model": "deepseek-ai/DeepSeek-R1",
+    "priority": 12,
+    "dailyLimit": 100000,
+    "keyHint": "cloud.siliconflow.com → API Keys",
+    "envVar": "PROVIDER_SILICONFLOW_APIKEY",
+    "free": true,
+    "category": "reasoning",
+    "context_window": 131072
+  },
+  "siliconflow-qwen3-coder": {
+    "endpoint": "https://api.siliconflow.com/v1/chat/completions",
+    "model": "Qwen/Qwen3-Coder-480B-A35B-Instruct",
+    "priority": 12,
+    "dailyLimit": 100000,
+    "keyHint": "cloud.siliconflow.com → API Keys",
+    "envVar": "PROVIDER_SILICONFLOW_APIKEY",
+    "free": true,
+    "category": "coding",
+    "context_window": 131072
+  },
+  "siliconflow-qwen3-235b": {
+    "endpoint": "https://api.siliconflow.com/v1/chat/completions",
+    "model": "Qwen/Qwen3-235B-A22B-Instruct-2507",
+    "priority": 13,
+    "dailyLimit": 100000,
+    "keyHint": "cloud.siliconflow.com → API Keys",
+    "envVar": "PROVIDER_SILICONFLOW_APIKEY",
+    "free": true,
+    "category": "reasoning",
+    "context_window": 131072
+  },
+  "siliconflow-kimi-k2": {
+    "endpoint": "https://api.siliconflow.com/v1/chat/completions",
+    "model": "moonshotai/Kimi-K2.5",
+    "priority": 13,
+    "dailyLimit": 100000,
+    "keyHint": "cloud.siliconflow.com → API Keys",
+    "envVar": "PROVIDER_SILICONFLOW_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 131072
+  },
+  "siliconflow-glm": {
+    "endpoint": "https://api.siliconflow.com/v1/chat/completions",
+    "model": "zai-org/GLM-4.7",
+    "priority": 14,
+    "dailyLimit": 100000,
+    "keyHint": "cloud.siliconflow.com → API Keys",
+    "envVar": "PROVIDER_SILICONFLOW_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 131072
+  },
+  "deepinfra-gpt-oss-120b": {
+    "endpoint": "https://api.deepinfra.com/v1/openai/chat/completions",
+    "model": "openai/gpt-oss-120b",
+    "priority": 12,
+    "dailyLimit": 100000,
+    "keyHint": "deepinfra.com → API Keys",
+    "envVar": "PROVIDER_DEEPINFRA_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 131072
+  },
+  "deepinfra-gpt-oss-20b": {
+    "endpoint": "https://api.deepinfra.com/v1/openai/chat/completions",
+    "model": "openai/gpt-oss-20b",
+    "priority": 13,
+    "dailyLimit": 100000,
+    "keyHint": "deepinfra.com → API Keys",
+    "envVar": "PROVIDER_DEEPINFRA_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 131072
+  },
+  "deepinfra-nemotron-super-120b": {
+    "endpoint": "https://api.deepinfra.com/v1/openai/chat/completions",
+    "model": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B",
+    "priority": 12,
+    "dailyLimit": 100000,
+    "keyHint": "deepinfra.com → API Keys",
+    "envVar": "PROVIDER_DEEPINFRA_APIKEY",
+    "free": true,
+    "category": "reasoning",
+    "context_window": 131072
+  },
+  "deepinfra-llama-4-maverick": {
+    "endpoint": "https://api.deepinfra.com/v1/openai/chat/completions",
+    "model": "meta-llama/Llama-4-Maverick-17B-128E-Instruct",
+    "priority": 14,
+    "dailyLimit": 100000,
+    "keyHint": "deepinfra.com → API Keys",
+    "envVar": "PROVIDER_DEEPINFRA_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 1000000
+  },
+  "hyperbolic-deepseek-v3": {
+    "endpoint": "https://api.hyperbolic.xyz/v1/chat/completions",
+    "model": "deepseek-ai/DeepSeek-V3",
+    "priority": 14,
+    "dailyLimit": 100000,
+    "keyHint": "app.hyperbolic.xyz → API Keys",
+    "envVar": "PROVIDER_HYPERBOLIC_APIKEY",
+    "free": true,
+    "category": "reasoning",
+    "context_window": 131072
+  },
+  "hyperbolic-qwen3-coder": {
+    "endpoint": "https://api.hyperbolic.xyz/v1/chat/completions",
+    "model": "Qwen/Qwen2.5-Coder-32B-Instruct",
+    "priority": 14,
+    "dailyLimit": 100000,
+    "keyHint": "app.hyperbolic.xyz → API Keys",
+    "envVar": "PROVIDER_HYPERBOLIC_APIKEY",
+    "free": true,
+    "category": "coding",
+    "context_window": 131072
+  },
+  "hyperbolic-llama-70b": {
+    "endpoint": "https://api.hyperbolic.xyz/v1/chat/completions",
+    "model": "meta-llama/Llama-3.3-70B-Instruct",
+    "priority": 15,
+    "dailyLimit": 100000,
+    "keyHint": "app.hyperbolic.xyz → API Keys",
+    "envVar": "PROVIDER_HYPERBOLIC_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 131072
+  },
+  "cohere-command-a": {
+    "endpoint": "https://api.cohere.com/compatibility/v1/chat/completions",
+    "model": "command-a-03-2025",
+    "priority": 15,
+    "dailyLimit": 100000,
+    "keyHint": "dashboard.cohere.com → API Keys",
+    "envVar": "PROVIDER_COHERE_APIKEY",
+    "free": true,
+    "category": "reasoning",
+    "context_window": 131072
+  },
+  "cohere-command-r-plus": {
+    "endpoint": "https://api.cohere.com/compatibility/v1/chat/completions",
+    "model": "command-r-plus-08-2024",
+    "priority": 16,
+    "dailyLimit": 100000,
+    "keyHint": "dashboard.cohere.com → API Keys",
+    "envVar": "PROVIDER_COHERE_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 131072
+  },
+  "llm7-gpt4o-mini": {
+    "endpoint": "https://api.llm7.io/v1/chat/completions",
+    "model": "gpt-4o-mini-2024-07-18",
+    "priority": 13,
+    "dailyLimit": 100000,
+    "keyHint": "llm7.io (бесплатный ключ при регистрации)",
+    "envVar": "PROVIDER_LLM7_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 131072
+  },
+  "llm7-deepseek-r1": {
+    "endpoint": "https://api.llm7.io/v1/chat/completions",
+    "model": "deepseek-r1-0528",
+    "priority": 13,
+    "dailyLimit": 100000,
+    "keyHint": "llm7.io (бесплатный ключ при регистрации)",
+    "envVar": "PROVIDER_LLM7_APIKEY",
+    "free": true,
+    "category": "reasoning",
+    "context_window": 131072
+  },
+  "nara-mistral-large": {
+    "endpoint": "https://router.bynara.id/v1/chat/completions",
+    "model": "mistral-large",
+    "priority": 13,
+    "dailyLimit": 100000,
+    "keyHint": "nara — бесплатный ключ через Telegram-канал",
+    "envVar": "PROVIDER_NARA_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 252000
+  },
+  "nara-mistral-medium": {
+    "endpoint": "https://router.bynara.id/v1/chat/completions",
+    "model": "mistral-medium-3-5",
+    "priority": 14,
+    "dailyLimit": 100000,
+    "keyHint": "nara — бесплатный ключ через Telegram-канал",
+    "envVar": "PROVIDER_NARA_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 256000
+  },
+  "nara-tencent-hy3": {
+    "endpoint": "https://router.bynara.id/v1/chat/completions",
+    "model": "tencent-hy3",
+    "priority": 14,
+    "dailyLimit": 100000,
+    "keyHint": "nara — бесплатный ключ через Telegram-канал",
+    "envVar": "PROVIDER_NARA_APIKEY",
+    "free": true,
+    "category": "general",
+    "context_window": 1000000
   }
 }

package/server.js

@@ -4,7 +4,7 @@ const fs = require('fs');
 const path = require('path');
 const { LRUCache } = require('./lib/cache');
 const { PROVIDERS, MODEL_MAP, callProvider, reloadProviders } = require('./lib/providers');
-const { loadState, initHealth, isCircuitOpen, recordSuccess, recordFailure, recordRequest, recordTokens, getHealth, getStats, getReliability, recordRecent, recordRpm, getRecent, getRpm, recordSelection, getLastSelection, getBandit, recordBandit, warmBanditPriors, getContextStats } = require('./lib/health');
+const { loadState, initHealth, isCircuitOpen, recordSuccess, recordFailure, recordRequest, recordTokens, getHealth, getStats, getReliability, recordRecent, recordRpm, getRecent, getRpm, recordSelection, getLastSelection, getBandit, recordBandit, warmBanditPriors, getContextStats, getHourly } = require('./lib/health');
 const { checkRateLimit } = require('./lib/rateLimit');
 const { handleDashboard } = require('./lib/dashboard');
 const { acquire, stats: poolStats } = require('./lib/pool');
@@ -99,6 +99,30 @@ const METHODOLOGY_CONFIG = Object.assign(
   (config.methodology && typeof config.methodology === 'object') ? config.methodology : {}
 );
 
+// --- Самопроверка ответа второй моделью (vetting) ---
+// Опционально: после не-stream ответа отправляем краткий чек другой модели.
+// Выключено по умолчанию (жжёт 2-й free-лимит). Включается config.vetting.enabled.
+const { shouldVet, vetAnswer, VETTING_DEFAULTS } = require('./lib/vetting');
+const VETTING_CONFIG = Object.assign(
+  { ...VETTING_DEFAULTS },
+  (config.vetting && typeof config.vetting === 'object') ? config.vetting : {}
+);
+
+// --- Стратегия роутинга ---
+// Опциональные модификаторы равномерности (round-robin / least-used).
+// По умолчанию 'weighted' — поведение без изменений.
+const { makeWeightModifier } = require('./lib/strategy');
+const ROUTING_STRATEGY = (config.routing && config.routing.strategy) || 'weighted';
+
+// --- Сжатие промпта (Caveman-стиль) ---
+// Опционально убирает вежливость/заполнители из последнего user-сообщения,
+// экономя токены. config.compress.enabled=true включает.
+const { compressMessages } = require('./lib/compress');
+const COMPRESS_CONFIG = Object.assign(
+  { enabled: false, minLen: 60 },
+  (config.compress && typeof config.compress === 'object') ? config.compress : {}
+);
+
 // --- Веб-поиск для search-задач ---
 // Бесплатный поиск фактов (DuckDuckGo, без ключа) для запросов-поиска, чтобы
 // модель не галлюцинировала («что такое минимакс дизайн» → реальная инфа про
@@ -459,6 +483,15 @@ async function handleChatCompletion(req, res, body) {
   if (METHODOLOGY_CONFIG.enabled && Array.isArray(body.messages)) {
     try {
       taskCategory = classifyTask(body.messages);
+      // Сжатие промпта: убираем вежливость/заполнители ДО методолога (методолог
+      // не должен суммировать сжатый текст). Опционально (config.compress).
+      if (COMPRESS_CONFIG.enabled) {
+        const compressed = compressMessages(body.messages, COMPRESS_CONFIG);
+        if (compressed !== body.messages && Array.isArray(compressed)) {
+          body.messages = compressed;
+          measure.compressed = 1;
+        }
+      }
       const injected = injectMethodology(body.messages, taskCategory, METHODOLOGY_CONFIG);
       if (injected !== body.messages) {
         body.messages = injected;
@@ -608,6 +641,9 @@ async function handleChatCompletion(req, res, body) {
         .sort((a, b) => (b.provider.context_window || 0) - (a.provider.context_window || 0))
         .slice(0, 1);
     } else {
+      const usedTodayList = {};
+      for (const [k] of pool) usedTodayList[k] = usedTodayFor(k);
+      const strategyModifier = makeWeightModifier(ROUTING_STRATEGY, { keys: pool.map(([k]) => k), usedTodayList });
       const scored = pool.map(([key, provider]) => {
         const h = getHealth()[key];
         let score = h.score || 50;
@@ -640,6 +676,9 @@ async function handleChatCompletion(req, res, body) {
         const usedToday = usedTodayFor(key);
         if (dailyLimit > 0 && usedToday >= dailyLimit) weight *= 0.03;
         else if (dailyLimit > 0 && usedToday >= dailyLimit * 0.9) weight *= 0.4;
+        // Стратегия роутинга: равномерность (round-robin / least-used) как
+        // лёгкий модификатор к базовому weight — не ломает основной скоринг.
+        weight *= strategyModifier(key);
         return { key, provider, weight };
       });
 
@@ -836,7 +875,7 @@ async function handleChatCompletion(req, res, body) {
                 model: provider.model,
                 choices: [{ index: 0, message: { role: 'assistant', content: full }, finish_reason: 'stop' }],
                 usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 },
-              });
+              }, key);
             }
             commit(200);
             res.end();
@@ -946,7 +985,7 @@ async function handleChatCompletion(req, res, body) {
               model: provider.model,
               choices: [{ index: 0, message: { role: 'assistant', content: full }, finish_reason: 'stop' }],
               usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 },
-            });
+            }, key);
           }
           commit(200);
           res.end();
@@ -971,8 +1010,31 @@ async function handleChatCompletion(req, res, body) {
         measure.provider = key;
         measure.real = (result.data.usage && result.data.usage.prompt_tokens) ? result.data.usage.prompt_tokens : measure.sentTokens || 0;
         measure.win = PROVIDERS[key]?.context_window || 0;
+        // Самопроверка второй моделью (опционально): только не-stream, только по конфигу.
+        if (VETTING_CONFIG.enabled && !isStreaming) {
+          const answer = result.data?.choices?.[0]?.message?.content || '';
+          if (shouldVet({ config: VETTING_CONFIG, complexity, category: taskCategory, answerLen: answer.length })) {
+            const picks = Object.entries(PROVIDERS)
+              .filter(([pk, p]) => p.enabled && pk !== key && !isCircuitOpen(pk) &&
+                getHealth()[pk]?.status === 'up' && p.vision !== true)
+              .map(([_, p]) => p);
+            try {
+              const verdict = await vetAnswer({ answer, callProvider, picks, config: VETTING_CONFIG });
+              if (verdict.checked && !verdict.ok && verdict.note) {
+                const msg = result.data.choices[0].message;
+                msg.content = (msg.content || '') + '\n\n> ⚠️ Проверка второй моделью: ' + verdict.note;
+                measure.vetted = 1;
+                measure.vetNote = verdict.note;
+              } else {
+                measure.vetted = 0;
+              }
+            } catch (vetErr) {
+              measure.vetted = 0;
+            }
+          }
+        }
         commit(200);
-        cache.set(effectiveModel, body.messages, body.temperature, result.data);
+        cache.set(effectiveModel, body.messages, body.temperature, result.data, key);
         recordTokens(key, result.usage);
         res.writeHead(200, { 'Content-Type': 'application/json' });
         res.end(JSON.stringify(result.data));
@@ -1059,7 +1121,7 @@ if (isTooShort(result.data, lastUserText(body.messages))) {
           measure.real = (result.data.usage && result.data.usage.prompt_tokens) ? result.data.usage.prompt_tokens : measure.sentTokens || 0;
           measure.win = PROVIDERS[key]?.context_window || 0;
           commit(200);
-          cache.set(effectiveModel, body.messages, body.temperature, result.data);
+          cache.set(effectiveModel, body.messages, body.temperature, result.data, key);
           recordTokens(key, result.usage);
           res.writeHead(200, { 'Content-Type': 'application/json' });
           res.end(JSON.stringify(result.data));
@@ -1117,7 +1179,7 @@ if (isTooShort(result.data, lastUserText(body.messages))) {
                 model: provider.model,
                 choices: [{ index: 0, message: { role: 'assistant', content: full }, finish_reason: 'stop' }],
                 usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 },
-              });
+              }, key);
             }
             commit(200);
             res.end();
@@ -1346,6 +1408,7 @@ const server = http.createServer(async (req, res) => {
         return [k, { status: v.status, score: v.score, latency_ms: v.latency, reason, reliability }];
       })),
       cache: cache.stats(),
+      hourly: (() => { try { return getHourly(); } catch { return []; } })(),
       limits,
       pool: poolStats(),
       last_selection: getLastSelection(),