@everworker/oneringai 0.4.5 → 0.4.7

package/dist/index.cjs

@@ -18,6 +18,7 @@ var spawn = require('cross-spawn');
 var process2 = require('process');
 var stream = require('stream');
 var fs17 = require('fs/promises');
+var events = require('events');
 var simpleIcons = require('simple-icons');
 var child_process = require('child_process');
 var util = require('util');
@@ -12631,18 +12632,30 @@ function isVendor(value) {
 // src/domain/entities/Model.ts
 var LLM_MODELS = {
   [Vendor.OpenAI]: {
+    // GPT-5.3 Series
+    GPT_5_3_CODEX: "gpt-5.3-codex",
+    GPT_5_3_CHAT: "gpt-5.3-chat-latest",
     // GPT-5.2 Series (Current Flagship)
     GPT_5_2: "gpt-5.2",
     GPT_5_2_PRO: "gpt-5.2-pro",
+    GPT_5_2_CODEX: "gpt-5.2-codex",
+    GPT_5_2_CHAT: "gpt-5.2-chat-latest",
+    // GPT-5.1 Series
+    GPT_5_1: "gpt-5.1",
+    GPT_5_1_CODEX: "gpt-5.1-codex",
+    GPT_5_1_CODEX_MAX: "gpt-5.1-codex-max",
+    GPT_5_1_CODEX_MINI: "gpt-5.1-codex-mini",
+    GPT_5_1_CHAT: "gpt-5.1-chat-latest",
     // GPT-5 Series
     GPT_5: "gpt-5",
     GPT_5_MINI: "gpt-5-mini",
     GPT_5_NANO: "gpt-5-nano",
+    GPT_5_CHAT: "gpt-5-chat-latest",
     // GPT-4.1 Series
     GPT_4_1: "gpt-4.1",
     GPT_4_1_MINI: "gpt-4.1-mini",
     GPT_4_1_NANO: "gpt-4.1-nano",
-    // GPT-4o Series (Legacy, Audio Capable)
+    // GPT-4o Series (Legacy)
     GPT_4O: "gpt-4o",
     GPT_4O_MINI: "gpt-4o-mini",
     // Reasoning Models (o-series)
@@ -12650,18 +12663,26 @@ var LLM_MODELS = {
     O1: "o1"
   },
   [Vendor.Anthropic]: {
-    // Claude 4.5 Series (Current)
+    // Claude 4.6 Series (Current)
+    CLAUDE_OPUS_4_6: "claude-opus-4-6",
+    CLAUDE_SONNET_4_6: "claude-sonnet-4-6",
+    // Claude 4.5 Series
     CLAUDE_OPUS_4_5: "claude-opus-4-5-20251101",
     CLAUDE_SONNET_4_5: "claude-sonnet-4-5-20250929",
     CLAUDE_HAIKU_4_5: "claude-haiku-4-5-20251001",
     // Claude 4.x Legacy
     CLAUDE_OPUS_4_1: "claude-opus-4-1-20250805",
+    CLAUDE_OPUS_4: "claude-opus-4-20250514",
     CLAUDE_SONNET_4: "claude-sonnet-4-20250514",
     CLAUDE_SONNET_3_7: "claude-3-7-sonnet-20250219",
-    // Claude 3.x Legacy
+    // Claude 3.x Legacy (Deprecated)
     CLAUDE_HAIKU_3: "claude-3-haiku-20240307"
   },
   [Vendor.Google]: {
+    // Gemini 3.1 Series (Preview)
+    GEMINI_3_1_PRO_PREVIEW: "gemini-3.1-pro-preview",
+    GEMINI_3_1_FLASH_LITE_PREVIEW: "gemini-3.1-flash-lite-preview",
+    GEMINI_3_1_FLASH_IMAGE_PREVIEW: "gemini-3.1-flash-image-preview",
     // Gemini 3 Series (Preview)
     GEMINI_3_FLASH_PREVIEW: "gemini-3-flash-preview",
     GEMINI_3_PRO_PREVIEW: "gemini-3-pro-preview",
@@ -12693,12 +12714,91 @@ var MODEL_REGISTRY = {
   // ============================================================================
   // OpenAI Models (Verified from platform.openai.com)
   // ============================================================================
+  // GPT-5.3 Series
+  "gpt-5.3-codex": {
+    name: "gpt-5.3-codex",
+    provider: Vendor.OpenAI,
+    description: "Latest codex model for coding and agentic tasks. Reasoning.effort: low, medium, high, xhigh",
+    isActive: true,
+    releaseDate: "2026-02-01",
+    knowledgeCutoff: "2025-08-31",
+    features: {
+      reasoning: true,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      batchAPI: true,
+      promptCaching: true,
+      parameters: {
+        temperature: false,
+        topP: false,
+        frequencyPenalty: false,
+        presencePenalty: false
+      },
+      input: {
+        tokens: 4e5,
+        text: true,
+        image: true,
+        cpm: 1.75,
+        cpmCached: 0.175
+      },
+      output: {
+        tokens: 128e3,
+        text: true,
+        cpm: 14
+      }
+    }
+  },
+  "gpt-5.3-chat-latest": {
+    name: "gpt-5.3-chat-latest",
+    provider: Vendor.OpenAI,
+    description: "Latest GPT-5.3 chat model for general-purpose use",
+    isActive: true,
+    releaseDate: "2026-02-01",
+    knowledgeCutoff: "2025-08-31",
+    features: {
+      reasoning: false,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      batchAPI: true,
+      promptCaching: true,
+      parameters: {
+        temperature: false
+      },
+      input: {
+        tokens: 128e3,
+        text: true,
+        image: true,
+        cpm: 1.75,
+        cpmCached: 0.175
+      },
+      output: {
+        tokens: 16e3,
+        text: true,
+        cpm: 14
+      }
+    }
+  },
   // GPT-5.2 Series (Current Flagship)
   "gpt-5.2": {
     name: "gpt-5.2",
     provider: Vendor.OpenAI,
     description: "Flagship model for coding and agentic tasks. Reasoning.effort: none, low, medium, high, xhigh",
     isActive: true,
+    preferred: true,
     releaseDate: "2025-12-01",
     knowledgeCutoff: "2025-08-31",
     features: {
@@ -12724,7 +12824,8 @@ var MODEL_REGISTRY = {
         tokens: 4e5,
         text: true,
         image: true,
-        cpm: 1.75
+        cpm: 1.75,
+        cpmCached: 0.175
       },
       output: {
         tokens: 128e3,
@@ -12743,7 +12844,7 @@ var MODEL_REGISTRY = {
     features: {
       reasoning: true,
       streaming: true,
-      structuredOutput: true,
+      structuredOutput: false,
       functionCalling: true,
       fineTuning: false,
       predictedOutputs: false,
@@ -12772,6 +12873,276 @@ var MODEL_REGISTRY = {
       }
     }
   },
+  "gpt-5.2-codex": {
+    name: "gpt-5.2-codex",
+    provider: Vendor.OpenAI,
+    description: "GPT-5.2 codex for coding and agentic tasks. Reasoning.effort: low, medium, high, xhigh",
+    isActive: true,
+    preferred: true,
+    releaseDate: "2025-12-01",
+    knowledgeCutoff: "2025-08-31",
+    features: {
+      reasoning: true,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      batchAPI: true,
+      promptCaching: true,
+      parameters: {
+        temperature: false,
+        topP: false,
+        frequencyPenalty: false,
+        presencePenalty: false
+      },
+      input: {
+        tokens: 4e5,
+        text: true,
+        image: true,
+        cpm: 1.75,
+        cpmCached: 0.175
+      },
+      output: {
+        tokens: 128e3,
+        text: true,
+        cpm: 14
+      }
+    }
+  },
+  "gpt-5.2-chat-latest": {
+    name: "gpt-5.2-chat-latest",
+    provider: Vendor.OpenAI,
+    description: "GPT-5.2 chat model for general-purpose use",
+    isActive: true,
+    releaseDate: "2025-12-01",
+    knowledgeCutoff: "2025-08-31",
+    features: {
+      reasoning: false,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      batchAPI: true,
+      promptCaching: true,
+      input: {
+        tokens: 128e3,
+        text: true,
+        image: true,
+        cpm: 1.75,
+        cpmCached: 0.175
+      },
+      output: {
+        tokens: 16e3,
+        text: true,
+        cpm: 14
+      }
+    }
+  },
+  // GPT-5.1 Series
+  "gpt-5.1": {
+    name: "gpt-5.1",
+    provider: Vendor.OpenAI,
+    description: "Intelligent reasoning model for coding and agentic tasks. Reasoning.effort: none, low, medium, high",
+    isActive: true,
+    releaseDate: "2025-10-01",
+    knowledgeCutoff: "2024-09-30",
+    features: {
+      reasoning: true,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      batchAPI: true,
+      promptCaching: true,
+      parameters: {
+        temperature: false,
+        topP: false,
+        frequencyPenalty: false,
+        presencePenalty: false
+      },
+      input: {
+        tokens: 4e5,
+        text: true,
+        image: true,
+        cpm: 1.25,
+        cpmCached: 0.125
+      },
+      output: {
+        tokens: 128e3,
+        text: true,
+        cpm: 10
+      }
+    }
+  },
+  "gpt-5.1-codex": {
+    name: "gpt-5.1-codex",
+    provider: Vendor.OpenAI,
+    description: "GPT-5.1 codex for coding and agentic tasks with reasoning",
+    isActive: true,
+    releaseDate: "2025-10-01",
+    knowledgeCutoff: "2024-09-30",
+    features: {
+      reasoning: true,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      batchAPI: true,
+      promptCaching: true,
+      parameters: {
+        temperature: false,
+        topP: false,
+        frequencyPenalty: false,
+        presencePenalty: false
+      },
+      input: {
+        tokens: 4e5,
+        text: true,
+        image: true,
+        cpm: 1.25,
+        cpmCached: 0.125
+      },
+      output: {
+        tokens: 128e3,
+        text: true,
+        cpm: 10
+      }
+    }
+  },
+  "gpt-5.1-codex-max": {
+    name: "gpt-5.1-codex-max",
+    provider: Vendor.OpenAI,
+    description: "GPT-5.1 codex max for maximum reasoning depth on coding tasks",
+    isActive: true,
+    releaseDate: "2025-10-01",
+    knowledgeCutoff: "2024-09-30",
+    features: {
+      reasoning: true,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      batchAPI: true,
+      promptCaching: true,
+      parameters: {
+        temperature: false,
+        topP: false,
+        frequencyPenalty: false,
+        presencePenalty: false
+      },
+      input: {
+        tokens: 4e5,
+        text: true,
+        image: true,
+        cpm: 1.25,
+        cpmCached: 0.125
+      },
+      output: {
+        tokens: 128e3,
+        text: true,
+        cpm: 10
+      }
+    }
+  },
+  "gpt-5.1-codex-mini": {
+    name: "gpt-5.1-codex-mini",
+    provider: Vendor.OpenAI,
+    description: "GPT-5.1 codex mini for cost-efficient coding tasks",
+    isActive: true,
+    releaseDate: "2025-10-01",
+    knowledgeCutoff: "2024-09-30",
+    features: {
+      reasoning: true,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      batchAPI: true,
+      promptCaching: true,
+      parameters: {
+        temperature: false,
+        topP: false,
+        frequencyPenalty: false,
+        presencePenalty: false
+      },
+      input: {
+        tokens: 4e5,
+        text: true,
+        image: true,
+        cpm: 0.25,
+        cpmCached: 0.025
+      },
+      output: {
+        tokens: 128e3,
+        text: true,
+        cpm: 2
+      }
+    }
+  },
+  "gpt-5.1-chat-latest": {
+    name: "gpt-5.1-chat-latest",
+    provider: Vendor.OpenAI,
+    description: "GPT-5.1 chat model for general-purpose use",
+    isActive: true,
+    releaseDate: "2025-10-01",
+    knowledgeCutoff: "2024-09-30",
+    features: {
+      reasoning: false,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      batchAPI: true,
+      promptCaching: true,
+      input: {
+        tokens: 128e3,
+        text: true,
+        image: true,
+        cpm: 1.25,
+        cpmCached: 0.125
+      },
+      output: {
+        tokens: 16e3,
+        text: true,
+        cpm: 10
+      }
+    }
+  },
   // GPT-5 Series
   "gpt-5": {
     name: "gpt-5",
@@ -12803,7 +13174,8 @@ var MODEL_REGISTRY = {
         tokens: 4e5,
         text: true,
         image: true,
-        cpm: 1.25
+        cpm: 1.25,
+        cpmCached: 0.125
       },
       output: {
         tokens: 128e3,
@@ -12842,7 +13214,8 @@ var MODEL_REGISTRY = {
         tokens: 4e5,
         text: true,
         image: true,
-        cpm: 0.25
+        cpm: 0.25,
+        cpmCached: 0.025
       },
       output: {
         tokens: 128e3,
@@ -12881,7 +13254,8 @@ var MODEL_REGISTRY = {
         tokens: 4e5,
         text: true,
         image: true,
-        cpm: 0.05
+        cpm: 0.05,
+        cpmCached: 5e-3
       },
       output: {
         tokens: 128e3,
@@ -12890,6 +13264,40 @@ var MODEL_REGISTRY = {
       }
     }
   },
+  "gpt-5-chat-latest": {
+    name: "gpt-5-chat-latest",
+    provider: Vendor.OpenAI,
+    description: "GPT-5 chat model for general-purpose use",
+    isActive: true,
+    releaseDate: "2025-08-01",
+    knowledgeCutoff: "2024-09-30",
+    features: {
+      reasoning: false,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      batchAPI: true,
+      promptCaching: true,
+      input: {
+        tokens: 128e3,
+        text: true,
+        image: true,
+        cpm: 1.25,
+        cpmCached: 0.125
+      },
+      output: {
+        tokens: 16e3,
+        text: true,
+        cpm: 10
+      }
+    }
+  },
   // GPT-4.1 Series
   "gpt-4.1": {
     name: "gpt-4.1",
@@ -12897,7 +13305,7 @@ var MODEL_REGISTRY = {
     description: "GPT-4.1 specialized for coding with 1M token context window",
     isActive: true,
     releaseDate: "2025-04-14",
-    knowledgeCutoff: "2025-04-01",
+    knowledgeCutoff: "2024-06-01",
     features: {
       reasoning: false,
       streaming: true,
@@ -12915,7 +13323,8 @@ var MODEL_REGISTRY = {
         tokens: 1e6,
         text: true,
         image: true,
-        cpm: 2
+        cpm: 2,
+        cpmCached: 0.5
       },
       output: {
         tokens: 32768,
@@ -12930,7 +13339,7 @@ var MODEL_REGISTRY = {
     description: "Efficient GPT-4.1 model, beats GPT-4o in many benchmarks at 83% lower cost",
     isActive: true,
     releaseDate: "2025-04-14",
-    knowledgeCutoff: "2025-04-01",
+    knowledgeCutoff: "2024-06-01",
     features: {
       reasoning: false,
       streaming: true,
@@ -12948,7 +13357,8 @@ var MODEL_REGISTRY = {
         tokens: 1e6,
         text: true,
         image: true,
-        cpm: 0.4
+        cpm: 0.4,
+        cpmCached: 0.1
       },
       output: {
         tokens: 16384,
@@ -12963,7 +13373,7 @@ var MODEL_REGISTRY = {
     description: "Fastest and cheapest model with 1M context. 80.1% MMLU, ideal for classification/autocompletion",
     isActive: true,
     releaseDate: "2025-04-14",
-    knowledgeCutoff: "2025-04-01",
+    knowledgeCutoff: "2024-06-01",
     features: {
       reasoning: false,
       streaming: true,
@@ -12981,7 +13391,8 @@ var MODEL_REGISTRY = {
         tokens: 1e6,
         text: true,
         image: true,
-        cpm: 0.1
+        cpm: 0.1,
+        cpmCached: 0.025
       },
       output: {
         tokens: 16384,
@@ -12990,14 +13401,14 @@ var MODEL_REGISTRY = {
       }
     }
   },
-  // GPT-4o Series (Legacy, Audio Capable)
+  // GPT-4o Series (Legacy)
   "gpt-4o": {
     name: "gpt-4o",
     provider: Vendor.OpenAI,
-    description: "Versatile omni model with audio support. Legacy but still available",
+    description: "Versatile omni model. Legacy but still available",
     isActive: true,
     releaseDate: "2024-05-13",
-    knowledgeCutoff: "2024-04-01",
+    knowledgeCutoff: "2023-10-01",
     features: {
       reasoning: false,
       streaming: true,
@@ -13007,7 +13418,7 @@ var MODEL_REGISTRY = {
       predictedOutputs: true,
       realtime: true,
       vision: true,
-      audio: true,
+      audio: false,
       video: false,
       batchAPI: true,
       promptCaching: true,
@@ -13015,13 +13426,12 @@ var MODEL_REGISTRY = {
         tokens: 128e3,
         text: true,
         image: true,
-        audio: true,
-        cpm: 2.5
+        cpm: 2.5,
+        cpmCached: 1.25
       },
       output: {
         tokens: 16384,
         text: true,
-        audio: true,
         cpm: 10
       }
     }
@@ -13029,10 +13439,10 @@ var MODEL_REGISTRY = {
   "gpt-4o-mini": {
     name: "gpt-4o-mini",
     provider: Vendor.OpenAI,
-    description: "Fast, affordable omni model with audio support",
+    description: "Fast, affordable omni model",
     isActive: true,
     releaseDate: "2024-07-18",
-    knowledgeCutoff: "2024-04-01",
+    knowledgeCutoff: "2023-10-01",
     features: {
       reasoning: false,
       streaming: true,
@@ -13042,7 +13452,7 @@ var MODEL_REGISTRY = {
       predictedOutputs: false,
       realtime: true,
       vision: true,
-      audio: true,
+      audio: false,
       video: false,
       batchAPI: true,
       promptCaching: true,
@@ -13050,13 +13460,12 @@ var MODEL_REGISTRY = {
         tokens: 128e3,
         text: true,
         image: true,
-        audio: true,
-        cpm: 0.15
+        cpm: 0.15,
+        cpmCached: 0.075
       },
       output: {
         tokens: 16384,
         text: true,
-        audio: true,
         cpm: 0.6
       }
     }
@@ -13068,7 +13477,7 @@ var MODEL_REGISTRY = {
     description: "Fast reasoning model tailored for coding, math, and science",
     isActive: true,
     releaseDate: "2025-01-31",
-    knowledgeCutoff: "2024-10-01",
+    knowledgeCutoff: "2023-10-01",
     features: {
       reasoning: true,
       streaming: true,
@@ -13077,11 +13486,11 @@ var MODEL_REGISTRY = {
       fineTuning: false,
       predictedOutputs: false,
       realtime: false,
-      vision: true,
+      vision: false,
       audio: false,
       video: false,
       batchAPI: true,
-      promptCaching: false,
+      promptCaching: true,
       parameters: {
         temperature: false,
         topP: false,
@@ -13091,8 +13500,8 @@ var MODEL_REGISTRY = {
       input: {
         tokens: 2e5,
         text: true,
-        image: true,
-        cpm: 1.1
+        cpm: 1.1,
+        cpmCached: 0.55
       },
       output: {
         tokens: 1e5,
@@ -13107,7 +13516,7 @@ var MODEL_REGISTRY = {
     description: "Advanced reasoning model for complex problems",
     isActive: true,
     releaseDate: "2024-12-17",
-    knowledgeCutoff: "2024-10-01",
+    knowledgeCutoff: "2023-10-01",
     features: {
       reasoning: true,
       streaming: true,
@@ -13120,7 +13529,7 @@ var MODEL_REGISTRY = {
       audio: false,
       video: false,
       batchAPI: true,
-      promptCaching: false,
+      promptCaching: true,
       parameters: {
         temperature: false,
         topP: false,
@@ -13131,7 +13540,8 @@ var MODEL_REGISTRY = {
         tokens: 2e5,
         text: true,
         image: true,
-        cpm: 15
+        cpm: 15,
+        cpmCached: 7.5
       },
       output: {
         tokens: 1e5,
@@ -13141,13 +13551,88 @@ var MODEL_REGISTRY = {
     }
   },
   // ============================================================================
-  // Anthropic Models (Verified from platform.claude.com)
+  // Anthropic Models (Verified from platform.claude.com - March 2026)
   // ============================================================================
-  // Claude 4.5 Series (Current)
+  // Claude 4.6 Series (Current)
+  "claude-opus-4-6": {
+    name: "claude-opus-4-6",
+    provider: Vendor.Anthropic,
+    description: "The most intelligent model for building agents and coding. 128K output, adaptive thinking",
+    isActive: true,
+    preferred: true,
+    releaseDate: "2026-02-01",
+    knowledgeCutoff: "2025-05-01",
+    features: {
+      reasoning: false,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      extendedThinking: true,
+      batchAPI: true,
+      promptCaching: true,
+      input: {
+        tokens: 2e5,
+        // 1M with beta header
+        text: true,
+        image: true,
+        cpm: 5,
+        cpmCached: 0.5
+      },
+      output: {
+        tokens: 128e3,
+        text: true,
+        cpm: 25
+      }
+    }
+  },
+  "claude-sonnet-4-6": {
+    name: "claude-sonnet-4-6",
+    provider: Vendor.Anthropic,
+    description: "Best combination of speed and intelligence. Adaptive thinking, 1M context beta",
+    isActive: true,
+    preferred: true,
+    releaseDate: "2026-02-01",
+    knowledgeCutoff: "2025-08-01",
+    features: {
+      reasoning: false,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      extendedThinking: true,
+      batchAPI: true,
+      promptCaching: true,
+      input: {
+        tokens: 2e5,
+        // 1M with beta header
+        text: true,
+        image: true,
+        cpm: 3,
+        cpmCached: 0.3
+      },
+      output: {
+        tokens: 64e3,
+        text: true,
+        cpm: 15
+      }
+    }
+  },
+  // Claude 4.5 Series
   "claude-opus-4-5-20251101": {
     name: "claude-opus-4-5-20251101",
     provider: Vendor.Anthropic,
-    description: "Premium model combining maximum intelligence with practical performance",
+    description: "Legacy Opus 4.5. Premium model combining maximum intelligence with practical performance",
     isActive: true,
     releaseDate: "2025-11-01",
     knowledgeCutoff: "2025-05-01",
@@ -13182,7 +13667,7 @@ var MODEL_REGISTRY = {
   "claude-sonnet-4-5-20250929": {
     name: "claude-sonnet-4-5-20250929",
     provider: Vendor.Anthropic,
-    description: "Smart model for complex agents and coding. Best balance of intelligence, speed, cost",
+    description: "Legacy Sonnet 4.5. Smart model for complex agents and coding",
     isActive: true,
     releaseDate: "2025-09-29",
     knowledgeCutoff: "2025-01-01",
@@ -13202,6 +13687,7 @@ var MODEL_REGISTRY = {
       promptCaching: true,
       input: {
         tokens: 2e5,
+        // 1M with beta header
         text: true,
         image: true,
         cpm: 3,
@@ -13285,10 +13771,45 @@ var MODEL_REGISTRY = {
       }
     }
   },
+  "claude-opus-4-20250514": {
+    name: "claude-opus-4-20250514",
+    provider: Vendor.Anthropic,
+    description: "Legacy Opus 4. Agentic tasks and reasoning",
+    isActive: true,
+    releaseDate: "2025-05-14",
+    knowledgeCutoff: "2025-01-01",
+    features: {
+      reasoning: false,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      extendedThinking: true,
+      batchAPI: true,
+      promptCaching: true,
+      input: {
+        tokens: 2e5,
+        text: true,
+        image: true,
+        cpm: 15,
+        cpmCached: 1.5
+      },
+      output: {
+        tokens: 32e3,
+        text: true,
+        cpm: 75
+      }
+    }
+  },
   "claude-sonnet-4-20250514": {
     name: "claude-sonnet-4-20250514",
     provider: Vendor.Anthropic,
-    description: "Legacy Sonnet 4. Default for most users, supports 1M context beta",
+    description: "Legacy Sonnet 4. Supports 1M context beta",
     isActive: true,
     releaseDate: "2025-05-14",
     knowledgeCutoff: "2025-01-01",
@@ -13324,7 +13845,7 @@ var MODEL_REGISTRY = {
   "claude-3-7-sonnet-20250219": {
     name: "claude-3-7-sonnet-20250219",
     provider: Vendor.Anthropic,
-    description: "Claude 3.7 Sonnet with extended thinking, supports 128K output beta",
+    description: "Deprecated. Claude 3.7 Sonnet with extended thinking",
     isActive: true,
     releaseDate: "2025-02-19",
     knowledgeCutoff: "2024-10-01",
@@ -13351,17 +13872,16 @@ var MODEL_REGISTRY = {
       },
       output: {
         tokens: 64e3,
-        // 128K with beta header
         text: true,
         cpm: 15
       }
     }
   },
-  // Claude 3.x Legacy
+  // Claude 3.x Legacy (Deprecated - retiring April 19, 2026)
   "claude-3-haiku-20240307": {
     name: "claude-3-haiku-20240307",
     provider: Vendor.Anthropic,
-    description: "Fast legacy model. Recommend migrating to Haiku 4.5",
+    description: "Deprecated. Retiring April 19, 2026. Migrate to Haiku 4.5",
     isActive: true,
     releaseDate: "2024-03-07",
     knowledgeCutoff: "2023-08-01",
@@ -13394,16 +13914,124 @@ var MODEL_REGISTRY = {
     }
   },
   // ============================================================================
-  // Google Models (Verified from ai.google.dev)
+  // Google Models (Verified from ai.google.dev - March 2026)
   // ============================================================================
+  // Gemini 3.1 Series (Preview)
+  "gemini-3.1-pro-preview": {
+    name: "gemini-3.1-pro-preview",
+    provider: Vendor.Google,
+    description: "Advanced intelligence with powerful agentic and coding capabilities. Replaces gemini-3-pro-preview",
+    isActive: true,
+    preferred: true,
+    releaseDate: "2026-02-01",
+    knowledgeCutoff: "2025-01-01",
+    features: {
+      reasoning: true,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: true,
+      video: true,
+      batchAPI: true,
+      promptCaching: true,
+      input: {
+        tokens: 1048576,
+        text: true,
+        image: true,
+        audio: true,
+        video: true,
+        cpm: 2,
+        cpmCached: 0.2
+      },
+      output: {
+        tokens: 65536,
+        text: true,
+        cpm: 12
+      }
+    }
+  },
+  "gemini-3.1-flash-lite-preview": {
+    name: "gemini-3.1-flash-lite-preview",
+    provider: Vendor.Google,
+    description: "High performance, budget-friendly for high-volume agentic tasks and data extraction",
+    isActive: true,
+    releaseDate: "2026-03-01",
+    knowledgeCutoff: "2025-01-01",
+    features: {
+      reasoning: true,
+      streaming: true,
+      structuredOutput: true,
+      functionCalling: true,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: true,
+      video: true,
+      batchAPI: true,
+      promptCaching: true,
+      input: {
+        tokens: 1048576,
+        text: true,
+        image: true,
+        audio: true,
+        video: true,
+        cpm: 0.25
+      },
+      output: {
+        tokens: 65536,
+        text: true,
+        cpm: 1.5
+      }
+    }
+  },
+  "gemini-3.1-flash-image-preview": {
+    name: "gemini-3.1-flash-image-preview",
+    provider: Vendor.Google,
+    description: "High-efficiency image generation with up to 4K output, search grounding support",
+    isActive: true,
+    releaseDate: "2026-02-01",
+    knowledgeCutoff: "2025-01-01",
+    features: {
+      reasoning: true,
+      streaming: true,
+      structuredOutput: false,
+      functionCalling: false,
+      fineTuning: false,
+      predictedOutputs: false,
+      realtime: false,
+      vision: true,
+      audio: false,
+      video: false,
+      batchAPI: true,
+      promptCaching: false,
+      input: {
+        tokens: 131072,
+        text: true,
+        image: true,
+        cpm: 0.25
+      },
+      output: {
+        tokens: 32768,
+        text: true,
+        image: true,
+        cpm: 1.5
+      }
+    }
+  },
   // Gemini 3 Series (Preview)
   "gemini-3-flash-preview": {
     name: "gemini-3-flash-preview",
     provider: Vendor.Google,
-    description: "Pro-grade reasoning with Flash-level latency and efficiency",
+    description: "Most powerful agentic and coding model with frontier-class reasoning",
     isActive: true,
-    releaseDate: "2025-11-18",
-    knowledgeCutoff: "2025-08-01",
+    preferred: true,
+    releaseDate: "2025-12-01",
+    knowledgeCutoff: "2025-01-01",
     features: {
       reasoning: true,
       streaming: true,
@@ -13418,27 +14046,28 @@ var MODEL_REGISTRY = {
       batchAPI: true,
       promptCaching: true,
       input: {
-        tokens: 1e6,
+        tokens: 1048576,
         text: true,
         image: true,
         audio: true,
         video: true,
-        cpm: 0.15
+        cpm: 0.5,
+        cpmCached: 0.05
       },
       output: {
         tokens: 65536,
         text: true,
-        cpm: 0.6
+        cpm: 3
       }
     }
   },
   "gemini-3-pro-preview": {
     name: "gemini-3-pro-preview",
     provider: Vendor.Google,
-    description: "Most advanced reasoning Gemini model for complex tasks",
+    description: "Deprecated. Shutting down March 9, 2026. Migrate to gemini-3.1-pro-preview",
     isActive: true,
     releaseDate: "2025-11-18",
-    knowledgeCutoff: "2025-08-01",
+    knowledgeCutoff: "2025-01-01",
     features: {
       reasoning: true,
       streaming: true,
@@ -13453,7 +14082,7 @@ var MODEL_REGISTRY = {
       batchAPI: true,
       promptCaching: true,
       input: {
-        tokens: 1e6,
+        tokens: 1048576,
         text: true,
         image: true,
         audio: true,
@@ -13470,14 +14099,14 @@ var MODEL_REGISTRY = {
   "gemini-3-pro-image-preview": {
     name: "gemini-3-pro-image-preview",
     provider: Vendor.Google,
-    description: "Highest quality image generation model",
+    description: "Professional-grade image generation and editing with reasoning",
     isActive: true,
     releaseDate: "2025-11-18",
-    knowledgeCutoff: "2025-08-01",
+    knowledgeCutoff: "2025-01-01",
     features: {
       reasoning: true,
       streaming: true,
-      structuredOutput: false,
+      structuredOutput: true,
       functionCalling: false,
       fineTuning: false,
       predictedOutputs: false,
@@ -13486,15 +14115,15 @@ var MODEL_REGISTRY = {
       audio: false,
       video: false,
       batchAPI: true,
-      promptCaching: true,
+      promptCaching: false,
       input: {
-        tokens: 1e6,
+        tokens: 65536,
         text: true,
         image: true,
         cpm: 1.25
       },
       output: {
-        tokens: 65536,
+        tokens: 32768,
         text: true,
         image: true,
         cpm: 10
@@ -13505,7 +14134,7 @@ var MODEL_REGISTRY = {
   "gemini-2.5-pro": {
     name: "gemini-2.5-pro",
     provider: Vendor.Google,
-    description: "Advanced multimodal model built for deep reasoning and agents",
+    description: "Most advanced model for complex tasks with deep reasoning and coding",
     isActive: true,
     releaseDate: "2025-03-01",
     knowledgeCutoff: "2025-01-01",
@@ -13523,12 +14152,13 @@ var MODEL_REGISTRY = {
       batchAPI: true,
       promptCaching: true,
       input: {
-        tokens: 1e6,
+        tokens: 1048576,
         text: true,
         image: true,
         audio: true,
         video: true,
-        cpm: 1.25
+        cpm: 1.25,
+        cpmCached: 0.125
       },
       output: {
         tokens: 65536,
@@ -13540,7 +14170,7 @@ var MODEL_REGISTRY = {
   "gemini-2.5-flash": {
     name: "gemini-2.5-flash",
     provider: Vendor.Google,
-    description: "Fast, cost-effective model with excellent reasoning",
+    description: "Best price-performance for low-latency, high-volume tasks with reasoning",
     isActive: true,
     releaseDate: "2025-06-17",
     knowledgeCutoff: "2025-01-01",
@@ -13558,24 +14188,25 @@ var MODEL_REGISTRY = {
       batchAPI: true,
       promptCaching: true,
       input: {
-        tokens: 1e6,
+        tokens: 1048576,
         text: true,
         image: true,
         audio: true,
         video: true,
-        cpm: 0.15
+        cpm: 0.3,
+        cpmCached: 0.03
       },
       output: {
         tokens: 65536,
         text: true,
-        cpm: 0.6
+        cpm: 2.5
       }
     }
   },
   "gemini-2.5-flash-lite": {
     name: "gemini-2.5-flash-lite",
     provider: Vendor.Google,
-    description: "Lowest latency for high-volume tasks, summarization, classification",
+    description: "Fastest and most budget-friendly multimodal model in the 2.5 family",
     isActive: true,
     releaseDate: "2025-06-17",
     knowledgeCutoff: "2025-01-01",
@@ -13593,31 +14224,31 @@ var MODEL_REGISTRY = {
       batchAPI: true,
       promptCaching: true,
       input: {
-        tokens: 1e6,
+        tokens: 1048576,
         text: true,
         image: true,
         audio: true,
         video: true,
-        cpm: 0.075
+        cpm: 0.1
       },
       output: {
         tokens: 65536,
         text: true,
-        cpm: 0.3
+        cpm: 0.4
       }
     }
   },
   "gemini-2.5-flash-image": {
     name: "gemini-2.5-flash-image",
     provider: Vendor.Google,
-    description: "Image generation and editing model",
+    description: "Fast native image generation and editing (Nano Banana)",
     isActive: true,
-    releaseDate: "2025-09-01",
-    knowledgeCutoff: "2025-01-01",
+    releaseDate: "2025-10-01",
+    knowledgeCutoff: "2025-06-01",
     features: {
-      reasoning: true,
+      reasoning: false,
       streaming: true,
-      structuredOutput: false,
+      structuredOutput: true,
       functionCalling: false,
       fineTuning: false,
       predictedOutputs: false,
@@ -13628,13 +14259,13 @@ var MODEL_REGISTRY = {
       batchAPI: true,
       promptCaching: true,
       input: {
-        tokens: 1e6,
+        tokens: 65536,
         text: true,
         image: true,
         cpm: 0.15
       },
       output: {
-        tokens: 65536,
+        tokens: 32768,
         text: true,
         image: true,
         cpm: 0.6
@@ -13642,7 +14273,7 @@ var MODEL_REGISTRY = {
     }
   },
   // ============================================================================
-  // xAI Grok Models (Verified from docs.x.ai - January 2026)
+  // xAI Grok Models (Verified from docs.x.ai - March 2026)
   // ============================================================================
   // Grok 4.1 Series (2M context, fast)
   "grok-4-1-fast-reasoning": {
@@ -13663,13 +14294,14 @@ var MODEL_REGISTRY = {
       vision: true,
       audio: false,
       video: false,
-      batchAPI: false,
-      promptCaching: false,
+      batchAPI: true,
+      promptCaching: true,
       input: {
         tokens: 2e6,
         text: true,
         image: true,
-        cpm: 0.2
+        cpm: 0.2,
+        cpmCached: 0.05
       },
       output: {
         tokens: 65536,
@@ -13696,13 +14328,14 @@ var MODEL_REGISTRY = {
       vision: true,
       audio: false,
       video: false,
-      batchAPI: false,
-      promptCaching: false,
+      batchAPI: true,
+      promptCaching: true,
       input: {
         tokens: 2e6,
         text: true,
         image: true,
-        cpm: 0.2
+        cpm: 0.2,
+        cpmCached: 0.05
       },
       output: {
         tokens: 65536,
@@ -13730,12 +14363,13 @@ var MODEL_REGISTRY = {
       vision: false,
       audio: false,
       video: false,
-      batchAPI: false,
-      promptCaching: false,
+      batchAPI: true,
+      promptCaching: true,
       input: {
         tokens: 256e3,
         text: true,
-        cpm: 0.2
+        cpm: 0.2,
+        cpmCached: 0.02
       },
       output: {
         tokens: 32768,
@@ -13748,7 +14382,7 @@ var MODEL_REGISTRY = {
   "grok-4-fast-reasoning": {
     name: "grok-4-fast-reasoning",
     provider: Vendor.Grok,
-    description: "Fast Grok 4 with reasoning capabilities, 2M context window",
+    description: "Fast Grok 4 with reasoning capabilities, 2M context window, vision support",
     isActive: true,
     releaseDate: "2025-09-01",
     knowledgeCutoff: "2024-11-01",
@@ -13760,15 +14394,17 @@ var MODEL_REGISTRY = {
       fineTuning: false,
       predictedOutputs: false,
       realtime: false,
-      vision: false,
+      vision: true,
       audio: false,
       video: false,
-      batchAPI: false,
-      promptCaching: false,
+      batchAPI: true,
+      promptCaching: true,
       input: {
         tokens: 2e6,
         text: true,
-        cpm: 0.2
+        image: true,
+        cpm: 0.2,
+        cpmCached: 0.05
       },
       output: {
         tokens: 65536,
@@ -13795,13 +14431,14 @@ var MODEL_REGISTRY = {
       vision: true,
       audio: false,
       video: false,
-      batchAPI: false,
-      promptCaching: false,
+      batchAPI: true,
+      promptCaching: true,
       input: {
         tokens: 2e6,
         text: true,
         image: true,
-        cpm: 0.2
+        cpm: 0.2,
+        cpmCached: 0.05
       },
       output: {
         tokens: 65536,
@@ -13813,12 +14450,12 @@ var MODEL_REGISTRY = {
   "grok-4-0709": {
     name: "grok-4-0709",
     provider: Vendor.Grok,
-    description: "Grok 4 flagship model (July 2025 release), 256K context, vision support",
+    description: "Grok 4 flagship model (July 2025 release), 256K context, vision support, reasoning",
     isActive: true,
     releaseDate: "2025-07-09",
     knowledgeCutoff: "2024-11-01",
     features: {
-      reasoning: false,
+      reasoning: true,
       streaming: true,
       structuredOutput: true,
       functionCalling: true,
@@ -13828,13 +14465,14 @@ var MODEL_REGISTRY = {
       vision: true,
       audio: false,
       video: false,
-      batchAPI: false,
-      promptCaching: false,
+      batchAPI: true,
+      promptCaching: true,
       input: {
         tokens: 256e3,
         text: true,
         image: true,
-        cpm: 3
+        cpm: 3,
+        cpmCached: 0.75
       },
       output: {
         tokens: 32768,
@@ -13847,12 +14485,12 @@ var MODEL_REGISTRY = {
   "grok-3-mini": {
     name: "grok-3-mini",
     provider: Vendor.Grok,
-    description: "Lightweight, cost-efficient model for simpler tasks, 131K context",
+    description: "Lightweight, cost-efficient model with reasoning, 131K context",
     isActive: true,
     releaseDate: "2025-06-01",
     knowledgeCutoff: "2024-11-01",
     features: {
-      reasoning: false,
+      reasoning: true,
       streaming: true,
       structuredOutput: true,
       functionCalling: true,
@@ -13862,12 +14500,13 @@ var MODEL_REGISTRY = {
       vision: false,
       audio: false,
       video: false,
-      batchAPI: false,
-      promptCaching: false,
+      batchAPI: true,
+      promptCaching: true,
       input: {
         tokens: 131072,
         text: true,
-        cpm: 0.3
+        cpm: 0.3,
+        cpmCached: 0.07
       },
       output: {
         tokens: 32768,
@@ -13894,12 +14533,13 @@ var MODEL_REGISTRY = {
       vision: false,
       audio: false,
       video: false,
-      batchAPI: false,
-      promptCaching: false,
+      batchAPI: true,
+      promptCaching: true,
       input: {
         tokens: 131072,
         text: true,
-        cpm: 3
+        cpm: 3,
+        cpmCached: 0.75
       },
       output: {
         tokens: 32768,
@@ -13908,11 +14548,11 @@ var MODEL_REGISTRY = {
       }
     }
   },
-  // Grok 2 Series (Vision)
+  // Grok 2 Series (Legacy - not in current docs)
   "grok-2-vision-1212": {
     name: "grok-2-vision-1212",
     provider: Vendor.Grok,
-    description: "Vision-capable model for image understanding, 32K context",
+    description: "Legacy vision model for image understanding, 32K context. Not in current xAI docs",
     isActive: true,
     releaseDate: "2024-12-12",
     knowledgeCutoff: "2024-11-01",
@@ -17058,27 +17698,6 @@ ${formatValue(entry.value)}`).join("\n\n")
 // src/core/context-nextgen/plugins/ToolCatalogPluginNextGen.ts
 init_Logger();
 var DEFAULT_MAX_LOADED = 10;
-var TOOL_CATALOG_INSTRUCTIONS = `## Tool Catalog
-
-You have access to a dynamic tool catalog. Not all tools are loaded at once \u2014 use these metatools to discover and load what you need:
-
-**tool_catalog_search** \u2014 Browse available tool categories and search for specific tools.
-  - No params \u2192 list all available categories with descriptions
-  - \`category\` \u2192 list tools in that category
-  - \`query\` \u2192 keyword search across categories and tools
-
-**tool_catalog_load** \u2014 Load a category's tools so you can use them.
-  - Tools become available immediately after loading.
-  - If you need tools from a category, load it first.
-
-**tool_catalog_unload** \u2014 Unload a category to free token budget.
-  - Unloaded tools are no longer sent to you.
-  - Use when you're done with a category.
-
-**Best practices:**
-- Search first to find the right category before loading.
-- Unload categories you no longer need to keep context lean.
-- Categories marked [LOADED] are already available.`;
 var catalogSearchDefinition = {
   type: "function",
   function: {
@@ -17137,6 +17756,8 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
   name = "tool_catalog";
   /** category name → array of tool names that were loaded */
   _loadedCategories = /* @__PURE__ */ new Map();
+  /** Categories that cannot be unloaded */
+  _pinnedCategories = /* @__PURE__ */ new Set();
   /** Reference to the ToolManager for registering/disabling tools */
   _toolManager = null;
   /** Cached connector categories — discovered once in setToolManager() */
@@ -17152,12 +17773,17 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
       maxLoadedCategories: DEFAULT_MAX_LOADED,
       ...config
     };
+    if (this._config.pinned?.length) {
+      for (const cat of this._config.pinned) {
+        this._pinnedCategories.add(cat);
+      }
+    }
   }
   // ========================================================================
   // Plugin Interface
   // ========================================================================
   getInstructions() {
-    return TOOL_CATALOG_INSTRUCTIONS;
+    return this.buildInstructions();
   }
   async getContent() {
     const categories = this.getAllowedCategories();
@@ -17168,15 +17794,15 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
     if (loaded.length > 0) {
       lines.push(`**Loaded:** ${loaded.join(", ")}`);
     }
-    lines.push(`**Available categories:** ${categories.length}`);
+    lines.push(`**Available categories:** ${categories.length + this.getConnectorCategories().length}`);
     for (const cat of categories) {
       const tools = ToolCatalogRegistry.getToolsInCategory(cat.name);
-      const marker = this._loadedCategories.has(cat.name) ? " [LOADED]" : "";
-      lines.push(`- **${cat.displayName}** (${tools.length} tools)${marker}: ${cat.description}`);
+      const markers = this.getCategoryMarkers(cat.name);
+      lines.push(`- **${cat.displayName}** (${tools.length} tools)${markers}: ${cat.description}`);
     }
     for (const cc of this.getConnectorCategories()) {
-      const marker = this._loadedCategories.has(cc.name) ? " [LOADED]" : "";
-      lines.push(`- **${cc.displayName}** (${cc.toolCount} tools)${marker}: ${cc.description}`);
+      const markers = this.getCategoryMarkers(cc.name);
+      lines.push(`- **${cc.displayName}** (${cc.toolCount} tools)${markers}: ${cc.description}`);
     }
     const content = lines.join("\n");
     this.updateTokenCache(this.estimator.estimateTokens(content));
@@ -17187,7 +17813,8 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
       loadedCategories: Array.from(this._loadedCategories.entries()).map(([name, tools]) => ({
         category: name,
         toolCount: tools.length,
-        tools
+        tools,
+        pinned: this._pinnedCategories.has(name)
       }))
     };
   }
@@ -17214,11 +17841,14 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
     return [searchTool, loadTool, unloadTool];
   }
   isCompactable() {
-    return this._loadedCategories.size > 0;
+    for (const category of this._loadedCategories.keys()) {
+      if (!this._pinnedCategories.has(category)) return true;
+    }
+    return false;
   }
   async compact(targetTokensToFree) {
     if (!this._toolManager || this._loadedCategories.size === 0) return 0;
-    const categoriesByLastUsed = this.getCategoriesSortedByLastUsed();
+    const categoriesByLastUsed = this.getCategoriesSortedByLastUsed().filter((cat) => !this._pinnedCategories.has(cat));
     let freed = 0;
     for (const category of categoriesByLastUsed) {
       if (freed >= targetTokensToFree) break;
@@ -17259,6 +17889,7 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
   }
   destroy() {
     this._loadedCategories.clear();
+    this._pinnedCategories.clear();
     this._toolManager = null;
     this._connectorCategories = null;
     this._destroyed = true;
@@ -17272,11 +17903,20 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
   setToolManager(tm) {
     this._toolManager = tm;
     this._connectorCategories = ToolCatalogRegistry.discoverConnectorCategories({
-      scope: this._config.categoryScope,
       identities: this._config.identities
     });
+    for (const category of this._pinnedCategories) {
+      const result = this.executeLoad(category);
+      if (result.error) {
+        exports.logger.warn(
+          { category, error: result.error },
+          `[ToolCatalogPlugin] Failed to load pinned category '${category}'`
+        );
+      }
+    }
     if (this._config.autoLoadCategories?.length) {
       for (const category of this._config.autoLoadCategories) {
+        if (this._pinnedCategories.has(category)) continue;
         const result = this.executeLoad(category);
         if (result.error) {
           exports.logger.warn(
@@ -17291,6 +17931,10 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
   get loadedCategories() {
     return Array.from(this._loadedCategories.keys());
   }
+  /** Get set of pinned category names */
+  get pinnedCategories() {
+    return this._pinnedCategories;
+  }
   // ========================================================================
   // Metatool Implementations
   // ========================================================================
@@ -17311,6 +17955,7 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
       return {
         category,
         loaded,
+        pinned: this._pinnedCategories.has(category),
         tools: tools.map((t) => ({
           name: t.name,
           displayName: t.displayName,
@@ -17332,7 +17977,8 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
         displayName: cat.displayName,
         description: cat.description,
         toolCount: tools.length,
-        loaded: this._loadedCategories.has(cat.name)
+        loaded: this._loadedCategories.has(cat.name),
+        pinned: this._pinnedCategories.has(cat.name)
       });
     }
     for (const cc of connectorCats) {
@@ -17341,7 +17987,8 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
         displayName: cc.displayName,
         description: cc.description,
         toolCount: cc.toolCount,
-        loaded: this._loadedCategories.has(cc.name)
+        loaded: this._loadedCategories.has(cc.name),
+        pinned: this._pinnedCategories.has(cc.name)
       });
     }
     return { categories: result };
@@ -17351,20 +17998,28 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
     if (!this._toolManager) {
       return { error: "ToolManager not connected. Plugin not properly initialized." };
     }
-    if (!ToolCatalogRegistry.isCategoryAllowed(category, this._config.categoryScope)) {
-      return { error: `Category '${category}' is not available for this agent.` };
+    const isConnector = ToolCatalogRegistry.parseConnectorCategory(category) !== null;
+    if (isConnector) {
+      const allowed = this.getConnectorCategories().some((cc) => cc.name === category);
+      if (!allowed) {
+        return { error: `Category '${category}' is not available for this agent.` };
+      }
+    } else {
+      if (!ToolCatalogRegistry.isCategoryAllowed(category, this._config.categoryScope)) {
+        return { error: `Category '${category}' is not available for this agent.` };
+      }
     }
     if (this._loadedCategories.has(category)) {
       const toolNames2 = this._loadedCategories.get(category);
       return { loaded: toolNames2.length, tools: toolNames2, alreadyLoaded: true };
     }
-    if (this._loadedCategories.size >= this._config.maxLoadedCategories) {
+    const nonPinnedLoaded = this._loadedCategories.size - this._pinnedCategories.size;
+    if (!this._pinnedCategories.has(category) && nonPinnedLoaded >= this._config.maxLoadedCategories) {
       return {
         error: `Maximum loaded categories (${this._config.maxLoadedCategories}) reached. Unload a category first.`,
         loaded: Array.from(this._loadedCategories.keys())
       };
     }
-    const isConnector = ToolCatalogRegistry.parseConnectorCategory(category) !== null;
     let tools;
     if (isConnector) {
       tools = ToolCatalogRegistry.resolveConnectorCategoryTools(category);
@@ -17401,6 +18056,9 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
     if (!this._toolManager) {
       return { error: "ToolManager not connected." };
     }
+    if (this._pinnedCategories.has(category)) {
+      return { error: `Category '${category}' is pinned and cannot be unloaded.` };
+    }
     const toolNames = this._loadedCategories.get(category);
     if (!toolNames) {
       return { unloaded: 0, message: `Category '${category}' is not loaded.` };
@@ -17426,6 +18084,61 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
   getConnectorCategories() {
     return this._connectorCategories ?? [];
   }
+  /**
+   * Build status markers for a category (e.g., " [PINNED]", " [LOADED]", " [PINNED] [LOADED]")
+   */
+  getCategoryMarkers(name) {
+    const parts = [];
+    if (this._pinnedCategories.has(name)) parts.push("[PINNED]");
+    if (this._loadedCategories.has(name)) parts.push("[LOADED]");
+    return parts.length > 0 ? " " + parts.join(" ") : "";
+  }
+  /**
+   * Build dynamic instructions that include the list of available categories.
+   */
+  buildInstructions() {
+    const lines = [];
+    lines.push("## Tool Catalog");
+    lines.push("");
+    lines.push("Your core tools (memory, context, instructions, etc.) are always available.");
+    lines.push("Additional tool categories can be loaded on demand from the catalog below.");
+    lines.push("");
+    lines.push("**tool_catalog_search** \u2014 Browse available tool categories and search for specific tools.");
+    lines.push("  - No params \u2192 list all available categories with descriptions");
+    lines.push("  - `category` \u2192 list tools in that category");
+    lines.push("  - `query` \u2192 keyword search across categories and tools");
+    lines.push("");
+    lines.push("**tool_catalog_load** \u2014 Load a category's tools so you can use them.");
+    lines.push("  - Tools become available immediately after loading.");
+    lines.push("  - If you need tools from a category, load it first.");
+    lines.push("");
+    lines.push("**tool_catalog_unload** \u2014 Unload a category to free token budget.");
+    lines.push("  - Unloaded tools are no longer sent to you.");
+    lines.push("  - Use when you're done with a category.");
+    lines.push("  - Pinned categories cannot be unloaded.");
+    lines.push("");
+    const builtIn = this.getAllowedCategories();
+    const connectors = this.getConnectorCategories();
+    if (builtIn.length > 0 || connectors.length > 0) {
+      lines.push("**Available categories:**");
+      for (const cat of builtIn) {
+        const tools = ToolCatalogRegistry.getToolsInCategory(cat.name);
+        const pinned = this._pinnedCategories.has(cat.name) ? " [PINNED]" : "";
+        lines.push(`- ${cat.name} (${tools.length} tools)${pinned}: ${cat.description}`);
+      }
+      for (const cc of connectors) {
+        const pinned = this._pinnedCategories.has(cc.name) ? " [PINNED]" : "";
+        lines.push(`- ${cc.name} (${cc.toolCount} tools)${pinned}: ${cc.description}`);
+      }
+      lines.push("");
+    }
+    lines.push("**Best practices:**");
+    lines.push("- Search first to find the right category before loading.");
+    lines.push("- Unload categories you no longer need to keep context lean.");
+    lines.push("- Categories marked [LOADED] are already available.");
+    lines.push("- Categories marked [PINNED] are always available and cannot be unloaded.");
+    return lines.join("\n");
+  }
   keywordSearch(query) {
     const lq = query.toLowerCase();
     const results = [];
@@ -17463,12 +18176,17 @@ var ToolCatalogPluginNextGen = class extends BasePluginNextGen {
     return { query, results, totalMatches: results.length };
   }
   searchConnectorCategory(category) {
+    const allowed = this.getConnectorCategories().some((cc) => cc.name === category);
+    if (!allowed) {
+      return { error: `Category '${category}' is not available for this agent.` };
+    }
     const connectorName = ToolCatalogRegistry.parseConnectorCategory(category);
     const tools = ToolCatalogRegistry.resolveConnectorCategoryTools(category);
     const loaded = this._loadedCategories.has(category);
     return {
       category,
       loaded,
+      pinned: this._pinnedCategories.has(category),
       connectorName,
       tools: tools.map((t) => ({
         name: t.name,
@@ -20217,6 +20935,9 @@ var StreamEventType = /* @__PURE__ */ ((StreamEventType2) => {
   StreamEventType2["REASONING_DONE"] = "response.reasoning.done";
   StreamEventType2["RESPONSE_COMPLETE"] = "response.complete";
   StreamEventType2["ERROR"] = "response.error";
+  StreamEventType2["AUDIO_CHUNK_READY"] = "response.audio_chunk.ready";
+  StreamEventType2["AUDIO_CHUNK_ERROR"] = "response.audio_chunk.error";
+  StreamEventType2["AUDIO_STREAM_COMPLETE"] = "response.audio_stream.complete";
   return StreamEventType2;
 })(StreamEventType || {});
 function isStreamEvent(event, type) {
@@ -20246,6 +20967,15 @@ function isResponseComplete(event) {
 function isErrorEvent(event) {
   return event.type === "response.error" /* ERROR */;
 }
+function isAudioChunkReady(event) {
+  return event.type === "response.audio_chunk.ready" /* AUDIO_CHUNK_READY */;
+}
+function isAudioChunkError(event) {
+  return event.type === "response.audio_chunk.error" /* AUDIO_CHUNK_ERROR */;
+}
+function isAudioStreamComplete(event) {
+  return event.type === "response.audio_stream.complete" /* AUDIO_STREAM_COMPLETE */;
+}
 
 // src/infrastructure/providers/openai/OpenAIResponsesStreamConverter.ts
 var OpenAIResponsesStreamConverter = class {
@@ -20555,9 +21285,17 @@ var OpenAITextProvider = class extends BaseTextProvider {
           ...options.metadata && { metadata: options.metadata }
         };
         this.applyReasoningConfig(params, options);
+        console.log(
+          `[OpenAITextProvider] generate: calling OpenAI API (model=${options.model}, tools=${params.tools?.length ?? 0})`
+        );
+        const genStartTime = Date.now();
         const response = await this.client.responses.create(params);
+        console.log(
+          `[OpenAITextProvider] generate: response received (${Date.now() - genStartTime}ms)`
+        );
         return this.converter.convertResponse(response);
       } catch (error) {
+        console.error(`[OpenAITextProvider] generate error (model=${options.model}):`, error.message || error);
         this.handleError(error, options.model);
         throw error;
       }
@@ -20597,9 +21335,27 @@ var OpenAITextProvider = class extends BaseTextProvider {
         stream: true
       };
       this.applyReasoningConfig(params, options);
+      console.log(
+        `[OpenAITextProvider] streamGenerate: calling OpenAI API (model=${options.model}, tools=${params.tools?.length ?? 0})`
+      );
+      const streamStartTime = Date.now();
       const stream = await this.client.responses.create(params);
-      yield* this.streamConverter.convertStream(stream);
+      console.log(
+        `[OpenAITextProvider] streamGenerate: OpenAI stream opened (${Date.now() - streamStartTime}ms)`
+      );
+      let chunkCount = 0;
+      for await (const event of this.streamConverter.convertStream(stream)) {
+        chunkCount++;
+        yield event;
+      }
+      console.log(
+        `[OpenAITextProvider] streamGenerate: stream complete (${chunkCount} events, ${Date.now() - streamStartTime}ms total)`
+      );
     } catch (error) {
+      console.error(
+        `[OpenAITextProvider] streamGenerate error (model=${options.model}):`,
+        error.message || error
+      );
       this.handleError(error, options.model);
       throw error;
     }
@@ -21605,12 +22361,20 @@ var AnthropicTextProvider = class extends BaseTextProvider {
     return this.executeWithCircuitBreaker(async () => {
       try {
         const anthropicRequest = this.converter.convertRequest(options);
+        console.log(
+          `[AnthropicTextProvider] generate: calling Anthropic API (model=${options.model}, messages=${anthropicRequest.messages?.length ?? 0}, tools=${anthropicRequest.tools?.length ?? 0})`
+        );
+        const genStartTime = Date.now();
         const anthropicResponse = await this.client.messages.create({
           ...anthropicRequest,
           stream: false
         });
+        console.log(
+          `[AnthropicTextProvider] generate: response received (${Date.now() - genStartTime}ms)`
+        );
         return this.converter.convertResponse(anthropicResponse);
       } catch (error) {
+        console.error(`[AnthropicTextProvider] generate error (model=${options.model}):`, error.message || error);
         this.handleError(error, options.model);
         throw error;
       }
@@ -21622,13 +22386,31 @@ var AnthropicTextProvider = class extends BaseTextProvider {
   async *streamGenerate(options) {
     try {
       const anthropicRequest = this.converter.convertRequest(options);
+      console.log(
+        `[AnthropicTextProvider] streamGenerate: calling Anthropic API (model=${options.model}, messages=${anthropicRequest.messages?.length ?? 0}, tools=${anthropicRequest.tools?.length ?? 0})`
+      );
+      const streamStartTime = Date.now();
       const stream = await this.client.messages.create({
         ...anthropicRequest,
         stream: true
       });
+      console.log(
+        `[AnthropicTextProvider] streamGenerate: Anthropic stream opened (${Date.now() - streamStartTime}ms)`
+      );
       this.streamConverter.reset();
-      yield* this.streamConverter.convertStream(stream, options.model);
+      let chunkCount = 0;
+      for await (const event of this.streamConverter.convertStream(stream, options.model)) {
+        chunkCount++;
+        yield event;
+      }
+      console.log(
+        `[AnthropicTextProvider] streamGenerate: stream complete (${chunkCount} events, ${Date.now() - streamStartTime}ms total)`
+      );
     } catch (error) {
+      console.error(
+        `[AnthropicTextProvider] streamGenerate error (model=${options.model}):`,
+        error.message || error
+      );
       this.handleError(error, options.model);
       throw error;
     } finally {
@@ -22422,6 +23204,10 @@ var GoogleTextProvider = class extends BaseTextProvider {
             // First message only
           }, null, 2));
         }
+        console.log(
+          `[GoogleTextProvider] generate: calling Google API (model=${options.model}, contents=${googleRequest.contents?.length ?? 0} messages, tools=${googleRequest.tools?.[0]?.functionDeclarations?.length ?? 0} tools)`
+        );
+        const genStartTime = Date.now();
         const result = await this.client.models.generateContent({
           model: options.model,
           contents: googleRequest.contents,
@@ -22432,6 +23218,9 @@ var GoogleTextProvider = class extends BaseTextProvider {
             ...googleRequest.generationConfig
           }
         });
+        console.log(
+          `[GoogleTextProvider] generate: response received (${Date.now() - genStartTime}ms)`
+        );
         if (process.env.DEBUG_GOOGLE) {
           console.error("[DEBUG] Google Response:", JSON.stringify({
             candidates: result.candidates?.map((c) => ({
@@ -22450,6 +23239,7 @@ var GoogleTextProvider = class extends BaseTextProvider {
         }
         return response;
       } catch (error) {
+        console.error(`[GoogleTextProvider] generate error (model=${options.model}):`, error.message || error);
         this.converter.clearMappings();
         this.handleError(error, options.model);
         throw error;
@@ -22462,6 +23252,10 @@ var GoogleTextProvider = class extends BaseTextProvider {
   async *streamGenerate(options) {
     try {
       const googleRequest = await this.converter.convertRequest(options);
+      console.log(
+        `[GoogleTextProvider] streamGenerate: calling Google API (model=${options.model}, contents=${googleRequest.contents?.length ?? 0} messages, tools=${googleRequest.tools?.[0]?.functionDeclarations?.length ?? 0} tools)`
+      );
+      const streamStartTime = Date.now();
       const stream = await this.client.models.generateContentStream({
         model: options.model,
         contents: googleRequest.contents,
@@ -22472,13 +23266,27 @@ var GoogleTextProvider = class extends BaseTextProvider {
           ...googleRequest.generationConfig
         }
       });
+      console.log(
+        `[GoogleTextProvider] streamGenerate: Google stream opened (${Date.now() - streamStartTime}ms)`
+      );
       this.streamConverter.reset();
-      yield* this.streamConverter.convertStream(stream, options.model);
+      let chunkCount = 0;
+      for await (const event of this.streamConverter.convertStream(stream, options.model)) {
+        chunkCount++;
+        yield event;
+      }
+      console.log(
+        `[GoogleTextProvider] streamGenerate: stream complete (${chunkCount} events, ${Date.now() - streamStartTime}ms total)`
+      );
       if (!this.streamConverter.hasToolCalls()) {
         this.converter.clearMappings();
         this.streamConverter.clear();
       }
     } catch (error) {
+      console.error(
+        `[GoogleTextProvider] streamGenerate error (model=${options.model}):`,
+        error.message || error
+      );
       this.converter.clearMappings();
       this.streamConverter.clear();
       this.handleError(error, options.model);
@@ -22565,6 +23373,10 @@ var VertexAITextProvider = class extends BaseTextProvider {
   async generate(options) {
     try {
       const googleRequest = await this.converter.convertRequest(options);
+      console.log(
+        `[VertexAITextProvider] generate: calling Vertex AI (model=${options.model}, contents=${googleRequest.contents?.length ?? 0} messages, tools=${googleRequest.tools?.[0]?.functionDeclarations?.length ?? 0} tools)`
+      );
+      const genStartTime = Date.now();
       const result = await this.client.models.generateContent({
         model: options.model,
         contents: googleRequest.contents,
@@ -22575,8 +23387,12 @@ var VertexAITextProvider = class extends BaseTextProvider {
           ...googleRequest.generationConfig
         }
       });
+      console.log(
+        `[VertexAITextProvider] generate: response received (${Date.now() - genStartTime}ms)`
+      );
       return this.converter.convertResponse(result);
     } catch (error) {
+      console.error(`[VertexAITextProvider] generate error (model=${options.model}):`, error.message || error);
       this.handleError(error, options.model);
       throw error;
     }
@@ -22587,6 +23403,10 @@ var VertexAITextProvider = class extends BaseTextProvider {
   async *streamGenerate(options) {
     try {
       const googleRequest = await this.converter.convertRequest(options);
+      console.log(
+        `[VertexAITextProvider] streamGenerate: calling Vertex AI (model=${options.model}, contents=${googleRequest.contents?.length ?? 0} messages, tools=${googleRequest.tools?.[0]?.functionDeclarations?.length ?? 0} tools)`
+      );
+      const streamStartTime = Date.now();
       const stream = await this.client.models.generateContentStream({
         model: options.model,
         contents: googleRequest.contents,
@@ -22597,9 +23417,23 @@ var VertexAITextProvider = class extends BaseTextProvider {
           ...googleRequest.generationConfig
         }
       });
+      console.log(
+        `[VertexAITextProvider] streamGenerate: Vertex AI stream opened (${Date.now() - streamStartTime}ms)`
+      );
       const streamConverter = new GoogleStreamConverter();
-      yield* streamConverter.convertStream(stream, options.model);
+      let chunkCount = 0;
+      for await (const event of streamConverter.convertStream(stream, options.model)) {
+        chunkCount++;
+        yield event;
+      }
+      console.log(
+        `[VertexAITextProvider] streamGenerate: stream complete (${chunkCount} events, ${Date.now() - streamStartTime}ms total)`
+      );
     } catch (error) {
+      console.error(
+        `[VertexAITextProvider] streamGenerate error (model=${options.model}):`,
+        error.message || error
+      );
       this.handleError(error, options.model);
       throw error;
     }
@@ -24540,6 +25374,20 @@ var Agent = class _Agent extends BaseAgent {
       timestamp: /* @__PURE__ */ new Date(),
       duration: totalDuration
     });
+    const hasTextOutput = response.output_text?.trim() || response.output?.some(
+      (item) => "content" in item && Array.isArray(item.content) && item.content.some((c) => c.type === "output_text" /* OUTPUT_TEXT */ && c.text?.trim())
+    );
+    if (!hasTextOutput) {
+      console.warn(
+        `[Agent] WARNING: ${methodName} completed with zero text output (executionId=${executionId}, iterations=${this.executionContext?.metrics.iterationCount ?? "?"}, tokens=${response.usage?.total_tokens ?? 0})`
+      );
+      this.emit("execution:empty_output", {
+        executionId,
+        timestamp: /* @__PURE__ */ new Date(),
+        duration: totalDuration,
+        usage: response.usage
+      });
+    }
     const duration = Date.now() - startTime;
     this._logger.info({ duration }, `Agent ${methodName} completed`);
     exports.metrics.timing(`agent.${methodName}.duration`, duration, { model: this.model, connector: this.connector.name });
@@ -24594,6 +25442,17 @@ var Agent = class _Agent extends BaseAgent {
         }
         const iterationStartTime = Date.now();
         const prepared = await this._agentContext.prepare();
+        const b1 = prepared.budget;
+        const bd1 = b1.breakdown;
+        const bp1 = [
+          `sysPrompt=${bd1.systemPrompt}`,
+          `PI=${bd1.persistentInstructions}`,
+          bd1.pluginInstructions ? `pluginInstr=${bd1.pluginInstructions}` : "",
+          ...Object.entries(bd1.pluginContents || {}).map(([k, v]) => `plugin:${k}=${v}`)
+        ].filter(Boolean).join(" ");
+        console.log(
+          `[Agent] [Context] iteration=${iteration} tokens: ${b1.totalUsed}/${b1.maxTokens} (${b1.utilizationPercent.toFixed(1)}%) tools=${b1.toolsTokens} conversation=${b1.conversationTokens} system=${b1.systemMessageTokens} input=${b1.currentInputTokens}` + (bp1 ? ` | ${bp1}` : "") + (prepared.compacted ? ` COMPACTED: ${prepared.compactionLog.join("; ")}` : "")
+        );
         const response = await this.generateWithHooks(prepared.input, iteration, executionId);
         const toolCalls = this.extractToolCalls(response.output);
         this._agentContext.addAssistantResponse(response.output);
@@ -24708,13 +25567,23 @@ var Agent = class _Agent extends BaseAgent {
    * Build placeholder response for streaming finalization
    */
   _buildPlaceholderResponse(executionId, startTime, streamState) {
+    const outputText = streamState.getAllText();
+    const output = [];
+    if (outputText && outputText.trim()) {
+      output.push({
+        type: "message",
+        role: "assistant" /* ASSISTANT */,
+        content: [{ type: "output_text" /* OUTPUT_TEXT */, text: outputText }]
+      });
+    }
     return {
       id: executionId,
       object: "response",
       created_at: Math.floor(startTime / 1e3),
       status: "completed",
       model: this.model,
-      output: [],
+      output,
+      output_text: outputText || void 0,
       usage: streamState.usage
     };
   }
@@ -24734,6 +25603,17 @@ var Agent = class _Agent extends BaseAgent {
           break;
         }
         const prepared = await this._agentContext.prepare();
+        const b2 = prepared.budget;
+        const bd2 = b2.breakdown;
+        const bp2 = [
+          `sysPrompt=${bd2.systemPrompt}`,
+          `PI=${bd2.persistentInstructions}`,
+          bd2.pluginInstructions ? `pluginInstr=${bd2.pluginInstructions}` : "",
+          ...Object.entries(bd2.pluginContents || {}).map(([k, v]) => `plugin:${k}=${v}`)
+        ].filter(Boolean).join(" ");
+        console.log(
+          `[Agent] [Context] iteration=${iteration} tokens: ${b2.totalUsed}/${b2.maxTokens} (${b2.utilizationPercent.toFixed(1)}%) tools=${b2.toolsTokens} conversation=${b2.conversationTokens} system=${b2.systemMessageTokens} input=${b2.currentInputTokens}` + (bp2 ? ` | ${bp2}` : "") + (prepared.compacted ? ` COMPACTED: ${prepared.compactionLog.join("; ")}` : "")
+        );
         const iterationStreamState = new StreamState(executionId, this.model);
         const toolCallsMap = /* @__PURE__ */ new Map();
         yield* this.streamGenerateWithHooks(
@@ -33403,6 +34283,56 @@ var OpenAITTSProvider = class extends BaseMediaProvider {
       { model: options.model, voice: options.voice }
     );
   }
+  /**
+   * Check if streaming is supported for the given format
+   */
+  supportsStreaming(format) {
+    if (!format) return true;
+    return ["pcm", "wav", "mp3", "opus", "aac", "flac"].includes(format);
+  }
+  /**
+   * Stream TTS audio chunks as they arrive from the API
+   */
+  async *synthesizeStream(options) {
+    const format = this.mapFormat(options.format);
+    const requestParams = {
+      model: options.model,
+      input: options.input,
+      voice: options.voice,
+      response_format: format,
+      speed: options.speed
+    };
+    if (options.vendorOptions?.instructions) {
+      requestParams.instructions = options.vendorOptions.instructions;
+    }
+    this.logOperationStart("tts.synthesizeStream", {
+      model: options.model,
+      voice: options.voice,
+      inputLength: options.input.length,
+      format
+    });
+    try {
+      const response = await this.client.audio.speech.create(requestParams);
+      const body = response.body;
+      if (!body) {
+        throw new Error("No response body from OpenAI TTS API");
+      }
+      let totalBytes = 0;
+      for await (const chunk of body) {
+        const buf = Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk);
+        totalBytes += buf.length;
+        yield { audio: buf, isFinal: false };
+      }
+      yield { audio: Buffer.alloc(0), isFinal: true };
+      this.logOperationComplete("tts.synthesizeStream", {
+        model: options.model,
+        totalBytes
+      });
+    } catch (error) {
+      this.handleError(error);
+      throw error;
+    }
+  }
   /**
    * List available voices (returns static list for OpenAI)
    */
@@ -34004,13 +34934,13 @@ var TTS_MODEL_REGISTRY = {
     name: "gemini-2.5-flash-preview-tts",
     displayName: "Gemini 2.5 Flash TTS",
     provider: Vendor.Google,
-    description: "Google Gemini 2.5 Flash TTS - optimized for low latency",
+    description: "Google Gemini 2.5 Flash TTS - optimized for low latency, 30 voices, 70+ languages",
     isActive: true,
     releaseDate: "2025-01-01",
     sources: {
       documentation: "https://ai.google.dev/gemini-api/docs/speech-generation",
       pricing: "https://ai.google.dev/pricing",
-      lastVerified: "2026-01-25"
+      lastVerified: "2026-03-04"
     },
     capabilities: {
       voices: GEMINI_VOICES,
@@ -34029,20 +34959,27 @@ var TTS_MODEL_REGISTRY = {
         wordTimestamps: false
       },
       limits: { maxInputLength: 32e3 }
-      // 32k tokens
+      // 32k token context window
+    },
+    pricing: {
+      perMInputTokens: 0.5,
+      // $0.50 per 1M input tokens
+      perMOutputTokens: 10,
+      // $10.00 per 1M output tokens
+      currency: "USD"
     }
   },
   "gemini-2.5-pro-preview-tts": {
     name: "gemini-2.5-pro-preview-tts",
     displayName: "Gemini 2.5 Pro TTS",
     provider: Vendor.Google,
-    description: "Google Gemini 2.5 Pro TTS - optimized for quality",
+    description: "Google Gemini 2.5 Pro TTS - optimized for quality, 30 voices, 70+ languages",
     isActive: true,
     releaseDate: "2025-01-01",
     sources: {
       documentation: "https://ai.google.dev/gemini-api/docs/speech-generation",
       pricing: "https://ai.google.dev/pricing",
-      lastVerified: "2026-01-25"
+      lastVerified: "2026-03-04"
     },
     capabilities: {
       voices: GEMINI_VOICES,
@@ -34061,7 +34998,14 @@ var TTS_MODEL_REGISTRY = {
         wordTimestamps: false
       },
       limits: { maxInputLength: 32e3 }
-      // 32k tokens
+      // 32k token context window
+    },
+    pricing: {
+      perMInputTokens: 1,
+      // $1.00 per 1M input tokens
+      perMOutputTokens: 20,
+      // $20.00 per 1M output tokens
+      currency: "USD"
     }
   }
 };
@@ -34074,10 +35018,18 @@ function getTTSModelsWithFeature(feature) {
     (model) => model.isActive && model.capabilities.features[feature]
   );
 }
-function calculateTTSCost(modelName, characterCount) {
+function calculateTTSCost(modelName, characterCount, options) {
   const model = getTTSModelInfo(modelName);
   if (!model?.pricing) return null;
-  return characterCount / 1e3 * model.pricing.per1kCharacters;
+  if (model.pricing.per1kCharacters) {
+    return characterCount / 1e3 * model.pricing.per1kCharacters;
+  }
+  if (model.pricing.perMInputTokens && options?.inputTokens != null) {
+    const inputCost = options.inputTokens / 1e6 * model.pricing.perMInputTokens;
+    const outputCost = options.outputTokens ? options.outputTokens / 1e6 * (model.pricing.perMOutputTokens ?? 0) : 0;
+    return inputCost + outputCost;
+  }
+  return null;
 }
 var TextToSpeech = class _TextToSpeech {
   provider;
@@ -34123,6 +35075,35 @@ var TextToSpeech = class _TextToSpeech {
     const response = await this.synthesize(text, options);
     await fs17__namespace.writeFile(filePath, response.audio);
   }
+  // ======================== Streaming Methods ========================
+  /**
+   * Check if the underlying provider supports streaming TTS
+   */
+  supportsStreaming(format) {
+    const provider = this.provider;
+    return typeof provider.supportsStreaming === "function" && provider.supportsStreaming(format);
+  }
+  /**
+   * Stream TTS audio chunks as they arrive from the API.
+   * Falls back to buffered synthesis yielding a single chunk if provider doesn't support streaming.
+   */
+  async *synthesizeStream(text, options) {
+    const fullOptions = {
+      model: this.config.model ?? this.getDefaultModel(),
+      input: text,
+      voice: options?.voice ?? this.config.voice ?? this.getDefaultVoice(),
+      format: options?.format ?? this.config.format,
+      speed: options?.speed ?? this.config.speed,
+      vendorOptions: options?.vendorOptions
+    };
+    const provider = this.provider;
+    if (typeof provider.synthesizeStream === "function" && provider.supportsStreaming?.(fullOptions.format)) {
+      yield* provider.synthesizeStream(fullOptions);
+    } else {
+      const response = await this.provider.synthesize(fullOptions);
+      yield { audio: response.audio, isFinal: true };
+    }
+  }
   // ======================== Introspection Methods ========================
   /**
    * Get model information for current or specified model
@@ -35523,7 +36504,13 @@ var IMAGE_MODELS = {
     /** Imagen 4.0 Ultra: Highest quality */
     IMAGEN_4_ULTRA: "imagen-4.0-ultra-generate-001",
     /** Imagen 4.0 Fast: Optimized for speed */
-    IMAGEN_4_FAST: "imagen-4.0-fast-generate-001"
+    IMAGEN_4_FAST: "imagen-4.0-fast-generate-001",
+    /** Nano Banana 2: Gemini 3.1 Flash native image gen with 4K support */
+    GEMINI_3_1_FLASH_IMAGE: "gemini-3.1-flash-image-preview",
+    /** Nano Banana Pro: Gemini 3 Pro professional design engine with reasoning */
+    GEMINI_3_PRO_IMAGE: "gemini-3-pro-image-preview",
+    /** Nano Banana: Gemini 2.5 Flash native image gen/editing */
+    GEMINI_2_5_FLASH_IMAGE: "gemini-2.5-flash-image"
   },
   [Vendor.Grok]: {
     /** Grok Imagine Image: xAI image generation with editing support */
@@ -35707,7 +36694,7 @@ var IMAGE_MODEL_REGISTRY = {
     sources: {
       documentation: "https://ai.google.dev/gemini-api/docs/imagen",
       pricing: "https://ai.google.dev/pricing",
-      lastVerified: "2026-01-25"
+      lastVerified: "2026-03-04"
     },
     capabilities: {
       sizes: ["1024x1024"],
@@ -35818,7 +36805,7 @@ var IMAGE_MODEL_REGISTRY = {
     sources: {
       documentation: "https://ai.google.dev/gemini-api/docs/imagen",
       pricing: "https://ai.google.dev/pricing",
-      lastVerified: "2026-01-25"
+      lastVerified: "2026-03-04"
     },
     capabilities: {
       sizes: ["1024x1024"],
@@ -35915,7 +36902,8 @@ var IMAGE_MODEL_REGISTRY = {
       }
     },
     pricing: {
-      perImage: 0.08,
+      perImage: 0.06,
+      // Updated per official pricing page (was $0.08)
       currency: "USD"
     }
   },
@@ -35929,7 +36917,7 @@ var IMAGE_MODEL_REGISTRY = {
     sources: {
       documentation: "https://ai.google.dev/gemini-api/docs/imagen",
       pricing: "https://ai.google.dev/pricing",
-      lastVerified: "2026-01-25"
+      lastVerified: "2026-03-04"
     },
     capabilities: {
       sizes: ["1024x1024"],
@@ -36030,6 +37018,141 @@ var IMAGE_MODEL_REGISTRY = {
       currency: "USD"
     }
   },
+  // ======================== Google Nano Banana (Gemini Native Image) ========================
+  "gemini-3.1-flash-image-preview": {
+    name: "gemini-3.1-flash-image-preview",
+    displayName: "Nano Banana 2 (Gemini 3.1 Flash Image)",
+    provider: Vendor.Google,
+    description: "High-efficiency native image generation and editing with 4K support and thinking capabilities",
+    isActive: true,
+    releaseDate: "2026-02-01",
+    sources: {
+      documentation: "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-image-preview",
+      pricing: "https://ai.google.dev/pricing",
+      lastVerified: "2026-03-04"
+    },
+    capabilities: {
+      sizes: ["512x512", "1024x1024", "1536x1536", "auto"],
+      aspectRatios: ["1:1", "1:4", "4:1", "1:8", "8:1"],
+      maxImagesPerRequest: 4,
+      outputFormats: ["png", "jpeg"],
+      features: {
+        generation: true,
+        editing: true,
+        variations: false,
+        styleControl: false,
+        qualityControl: true,
+        // Multiple resolution tiers: 0.5K, 1K, 2K, 4K
+        transparency: false,
+        promptRevision: false
+      },
+      limits: { maxPromptLength: 131072 },
+      // 131K input tokens
+      vendorOptions: {
+        outputImageResolution: {
+          type: "enum",
+          label: "Resolution",
+          description: "Output image resolution tier",
+          enum: ["0.5K", "1K", "2K", "4K"],
+          default: "1K",
+          controlType: "select"
+        }
+      }
+    },
+    pricing: {
+      // Per-image, varies by resolution: $0.045 (512px), $0.067 (1K), $0.101 (2K), $0.151 (4K)
+      perImageStandard: 0.067,
+      // 1K default
+      perImageHD: 0.151,
+      // 4K
+      currency: "USD"
+    }
+  },
+  "gemini-3-pro-image-preview": {
+    name: "gemini-3-pro-image-preview",
+    displayName: "Nano Banana Pro (Gemini 3 Pro Image)",
+    provider: Vendor.Google,
+    description: "Professional design engine with reasoning for studio-quality 4K visuals, complex layouts, and precise text rendering",
+    isActive: true,
+    releaseDate: "2025-11-01",
+    sources: {
+      documentation: "https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-image-preview",
+      pricing: "https://ai.google.dev/pricing",
+      lastVerified: "2026-03-04"
+    },
+    capabilities: {
+      sizes: ["1024x1024", "auto"],
+      aspectRatios: ["1:1", "3:4", "4:3", "9:16", "16:9"],
+      maxImagesPerRequest: 4,
+      outputFormats: ["png", "jpeg"],
+      features: {
+        generation: true,
+        editing: true,
+        variations: false,
+        styleControl: true,
+        // Reasoning-driven design
+        qualityControl: true,
+        // 1K, 2K, 4K tiers
+        transparency: false,
+        promptRevision: false
+      },
+      limits: { maxPromptLength: 65536 },
+      // 65K input tokens
+      vendorOptions: {
+        outputImageResolution: {
+          type: "enum",
+          label: "Resolution",
+          description: "Output image resolution tier",
+          enum: ["1K", "2K", "4K"],
+          default: "1K",
+          controlType: "select"
+        }
+      }
+    },
+    pricing: {
+      // $0.134 per 1K/2K image, $0.24 per 4K image
+      perImageStandard: 0.134,
+      // 1K/2K
+      perImageHD: 0.24,
+      // 4K
+      currency: "USD"
+    }
+  },
+  "gemini-2.5-flash-image": {
+    name: "gemini-2.5-flash-image",
+    displayName: "Nano Banana (Gemini 2.5 Flash Image)",
+    provider: Vendor.Google,
+    description: "Native image generation and editing designed for fast, creative workflows",
+    isActive: true,
+    releaseDate: "2025-10-01",
+    sources: {
+      documentation: "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-image",
+      pricing: "https://ai.google.dev/pricing",
+      lastVerified: "2026-03-04"
+    },
+    capabilities: {
+      sizes: ["1024x1024", "auto"],
+      aspectRatios: ["1:1", "3:4", "4:3", "9:16", "16:9"],
+      maxImagesPerRequest: 4,
+      outputFormats: ["png", "jpeg"],
+      features: {
+        generation: true,
+        editing: true,
+        variations: false,
+        styleControl: false,
+        qualityControl: false,
+        transparency: false,
+        promptRevision: false
+      },
+      limits: { maxPromptLength: 65536 }
+      // 65K input tokens
+    },
+    pricing: {
+      perImage: 0.039,
+      // $0.039 per image
+      currency: "USD"
+    }
+  },
   // ======================== xAI Grok ========================
   "grok-imagine-image": {
     name: "grok-imagine-image",
@@ -36041,11 +37164,11 @@ var IMAGE_MODEL_REGISTRY = {
     sources: {
       documentation: "https://docs.x.ai/docs/guides/image-generation",
       pricing: "https://docs.x.ai/docs/models",
-      lastVerified: "2026-02-01"
+      lastVerified: "2026-03-04"
     },
     capabilities: {
       sizes: ["1024x1024"],
-      aspectRatios: ["1:1", "4:3", "3:4", "16:9", "9:16", "3:2", "2:3"],
+      aspectRatios: ["1:1", "4:3", "3:4", "16:9", "9:16", "3:2", "2:3", "2:1", "1:2"],
       maxImagesPerRequest: 10,
       outputFormats: ["png", "jpeg"],
       features: {
@@ -37301,9 +38424,9 @@ var OPENAI_SOURCES = {
   lastVerified: "2026-01-25"
 };
 var GOOGLE_SOURCES = {
-  documentation: "https://docs.cloud.google.com/vertex-ai/generative-ai/docs/video/overview",
-  apiReference: "https://docs.cloud.google.com/vertex-ai/generative-ai/docs/model-reference/veo-video-generation",
-  lastVerified: "2026-01-25"
+  documentation: "https://ai.google.dev/gemini-api/docs/video",
+  apiReference: "https://ai.google.dev/gemini-api/docs/models/veo",
+  lastVerified: "2026-03-04"
 };
 var GROK_SOURCES = {
   documentation: "https://docs.x.ai/docs/guides/video-generations",
@@ -37377,14 +38500,16 @@ var VIDEO_MODEL_REGISTRY = {
     sources: GOOGLE_SOURCES,
     capabilities: {
       durations: [5, 6, 7, 8],
-      resolutions: [],
-      // Veo 2.0 uses aspectRatio only, no resolution control
+      resolutions: ["720p"],
+      // Veo 2 only supports 720p
       aspectRatios: ["16:9", "9:16"],
       maxFps: 24,
       audio: false,
-      imageToVideo: true,
+      imageToVideo: false,
+      // Veo 2 does not support reference images
       videoExtension: false,
       frameControl: true,
+      // First/last frame interpolation supported
       features: {
         upscaling: false,
         styleControl: false,
@@ -37393,7 +38518,8 @@ var VIDEO_MODEL_REGISTRY = {
       }
     },
     pricing: {
-      perSecond: 0.03,
+      perSecond: 0.35,
+      // Updated per official pricing page (was $0.03)
       currency: "USD"
     }
   },
@@ -37405,14 +38531,18 @@ var VIDEO_MODEL_REGISTRY = {
     sources: GOOGLE_SOURCES,
     capabilities: {
       durations: [4, 6, 8],
-      resolutions: ["720p"],
-      // Fast model only supports 720p
+      resolutions: ["720p", "1080p", "4k"],
+      // 1080p/4k require 8s duration
       aspectRatios: ["16:9", "9:16"],
       maxFps: 24,
       audio: true,
+      // Native audio generation
       imageToVideo: true,
-      videoExtension: false,
-      frameControl: false,
+      // Up to 3 reference images
+      videoExtension: true,
+      // Supported (720p only)
+      frameControl: true,
+      // First/last frame interpolation
       features: {
         upscaling: false,
         styleControl: false,
@@ -37421,7 +38551,8 @@ var VIDEO_MODEL_REGISTRY = {
       }
     },
     pricing: {
-      perSecond: 0.75,
+      perSecond: 0.15,
+      // $0.15 for 720p/1080p, $0.35 for 4K
       currency: "USD"
     }
   },
@@ -37438,8 +38569,11 @@ var VIDEO_MODEL_REGISTRY = {
       aspectRatios: ["16:9", "9:16"],
       maxFps: 30,
       audio: true,
+      // Native audio generation
       imageToVideo: true,
+      // Up to 3 reference images
       videoExtension: true,
+      // Supported (720p only)
       frameControl: true,
       features: {
         upscaling: true,
@@ -37449,7 +38583,8 @@ var VIDEO_MODEL_REGISTRY = {
       }
     },
     pricing: {
-      perSecond: 0.75,
+      perSecond: 0.4,
+      // $0.40 for 720p/1080p, $0.60 for 4K
       currency: "USD"
     }
   },
@@ -37673,6 +38808,694 @@ var VideoGeneration = class _VideoGeneration {
   }
 };
 
+// src/capabilities/speech/SentenceSplitter.ts
+var DEFAULT_ABBREVIATIONS = /* @__PURE__ */ new Set([
+  "dr.",
+  "mr.",
+  "mrs.",
+  "ms.",
+  "prof.",
+  "sr.",
+  "jr.",
+  "st.",
+  "ave.",
+  "blvd.",
+  "rd.",
+  "u.s.",
+  "u.k.",
+  "u.s.a.",
+  "u.n.",
+  "e.g.",
+  "i.e.",
+  "etc.",
+  "vs.",
+  "viz.",
+  "approx.",
+  "dept.",
+  "est.",
+  "inc.",
+  "ltd.",
+  "corp.",
+  "no.",
+  "vol.",
+  "rev.",
+  "gen.",
+  "gov.",
+  "jan.",
+  "feb.",
+  "mar.",
+  "apr.",
+  "jun.",
+  "jul.",
+  "aug.",
+  "sep.",
+  "oct.",
+  "nov.",
+  "dec.",
+  "fig.",
+  "eq.",
+  "ref.",
+  "sec.",
+  "ch.",
+  "min.",
+  "max.",
+  "avg."
+]);
+var DEFAULT_OPTIONS = {
+  minChunkLength: 20,
+  maxChunkLength: 500,
+  skipCodeBlocks: true,
+  stripMarkdown: true,
+  additionalAbbreviations: []
+};
+var SentenceChunkingStrategy = class {
+  buffer = "";
+  inCodeBlock = false;
+  codeBlockBuffer = "";
+  options;
+  abbreviations;
+  constructor(options) {
+    this.options = { ...DEFAULT_OPTIONS, ...options };
+    this.abbreviations = /* @__PURE__ */ new Set([
+      ...DEFAULT_ABBREVIATIONS,
+      ...this.options.additionalAbbreviations.map((a) => a.toLowerCase())
+    ]);
+  }
+  feed(delta) {
+    this.buffer += delta;
+    return this.extractChunks();
+  }
+  flush() {
+    if (this.inCodeBlock) {
+      this.codeBlockBuffer = "";
+      this.inCodeBlock = false;
+    }
+    const text = this.cleanForSpeech(this.buffer.trim());
+    this.buffer = "";
+    return text.length > 0 ? text : null;
+  }
+  reset() {
+    this.buffer = "";
+    this.inCodeBlock = false;
+    this.codeBlockBuffer = "";
+  }
+  // ======================== Private Methods ========================
+  extractChunks() {
+    const chunks = [];
+    if (this.options.skipCodeBlocks) {
+      this.processCodeBlocks();
+    }
+    let paragraphIdx = this.buffer.indexOf("\n\n");
+    while (paragraphIdx !== -1) {
+      const chunk = this.buffer.slice(0, paragraphIdx).trim();
+      this.buffer = this.buffer.slice(paragraphIdx + 2);
+      if (chunk.length > 0) {
+        const cleaned = this.cleanForSpeech(chunk);
+        if (cleaned.length > 0) {
+          chunks.push(cleaned);
+        }
+      }
+      paragraphIdx = this.buffer.indexOf("\n\n");
+    }
+    let sentenceEnd = this.findSentenceBoundary();
+    while (sentenceEnd !== -1) {
+      const sentence = this.buffer.slice(0, sentenceEnd).trim();
+      this.buffer = this.buffer.slice(sentenceEnd).trimStart();
+      if (sentence.length > 0) {
+        const cleaned = this.cleanForSpeech(sentence);
+        if (cleaned.length > 0) {
+          chunks.push(cleaned);
+        }
+      }
+      sentenceEnd = this.findSentenceBoundary();
+    }
+    if (this.buffer.length > this.options.maxChunkLength) {
+      const splitChunks = this.splitLongText(this.buffer);
+      this.buffer = splitChunks.pop() ?? "";
+      for (const chunk of splitChunks) {
+        const cleaned = this.cleanForSpeech(chunk.trim());
+        if (cleaned.length > 0) {
+          chunks.push(cleaned);
+        }
+      }
+    }
+    return this.mergeSmallChunks(chunks);
+  }
+  /**
+   * Track and remove fenced code blocks from the buffer.
+   * Text inside code blocks is discarded (not spoken).
+   */
+  processCodeBlocks() {
+    let idx = 0;
+    let result = "";
+    while (idx < this.buffer.length) {
+      if (this.buffer.startsWith("```", idx)) {
+        if (this.inCodeBlock) {
+          this.inCodeBlock = false;
+          this.codeBlockBuffer = "";
+          idx += 3;
+          const newline = this.buffer.indexOf("\n", idx);
+          idx = newline !== -1 ? newline + 1 : this.buffer.length;
+        } else {
+          this.inCodeBlock = true;
+          this.codeBlockBuffer = "";
+          idx += 3;
+          const newline = this.buffer.indexOf("\n", idx);
+          idx = newline !== -1 ? newline + 1 : this.buffer.length;
+        }
+      } else if (this.inCodeBlock) {
+        this.codeBlockBuffer += this.buffer[idx];
+        idx++;
+      } else {
+        result += this.buffer[idx];
+        idx++;
+      }
+    }
+    this.buffer = result;
+  }
+  /**
+   * Find the position right after the next sentence boundary.
+   * Returns -1 if no complete sentence boundary found.
+   */
+  findSentenceBoundary() {
+    const terminators = [".", "?", "!"];
+    for (let i = 0; i < this.buffer.length; i++) {
+      const ch = this.buffer.charAt(i);
+      if (!terminators.includes(ch)) continue;
+      if (i + 1 >= this.buffer.length) return -1;
+      const nextChar = this.buffer[i + 1];
+      if (nextChar !== " " && nextChar !== "\n" && nextChar !== "\r" && nextChar !== "	") {
+        continue;
+      }
+      if (ch === ".") {
+        if (this.isAbbreviation(i)) continue;
+        if (this.isDecimalNumber(i)) continue;
+        if (this.isEllipsis(i)) continue;
+      }
+      const candidate = this.buffer.slice(0, i + 1).trim();
+      if (candidate.length < this.options.minChunkLength) continue;
+      return i + 1;
+    }
+    return -1;
+  }
+  /**
+   * Check if the period at position `pos` is part of a known abbreviation.
+   */
+  isAbbreviation(pos) {
+    let wordStart = pos - 1;
+    while (wordStart >= 0 && this.buffer[wordStart] !== " " && this.buffer[wordStart] !== "\n") {
+      wordStart--;
+    }
+    wordStart++;
+    const word = this.buffer.slice(wordStart, pos + 1).toLowerCase();
+    return this.abbreviations.has(word);
+  }
+  /**
+   * Check if the period at position `pos` is a decimal point.
+   * e.g., 3.14, $1.50
+   */
+  isDecimalNumber(pos) {
+    if (pos === 0 || pos + 1 >= this.buffer.length) return false;
+    const before = this.buffer.charAt(pos - 1);
+    const after = this.buffer.charAt(pos + 1);
+    return /\d/.test(before) && /\d/.test(after);
+  }
+  /**
+   * Check if the period at position `pos` is part of an ellipsis (...).
+   */
+  isEllipsis(pos) {
+    if (pos >= 2 && this.buffer[pos - 1] === "." && this.buffer[pos - 2] === ".") return true;
+    if (pos + 1 < this.buffer.length && this.buffer[pos + 1] === ".") return true;
+    return false;
+  }
+  /**
+   * Split text that exceeds maxChunkLength at clause boundaries.
+   */
+  splitLongText(text) {
+    const max = this.options.maxChunkLength;
+    const chunks = [];
+    let remaining = text;
+    while (remaining.length > max) {
+      let splitPos = -1;
+      const clauseBreaks = [",", ";", ":", " \u2014", " \u2013", " -"];
+      for (const brk of clauseBreaks) {
+        const searchRegion = remaining.slice(0, max);
+        const lastPos = searchRegion.lastIndexOf(brk);
+        if (lastPos > this.options.minChunkLength) {
+          splitPos = lastPos + brk.length;
+          break;
+        }
+      }
+      if (splitPos === -1) {
+        const searchRegion = remaining.slice(0, max);
+        splitPos = searchRegion.lastIndexOf(" ");
+        if (splitPos <= this.options.minChunkLength) {
+          splitPos = max;
+        }
+      }
+      chunks.push(remaining.slice(0, splitPos));
+      remaining = remaining.slice(splitPos);
+    }
+    chunks.push(remaining);
+    return chunks;
+  }
+  /**
+   * Merge chunks that are shorter than minChunkLength with the next chunk.
+   */
+  mergeSmallChunks(chunks) {
+    if (chunks.length <= 1) return chunks;
+    const merged = [];
+    let accumulator = "";
+    for (const chunk of chunks) {
+      if (accumulator.length > 0) {
+        accumulator += " " + chunk;
+      } else {
+        accumulator = chunk;
+      }
+      if (accumulator.length >= this.options.minChunkLength) {
+        merged.push(accumulator);
+        accumulator = "";
+      }
+    }
+    if (accumulator.length > 0) {
+      if (merged.length > 0) {
+        merged[merged.length - 1] += " " + accumulator;
+      } else {
+        merged.push(accumulator);
+      }
+    }
+    return merged;
+  }
+  /**
+   * Strip markdown formatting from text for natural speech.
+   */
+  cleanForSpeech(text) {
+    if (!this.options.stripMarkdown) return text;
+    let cleaned = text;
+    cleaned = cleaned.replace(/`([^`]+)`/g, "$1");
+    cleaned = cleaned.replace(/\*\*([^*]+)\*\*/g, "$1");
+    cleaned = cleaned.replace(/__([^_]+)__/g, "$1");
+    cleaned = cleaned.replace(/(?<!\*)\*([^*]+)\*(?!\*)/g, "$1");
+    cleaned = cleaned.replace(/(?<!_)_([^_]+)_(?!_)/g, "$1");
+    cleaned = cleaned.replace(/~~([^~]+)~~/g, "$1");
+    cleaned = cleaned.replace(/\[([^\]]+)\]\([^)]+\)/g, "$1");
+    cleaned = cleaned.replace(/!\[([^\]]*)\]\([^)]+\)/g, "");
+    cleaned = cleaned.replace(/^#{1,6}\s+/gm, "");
+    cleaned = cleaned.replace(/^[-*+]\s+/gm, "");
+    cleaned = cleaned.replace(/^\d+\.\s+/gm, "");
+    cleaned = cleaned.replace(/^>\s+/gm, "");
+    cleaned = cleaned.replace(/^[-*_]{3,}\s*$/gm, "");
+    cleaned = cleaned.replace(/\n+/g, " ");
+    cleaned = cleaned.replace(/\s{2,}/g, " ");
+    return cleaned.trim();
+  }
+};
+
+// src/capabilities/speech/VoiceStream.ts
+var VoiceStream = class _VoiceStream extends events.EventEmitter {
+  tts;
+  chunker;
+  format;
+  speed;
+  maxConcurrentTTS;
+  maxQueuedChunks;
+  vendorOptions;
+  streaming;
+  // Pipeline state
+  chunkIndex = 0;
+  totalCharacters = 0;
+  totalDuration = 0;
+  activeJobs = /* @__PURE__ */ new Map();
+  activeTTSCount = 0;
+  interrupted = false;
+  lastResponseId = "";
+  _isDestroyed = false;
+  // Semaphore for TTS concurrency control
+  slotWaiters = [];
+  // Audio event buffer for interleaving with text events
+  audioEventBuffer = [];
+  // Async notification: resolves when new events are pushed to audioEventBuffer
+  bufferNotify = null;
+  // Queue backpressure
+  queueWaiters = [];
+  /**
+   * Create a new VoiceStream instance
+   */
+  static create(config) {
+    return new _VoiceStream(config);
+  }
+  constructor(config) {
+    super();
+    this.tts = TextToSpeech.create({
+      connector: config.ttsConnector,
+      model: config.ttsModel,
+      voice: config.voice
+    });
+    this.chunker = config.chunkingStrategy ?? new SentenceChunkingStrategy(config.chunkingOptions);
+    this.format = config.format ?? "mp3";
+    this.speed = config.speed ?? 1;
+    this.maxConcurrentTTS = config.maxConcurrentTTS ?? 2;
+    this.maxQueuedChunks = config.maxQueuedChunks ?? 5;
+    this.vendorOptions = config.vendorOptions;
+    this.streaming = config.streaming ?? false;
+  }
+  // ======================== Public API ========================
+  /**
+   * Transform an agent text stream into an augmented stream with audio events.
+   * Original text events pass through unchanged; audio events are interleaved.
+   *
+   * The generator yields events in this order:
+   * 1. All original StreamEvents (pass-through)
+   * 2. AudioChunkReady/AudioChunkError events as TTS completes
+   * 3. AudioStreamComplete as the final audio event
+   */
+  async *wrap(textStream) {
+    this.reset();
+    try {
+      for await (const event of textStream) {
+        yield event;
+        if (event.response_id) {
+          this.lastResponseId = event.response_id;
+        }
+        if (event.type === "response.output_text.delta" /* OUTPUT_TEXT_DELTA */ && !this.interrupted) {
+          const completedChunks = this.chunker.feed(event.delta);
+          for (const chunk of completedChunks) {
+            await this.scheduleTTS(chunk);
+          }
+        }
+        if ((event.type === "response.output_text.done" /* OUTPUT_TEXT_DONE */ || event.type === "response.complete" /* RESPONSE_COMPLETE */) && !this.interrupted) {
+          const remaining = this.chunker.flush();
+          if (remaining) {
+            await this.scheduleTTS(remaining);
+          }
+        }
+        yield* this.drainAudioBuffer();
+      }
+      while (this.activeJobs.size > 0 || this.audioEventBuffer.length > 0) {
+        if (this.audioEventBuffer.length === 0) {
+          await Promise.race([
+            this.waitForBufferNotify(),
+            ...Array.from(this.activeJobs.values()).map((j) => j.promise)
+          ]);
+        }
+        yield* this.drainAudioBuffer();
+      }
+      if (this.chunkIndex > 0) {
+        const completeEvent = {
+          type: "response.audio_stream.complete" /* AUDIO_STREAM_COMPLETE */,
+          response_id: this.lastResponseId,
+          total_chunks: this.chunkIndex,
+          total_characters: this.totalCharacters,
+          total_duration_seconds: this.totalDuration > 0 ? this.totalDuration : void 0
+        };
+        yield completeEvent;
+        this.emit("audio:complete", {
+          totalChunks: this.chunkIndex,
+          totalDurationSeconds: this.totalDuration > 0 ? this.totalDuration : void 0
+        });
+      }
+    } finally {
+      this.cleanup();
+    }
+  }
+  /**
+   * Interrupt audio generation. Cancels pending TTS and flushes queue.
+   * Call this when the user sends a new message mid-speech.
+   * Active HTTP requests cannot be cancelled but their results will be discarded.
+   */
+  interrupt() {
+    this.interrupted = true;
+    const pendingCount = this.activeJobs.size;
+    this.activeJobs.clear();
+    this.activeTTSCount = 0;
+    this.audioEventBuffer = [];
+    this.releaseAllWaiters();
+    this.chunker.reset();
+    this.emit("audio:interrupted", { pendingChunks: pendingCount });
+  }
+  /**
+   * Reset state for a new stream. Called automatically by wrap().
+   */
+  reset() {
+    this.chunkIndex = 0;
+    this.totalCharacters = 0;
+    this.totalDuration = 0;
+    this.activeJobs.clear();
+    this.activeTTSCount = 0;
+    this.interrupted = false;
+    this.lastResponseId = "";
+    this.audioEventBuffer = [];
+    this.bufferNotify = null;
+    this.slotWaiters = [];
+    this.queueWaiters = [];
+    this.chunker.reset();
+  }
+  destroy() {
+    this.interrupt();
+    this._isDestroyed = true;
+    this.removeAllListeners();
+  }
+  get isDestroyed() {
+    return this._isDestroyed;
+  }
+  // ======================== Private Methods ========================
+  /**
+   * Schedule a text chunk for TTS synthesis.
+   * Awaits a free queue slot if backpressure is active (lossless).
+   */
+  async scheduleTTS(text) {
+    if (this.interrupted || this._isDestroyed) return;
+    const cleanText = text.trim();
+    if (cleanText.length === 0) return;
+    while (this.activeJobs.size >= this.maxQueuedChunks && !this.interrupted) {
+      await this.waitForQueueSlot();
+    }
+    if (this.interrupted) return;
+    const index = this.chunkIndex++;
+    this.totalCharacters += cleanText.length;
+    const job = {
+      index,
+      text: cleanText,
+      promise: this.executeTTS(index, cleanText)
+    };
+    this.activeJobs.set(index, job);
+    job.promise.finally(() => {
+      this.activeJobs.delete(index);
+      this.releaseQueueWaiter();
+    });
+  }
+  /**
+   * Execute TTS for a single text chunk.
+   * Respects concurrency semaphore.
+   * Branches on streaming mode: yields sub-chunks or a single buffered chunk.
+   */
+  async executeTTS(index, text) {
+    while (this.activeTTSCount >= this.maxConcurrentTTS && !this.interrupted) {
+      await this.waitForTTSSlot();
+    }
+    if (this.interrupted) return;
+    this.activeTTSCount++;
+    try {
+      const ttsStart = Date.now();
+      if (this.streaming && this.tts.supportsStreaming(this.format)) {
+        let subIndex = 0;
+        const streamFormat = this.format === "mp3" ? "pcm" : this.format;
+        const MIN_BUFFER_BYTES = 6e3;
+        const pendingBuffers = [];
+        let pendingSize = 0;
+        const flushPending = () => {
+          if (pendingSize === 0) return;
+          const merged = Buffer.concat(pendingBuffers, pendingSize);
+          pendingBuffers.length = 0;
+          pendingSize = 0;
+          const currentSubIndex = subIndex++;
+          const audioEvent = {
+            type: "response.audio_chunk.ready" /* AUDIO_CHUNK_READY */,
+            response_id: this.lastResponseId,
+            chunk_index: index,
+            sub_index: currentSubIndex,
+            text: currentSubIndex === 0 ? text : "",
+            audio_base64: merged.toString("base64"),
+            format: streamFormat
+          };
+          this.pushAudioEvent(audioEvent);
+        };
+        for await (const chunk of this.tts.synthesizeStream(text, {
+          format: streamFormat,
+          speed: this.speed,
+          vendorOptions: this.vendorOptions
+        })) {
+          if (this.interrupted) return;
+          if (chunk.audio.length > 0) {
+            pendingBuffers.push(chunk.audio);
+            pendingSize += chunk.audio.length;
+            if (pendingSize >= MIN_BUFFER_BYTES) {
+              flushPending();
+            }
+          }
+          if (chunk.isFinal) {
+            break;
+          }
+        }
+        flushPending();
+        console.log(`[VoiceStream] TTS chunk ${index} streamed ${subIndex} sub-chunks in ${Date.now() - ttsStart}ms, text: "${text.slice(0, 40)}..."`);
+        this.emit("audio:ready", { chunkIndex: index, text });
+      } else {
+        const response = await this.tts.synthesize(text, {
+          format: this.format,
+          speed: this.speed,
+          vendorOptions: this.vendorOptions
+        });
+        if (this.interrupted) return;
+        if (response.durationSeconds) {
+          this.totalDuration += response.durationSeconds;
+        }
+        const audioEvent = {
+          type: "response.audio_chunk.ready" /* AUDIO_CHUNK_READY */,
+          response_id: this.lastResponseId,
+          chunk_index: index,
+          text,
+          audio_base64: response.audio.toString("base64"),
+          format: response.format,
+          duration_seconds: response.durationSeconds,
+          characters_used: response.charactersUsed
+        };
+        this.pushAudioEvent(audioEvent);
+        console.log(`[VoiceStream] TTS chunk ${index} ready in ${Date.now() - ttsStart}ms, text: "${text.slice(0, 40)}..."`);
+        this.emit("audio:ready", {
+          chunkIndex: index,
+          text,
+          durationSeconds: response.durationSeconds
+        });
+      }
+    } catch (error) {
+      if (this.interrupted) return;
+      const errorEvent = {
+        type: "response.audio_chunk.error" /* AUDIO_CHUNK_ERROR */,
+        response_id: this.lastResponseId,
+        chunk_index: index,
+        text,
+        error: error.message
+      };
+      this.pushAudioEvent(errorEvent);
+      this.emit("audio:error", {
+        chunkIndex: index,
+        text,
+        error
+      });
+    } finally {
+      this.activeTTSCount--;
+      this.releaseTTSSlot();
+    }
+  }
+  /**
+   * Drain the audio event buffer, yielding all ready events.
+   */
+  *drainAudioBuffer() {
+    while (this.audioEventBuffer.length > 0) {
+      yield this.audioEventBuffer.shift();
+    }
+  }
+  // ======================== Buffer Notification ========================
+  /**
+   * Push an audio event and wake up the consumer in wrap()
+   */
+  pushAudioEvent(event) {
+    this.audioEventBuffer.push(event);
+    if (this.bufferNotify) {
+      this.bufferNotify();
+      this.bufferNotify = null;
+    }
+  }
+  /**
+   * Wait until a new event is pushed to the audio buffer
+   */
+  waitForBufferNotify() {
+    return new Promise((resolve4) => {
+      this.bufferNotify = resolve4;
+    });
+  }
+  // ======================== Semaphore / Backpressure ========================
+  waitForTTSSlot() {
+    return new Promise((resolve4) => {
+      this.slotWaiters.push(resolve4);
+    });
+  }
+  releaseTTSSlot() {
+    const waiter = this.slotWaiters.shift();
+    if (waiter) waiter();
+  }
+  waitForQueueSlot() {
+    return new Promise((resolve4) => {
+      this.queueWaiters.push(resolve4);
+    });
+  }
+  releaseQueueWaiter() {
+    const waiter = this.queueWaiters.shift();
+    if (waiter) waiter();
+  }
+  releaseAllWaiters() {
+    for (const waiter of this.slotWaiters) waiter();
+    this.slotWaiters = [];
+    for (const waiter of this.queueWaiters) waiter();
+    this.queueWaiters = [];
+    if (this.bufferNotify) {
+      this.bufferNotify();
+      this.bufferNotify = null;
+    }
+  }
+  cleanup() {
+    this.releaseAllWaiters();
+  }
+};
+
+// src/capabilities/speech/AudioPlaybackQueue.ts
+var AudioPlaybackQueue = class {
+  buffer = /* @__PURE__ */ new Map();
+  nextPlayIndex = 0;
+  onReady;
+  constructor(onReady) {
+    this.onReady = onReady;
+  }
+  /**
+   * Enqueue an audio chunk event. If it's the next expected chunk,
+   * it (and any subsequent buffered chunks) are immediately delivered
+   * to the callback in order.
+   */
+  enqueue(event) {
+    this.buffer.set(event.chunk_index, event);
+    this.drain();
+  }
+  /**
+   * Reset the queue (e.g., on interruption or new stream).
+   */
+  reset() {
+    this.buffer.clear();
+    this.nextPlayIndex = 0;
+  }
+  /**
+   * Number of chunks currently buffered waiting for earlier chunks.
+   */
+  get pendingCount() {
+    return this.buffer.size;
+  }
+  /**
+   * The next chunk index expected for playback.
+   */
+  get nextExpectedIndex() {
+    return this.nextPlayIndex;
+  }
+  // ======================== Private ========================
+  drain() {
+    while (this.buffer.has(this.nextPlayIndex)) {
+      const event = this.buffer.get(this.nextPlayIndex);
+      this.buffer.delete(this.nextPlayIndex);
+      this.nextPlayIndex++;
+      this.onReady(event);
+    }
+  }
+};
+
 // src/capabilities/search/SearchProvider.ts
 init_Connector();
 
@@ -42352,6 +44175,14 @@ var SERVICE_DEFINITIONS = [
     baseURL: "https://aws.amazon.com",
     docsURL: "https://docs.aws.amazon.com/"
   },
+  {
+    id: "cloudflare",
+    name: "Cloudflare",
+    category: "cloud",
+    urlPattern: /api\.cloudflare\.com/i,
+    baseURL: "https://api.cloudflare.com/client/v4",
+    docsURL: "https://developers.cloudflare.com/api/"
+  },
   // ============ Storage ============
   {
     id: "dropbox",
@@ -42395,6 +44226,14 @@ var SERVICE_DEFINITIONS = [
     baseURL: "https://api.postmarkapp.com",
     docsURL: "https://postmarkapp.com/developer"
   },
+  {
+    id: "mailgun",
+    name: "Mailgun",
+    category: "email",
+    urlPattern: /api\.mailgun\.net|api\.eu\.mailgun\.net/i,
+    baseURL: "https://api.mailgun.net/v3",
+    docsURL: "https://documentation.mailgun.com/docs/mailgun/api-reference/"
+  },
   // ============ Monitoring & Observability ============
   {
     id: "datadog",
@@ -45134,6 +46973,43 @@ var awsTemplate = {
   ]
 };
 
+// src/connectors/vendors/templates/cloudflare.ts
+var cloudflareTemplate = {
+  id: "cloudflare",
+  name: "Cloudflare",
+  serviceType: "cloudflare",
+  baseURL: "https://api.cloudflare.com/client/v4",
+  docsURL: "https://developers.cloudflare.com/api/",
+  credentialsSetupURL: "https://dash.cloudflare.com/profile/api-tokens",
+  category: "cloud",
+  notes: "API Tokens (recommended) are scoped and more secure. Global API Key requires email and has full account access.",
+  authTemplates: [
+    {
+      id: "api-token",
+      name: "API Token",
+      type: "api_key",
+      description: "Scoped API token (recommended). Create at dash.cloudflare.com > My Profile > API Tokens",
+      requiredFields: ["apiKey"],
+      defaults: {
+        type: "api_key",
+        headerName: "Authorization",
+        headerPrefix: "Bearer"
+      }
+    },
+    {
+      id: "global-api-key",
+      name: "Global API Key",
+      type: "api_key",
+      description: "Legacy global API key + email. Has full account access. Prefer API Tokens for least-privilege access",
+      requiredFields: ["apiKey", "username"],
+      defaults: {
+        type: "api_key",
+        headerName: "X-Auth-Key"
+      }
+    }
+  ]
+};
+
 // src/connectors/vendors/templates/dropbox.ts
 var dropboxTemplate = {
   id: "dropbox",
@@ -45322,6 +47198,30 @@ var postmarkTemplate = {
     }
   ]
 };
+var mailgunTemplate = {
+  id: "mailgun",
+  name: "Mailgun",
+  serviceType: "mailgun",
+  baseURL: "https://api.mailgun.net/v3",
+  docsURL: "https://documentation.mailgun.com/docs/mailgun/api-reference/",
+  credentialsSetupURL: "https://app.mailgun.com/settings/api_security",
+  category: "email",
+  notes: "EU region uses api.eu.mailgun.net. Most endpoints require /v3/<domain> in the path.",
+  authTemplates: [
+    {
+      id: "api-key",
+      name: "API Key",
+      type: "api_key",
+      description: "Private API key for full account access. Find at Settings > API Security",
+      requiredFields: ["apiKey"],
+      defaults: {
+        type: "api_key",
+        headerName: "Authorization",
+        headerPrefix: "Basic"
+      }
+    }
+  ]
+};
 
 // src/connectors/vendors/templates/monitoring.ts
 var datadogTemplate = {
@@ -45770,6 +47670,7 @@ var allVendorTemplates = [
   rampTemplate,
   // Cloud
   awsTemplate,
+  cloudflareTemplate,
   // Storage
   dropboxTemplate,
   boxTemplate,
@@ -45777,6 +47678,7 @@ var allVendorTemplates = [
   sendgridTemplate,
   mailchimpTemplate,
   postmarkTemplate,
+  mailgunTemplate,
   // Monitoring
   datadogTemplate,
   pagerdutyTemplate,
@@ -49134,7 +51036,8 @@ SANDBOX API:
 4. connectors.get(name) \u2014 Connector info: { displayName, description, baseURL, serviceType }
 
 VARIABLES:
-   \u2022 input \u2014 data passed via the "input" parameter (default: {})
+   \u2022 input \u2014 data passed via the "input" parameter (default: {}). Always a parsed object/array, never a string.
+     CRITICAL: You MUST pass actual data values directly. Template placeholders ({{results}}, {{param.name}}, etc.) are NOT supported and will be passed as literal strings. If you need data from a previous tool call, include the actual returned data in the input object.
    \u2022 output \u2014 SET THIS to return your result to the caller
 
 GLOBALS: console.log/error/warn, JSON, Math, Date, Buffer, Promise, Array, Object, String, Number, Boolean, setTimeout, setInterval, URL, URLSearchParams, RegExp, Map, Set, Error, TextEncoder, TextDecoder
@@ -49157,7 +51060,8 @@ const resp = await authenticatedFetch('/chat.postMessage', {
 }, 'slack');
 output = await resp.json();
 ${accountIdExamples}
-// Data processing (no API needed)
+// Data processing \u2014 pass actual data via the input parameter, NOT template references
+// e.g. call with: { "code": "...", "input": { "data": [{"score": 0.9}, {"score": 0.5}] } }
 const items = input.data;
 output = items.filter(i => i.score > 0.8).sort((a, b) => b.score - a.score);
 
@@ -49181,7 +51085,7 @@ function createExecuteJavaScriptTool(options) {
               description: 'JavaScript code to execute. Set the "output" variable with your result. Code is auto-wrapped in async IIFE \u2014 you can use await directly. For explicit async control, wrap in (async () => { ... })().'
             },
             input: {
-              description: 'Optional data available as the "input" variable in your code. Can be any JSON value.'
+              description: 'Optional data available as the "input" variable in your code. IMPORTANT: Pass actual data directly as a JSON object/array. Template placeholders like {{results}} or {{param.name}} are NOT supported here and will be passed as literal strings. You must include the actual data values inline. Correct: "input": {"deals": [{"id":"1"}, ...]}. Wrong: "input": {"deals": "{{results}}"}.'
             },
             timeout: {
               type: "number",
@@ -49204,9 +51108,19 @@ function createExecuteJavaScriptTool(options) {
       try {
         const timeout = Math.min(Math.max(args.timeout || defaultTimeout, 0), maxTimeout);
         const registry = context?.connectorRegistry ?? exports.Connector.asRegistry();
+        let resolvedInput = args.input;
+        if (typeof resolvedInput === "string") {
+          const trimmed = resolvedInput.trim();
+          if (trimmed.startsWith("{") && trimmed.endsWith("}") || trimmed.startsWith("[") && trimmed.endsWith("]")) {
+            try {
+              resolvedInput = JSON.parse(trimmed);
+            } catch {
+            }
+          }
+        }
         const result = await executeInVM(
           args.code,
-          args.input,
+          resolvedInput,
           timeout,
           logs,
           context?.userId,
@@ -54622,6 +56536,7 @@ exports.APPROVAL_STATE_VERSION = APPROVAL_STATE_VERSION;
 exports.Agent = Agent;
 exports.AgentContextNextGen = AgentContextNextGen;
 exports.ApproximateTokenEstimator = ApproximateTokenEstimator;
+exports.AudioPlaybackQueue = AudioPlaybackQueue;
 exports.BaseMediaProvider = BaseMediaProvider;
 exports.BasePluginNextGen = BasePluginNextGen;
 exports.BaseProvider = BaseProvider;
@@ -54716,6 +56631,7 @@ exports.STT_MODELS = STT_MODELS;
 exports.STT_MODEL_REGISTRY = STT_MODEL_REGISTRY;
 exports.ScrapeProvider = ScrapeProvider;
 exports.SearchProvider = SearchProvider;
+exports.SentenceChunkingStrategy = SentenceChunkingStrategy;
 exports.SerperProvider = SerperProvider;
 exports.Services = Services;
 exports.SimpleScheduler = SimpleScheduler;
@@ -54751,6 +56667,7 @@ exports.VIDEO_MODELS = VIDEO_MODELS;
 exports.VIDEO_MODEL_REGISTRY = VIDEO_MODEL_REGISTRY;
 exports.Vendor = Vendor;
 exports.VideoGeneration = VideoGeneration;
+exports.VoiceStream = VoiceStream;
 exports.WorkingMemory = WorkingMemory;
 exports.WorkingMemoryPluginNextGen = WorkingMemoryPluginNextGen;
 exports.addJitter = addJitter;
@@ -54939,6 +56856,9 @@ exports.grep = grep;
 exports.hasClipboardImage = hasClipboardImage;
 exports.hasVendorLogo = hasVendorLogo;
 exports.hydrateCustomTool = hydrateCustomTool;
+exports.isAudioChunkError = isAudioChunkError;
+exports.isAudioChunkReady = isAudioChunkReady;
+exports.isAudioStreamComplete = isAudioStreamComplete;
 exports.isBlockedCommand = isBlockedCommand;
 exports.isErrorEvent = isErrorEvent;
 exports.isExcludedExtension = isExcludedExtension;