noumen 0.6.0 → 0.8.0

package/dist/chunk-7IQCQI2G.js

@@ -1,94 +0,0 @@
-// src/providers/resolve.ts
-var ENV_KEY_MAP = {
-  openai: "OPENAI_API_KEY",
-  anthropic: "ANTHROPIC_API_KEY",
-  gemini: "GEMINI_API_KEY",
-  openrouter: "OPENROUTER_API_KEY"
-};
-var DEFAULT_MODELS = {
-  openai: "gpt-5.4",
-  anthropic: "claude-opus-4.6",
-  gemini: "gemini-2.5-flash",
-  openrouter: "anthropic/claude-opus-4.6",
-  bedrock: "us.anthropic.claude-opus-4.6-v1:0",
-  vertex: "claude-opus-4.6",
-  ollama: "qwen2.5-coder:32b"
-};
-var SUPPORTED_PROVIDERS = Object.keys(DEFAULT_MODELS);
-function getProviderEnvKey(name) {
-  const envVar = ENV_KEY_MAP[name];
-  return envVar ? process.env[envVar] : void 0;
-}
-async function resolveProvider(input, opts) {
-  if (typeof input !== "string") return input;
-  const name = input;
-  if (!SUPPORTED_PROVIDERS.includes(name)) {
-    throw new Error(
-      `Unknown provider "${name}". Supported: ${SUPPORTED_PROVIDERS.join(", ")}`
-    );
-  }
-  const key = opts?.apiKey ?? getProviderEnvKey(name) ?? process.env.NOUMEN_API_KEY;
-  switch (name) {
-    case "openai": {
-      if (!key) throw new Error("OpenAI requires an API key. Set OPENAI_API_KEY or pass apiKey.");
-      const { OpenAIProvider } = await import("./providers/openai.js");
-      return new OpenAIProvider({ apiKey: key, model: opts?.model, baseURL: opts?.baseURL });
-    }
-    case "anthropic": {
-      if (!key) throw new Error("Anthropic requires an API key. Set ANTHROPIC_API_KEY or pass apiKey.");
-      const { AnthropicProvider } = await import("./providers/anthropic.js");
-      return new AnthropicProvider({ apiKey: key, model: opts?.model, baseURL: opts?.baseURL, cacheControl: { enabled: true } });
-    }
-    case "gemini": {
-      if (!key) throw new Error("Gemini requires an API key. Set GEMINI_API_KEY or pass apiKey.");
-      const { GeminiProvider } = await import("./providers/gemini.js");
-      return new GeminiProvider({ apiKey: key, model: opts?.model, baseURL: opts?.baseURL });
-    }
-    case "openrouter": {
-      if (!key) throw new Error("OpenRouter requires an API key. Set OPENROUTER_API_KEY or pass apiKey.");
-      const { OpenRouterProvider } = await import("./providers/openrouter.js");
-      return new OpenRouterProvider({ apiKey: key, model: opts?.model, appName: "noumen" });
-    }
-    case "bedrock": {
-      const { BedrockAnthropicProvider } = await import("./providers/bedrock.js");
-      return new BedrockAnthropicProvider({ model: opts?.model, baseURL: opts?.baseURL });
-    }
-    case "vertex": {
-      const { VertexAnthropicProvider } = await import("./providers/vertex.js");
-      return new VertexAnthropicProvider({ model: opts?.model, baseURL: opts?.baseURL });
-    }
-    case "ollama": {
-      const { OllamaProvider } = await import("./providers/ollama.js");
-      const base = opts?.baseURL ?? (process.env.OLLAMA_HOST ? `${process.env.OLLAMA_HOST.replace(/\/+$/, "")}/v1` : void 0);
-      return new OllamaProvider({ model: opts?.model, baseURL: base });
-    }
-    default:
-      throw new Error(`Unhandled provider: ${name}`);
-  }
-}
-async function detectProvider() {
-  if (process.env.ANTHROPIC_API_KEY) return "anthropic";
-  if (process.env.OPENAI_API_KEY) return "openai";
-  if (process.env.GEMINI_API_KEY) return "gemini";
-  if (process.env.OPENROUTER_API_KEY) return "openrouter";
-  if (process.env.AWS_ACCESS_KEY_ID || process.env.AWS_PROFILE) return "bedrock";
-  if (process.env.GOOGLE_APPLICATION_CREDENTIALS || process.env.GCLOUD_PROJECT) return "vertex";
-  if (process.env.OLLAMA_HOST) return "ollama";
-  try {
-    const host = process.env.OLLAMA_HOST ?? "http://localhost:11434";
-    const res = await fetch(`${host.replace(/\/+$/, "")}/api/tags`, {
-      signal: AbortSignal.timeout(500)
-    });
-    if (res.ok) return "ollama";
-  } catch {
-  }
-  return void 0;
-}
-
-export {
-  DEFAULT_MODELS,
-  SUPPORTED_PROVIDERS,
-  resolveProvider,
-  detectProvider
-};
-//# sourceMappingURL=chunk-7IQCQI2G.js.map

package/dist/chunk-7IQCQI2G.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/providers/resolve.ts"],"sourcesContent":["import type { AIProvider } from \"./types.js\";\n\nexport type ProviderName =\n  | \"openai\"\n  | \"anthropic\"\n  | \"gemini\"\n  | \"openrouter\"\n  | \"bedrock\"\n  | \"vertex\"\n  | \"ollama\";\n\nconst ENV_KEY_MAP: Record<string, string> = {\n  openai: \"OPENAI_API_KEY\",\n  anthropic: \"ANTHROPIC_API_KEY\",\n  gemini: \"GEMINI_API_KEY\",\n  openrouter: \"OPENROUTER_API_KEY\",\n};\n\nexport const DEFAULT_MODELS: Record<string, string> = {\n  openai: \"gpt-5.4\",\n  anthropic: \"claude-opus-4.6\",\n  gemini: \"gemini-2.5-flash\",\n  openrouter: \"anthropic/claude-opus-4.6\",\n  bedrock: \"us.anthropic.claude-opus-4.6-v1:0\",\n  vertex: \"claude-opus-4.6\",\n  ollama: \"qwen2.5-coder:32b\",\n};\n\nexport const SUPPORTED_PROVIDERS: ProviderName[] = Object.keys(DEFAULT_MODELS) as ProviderName[];\n\nfunction getProviderEnvKey(name: string): string | undefined {\n  const envVar = ENV_KEY_MAP[name];\n  return envVar ? process.env[envVar] : undefined;\n}\n\nexport interface ResolveProviderOptions {\n  apiKey?: string;\n  model?: string;\n  baseURL?: string;\n}\n\n/**\n * Resolve a provider from a name string or pass through an AIProvider instance.\n * API key resolution order:\n * 1. Explicit apiKey option\n * 2. Provider-specific env var (OPENAI_API_KEY, etc.)\n * 3. NOUMEN_API_KEY generic env var\n */\nexport async function resolveProvider(\n  input: AIProvider | ProviderName,\n  opts?: ResolveProviderOptions,\n): Promise<AIProvider> {\n  if (typeof input !== \"string\") return input;\n\n  const name = input;\n  if (!SUPPORTED_PROVIDERS.includes(name)) {\n    throw new Error(\n      `Unknown provider \"${name}\". Supported: ${SUPPORTED_PROVIDERS.join(\", \")}`,\n    );\n  }\n\n  const key =\n    opts?.apiKey ??\n    getProviderEnvKey(name) ??\n    process.env.NOUMEN_API_KEY;\n\n  switch (name) {\n    case \"openai\": {\n      if (!key) throw new Error(\"OpenAI requires an API key. Set OPENAI_API_KEY or pass apiKey.\");\n      const { OpenAIProvider } = await import(\"./openai.js\");\n      return new OpenAIProvider({ apiKey: key, model: opts?.model, baseURL: opts?.baseURL });\n    }\n    case \"anthropic\": {\n      if (!key) throw new Error(\"Anthropic requires an API key. Set ANTHROPIC_API_KEY or pass apiKey.\");\n      const { AnthropicProvider } = await import(\"./anthropic.js\");\n      return new AnthropicProvider({ apiKey: key, model: opts?.model, baseURL: opts?.baseURL, cacheControl: { enabled: true } });\n    }\n    case \"gemini\": {\n      if (!key) throw new Error(\"Gemini requires an API key. Set GEMINI_API_KEY or pass apiKey.\");\n      const { GeminiProvider } = await import(\"./gemini.js\");\n      return new GeminiProvider({ apiKey: key, model: opts?.model, baseURL: opts?.baseURL });\n    }\n    case \"openrouter\": {\n      if (!key) throw new Error(\"OpenRouter requires an API key. Set OPENROUTER_API_KEY or pass apiKey.\");\n      const { OpenRouterProvider } = await import(\"./openrouter.js\");\n      return new OpenRouterProvider({ apiKey: key, model: opts?.model, appName: \"noumen\" });\n    }\n    case \"bedrock\": {\n      const { BedrockAnthropicProvider } = await import(\"./bedrock.js\");\n      return new BedrockAnthropicProvider({ model: opts?.model, baseURL: opts?.baseURL });\n    }\n    case \"vertex\": {\n      const { VertexAnthropicProvider } = await import(\"./vertex.js\");\n      return new VertexAnthropicProvider({ model: opts?.model, baseURL: opts?.baseURL });\n    }\n    case \"ollama\": {\n      const { OllamaProvider } = await import(\"./ollama.js\");\n      const base = opts?.baseURL ?? (process.env.OLLAMA_HOST\n        ? `${process.env.OLLAMA_HOST.replace(/\\/+$/, \"\")}/v1`\n        : undefined);\n      return new OllamaProvider({ model: opts?.model, baseURL: base });\n    }\n    default:\n      throw new Error(`Unhandled provider: ${name}`);\n  }\n}\n\n/**\n * Auto-detect provider from available environment variables.\n */\nexport async function detectProvider(): Promise<ProviderName | undefined> {\n  if (process.env.ANTHROPIC_API_KEY) return \"anthropic\";\n  if (process.env.OPENAI_API_KEY) return \"openai\";\n  if (process.env.GEMINI_API_KEY) return \"gemini\";\n  if (process.env.OPENROUTER_API_KEY) return \"openrouter\";\n  if (process.env.AWS_ACCESS_KEY_ID || process.env.AWS_PROFILE) return \"bedrock\";\n  if (process.env.GOOGLE_APPLICATION_CREDENTIALS || process.env.GCLOUD_PROJECT) return \"vertex\";\n  if (process.env.OLLAMA_HOST) return \"ollama\";\n\n  try {\n    const host = process.env.OLLAMA_HOST ?? \"http://localhost:11434\";\n    const res = await fetch(`${host.replace(/\\/+$/, \"\")}/api/tags`, {\n      signal: AbortSignal.timeout(500),\n    });\n    if (res.ok) return \"ollama\";\n  } catch { /* not running */ }\n\n  return undefined;\n}\n"],"mappings":";AAWA,IAAM,cAAsC;AAAA,EAC1C,QAAQ;AAAA,EACR,WAAW;AAAA,EACX,QAAQ;AAAA,EACR,YAAY;AACd;AAEO,IAAM,iBAAyC;AAAA,EACpD,QAAQ;AAAA,EACR,WAAW;AAAA,EACX,QAAQ;AAAA,EACR,YAAY;AAAA,EACZ,SAAS;AAAA,EACT,QAAQ;AAAA,EACR,QAAQ;AACV;AAEO,IAAM,sBAAsC,OAAO,KAAK,cAAc;AAE7E,SAAS,kBAAkB,MAAkC;AAC3D,QAAM,SAAS,YAAY,IAAI;AAC/B,SAAO,SAAS,QAAQ,IAAI,MAAM,IAAI;AACxC;AAeA,eAAsB,gBACpB,OACA,MACqB;AACrB,MAAI,OAAO,UAAU,SAAU,QAAO;AAEtC,QAAM,OAAO;AACb,MAAI,CAAC,oBAAoB,SAAS,IAAI,GAAG;AACvC,UAAM,IAAI;AAAA,MACR,qBAAqB,IAAI,iBAAiB,oBAAoB,KAAK,IAAI,CAAC;AAAA,IAC1E;AAAA,EACF;AAEA,QAAM,MACJ,MAAM,UACN,kBAAkB,IAAI,KACtB,QAAQ,IAAI;AAEd,UAAQ,MAAM;AAAA,IACZ,KAAK,UAAU;AACb,UAAI,CAAC,IAAK,OAAM,IAAI,MAAM,gEAAgE;AAC1F,YAAM,EAAE,eAAe,IAAI,MAAM,OAAO,uBAAa;AACrD,aAAO,IAAI,eAAe,EAAE,QAAQ,KAAK,OAAO,MAAM,OAAO,SAAS,MAAM,QAAQ,CAAC;AAAA,IACvF;AAAA,IACA,KAAK,aAAa;AAChB,UAAI,CAAC,IAAK,OAAM,IAAI,MAAM,sEAAsE;AAChG,YAAM,EAAE,kBAAkB,IAAI,MAAM,OAAO,0BAAgB;AAC3D,aAAO,IAAI,kBAAkB,EAAE,QAAQ,KAAK,OAAO,MAAM,OAAO,SAAS,MAAM,SAAS,cAAc,EAAE,SAAS,KAAK,EAAE,CAAC;AAAA,IAC3H;AAAA,IACA,KAAK,UAAU;AACb,UAAI,CAAC,IAAK,OAAM,IAAI,MAAM,gEAAgE;AAC1F,YAAM,EAAE,eAAe,IAAI,MAAM,OAAO,uBAAa;AACrD,aAAO,IAAI,eAAe,EAAE,QAAQ,KAAK,OAAO,MAAM,OAAO,SAAS,MAAM,QAAQ,CAAC;AAAA,IACvF;AAAA,IACA,KAAK,cAAc;AACjB,UAAI,CAAC,IAAK,OAAM,IAAI,MAAM,wEAAwE;AAClG,YAAM,EAAE,mBAAmB,IAAI,MAAM,OAAO,2BAAiB;AAC7D,aAAO,IAAI,mBAAmB,EAAE,QAAQ,KAAK,OAAO,MAAM,OAAO,SAAS,SAAS,CAAC;AAAA,IACtF;AAAA,IACA,KAAK,WAAW;AACd,YAAM,EAAE,yBAAyB,IAAI,MAAM,OAAO,wBAAc;AAChE,aAAO,IAAI,yBAAyB,EAAE,OAAO,MAAM,OAAO,SAAS,MAAM,QAAQ,CAAC;AAAA,IACpF;AAAA,IACA,KAAK,UAAU;AACb,YAAM,EAAE,wBAAwB,IAAI,MAAM,OAAO,uBAAa;AAC9D,aAAO,IAAI,wBAAwB,EAAE,OAAO,MAAM,OAAO,SAAS,MAAM,QAAQ,CAAC;AAAA,IACnF;AAAA,IACA,KAAK,UAAU;AACb,YAAM,EAAE,eAAe,IAAI,MAAM,OAAO,uBAAa;AACrD,YAAM,OAAO,MAAM,YAAY,QAAQ,IAAI,cACvC,GAAG,QAAQ,IAAI,YAAY,QAAQ,QAAQ,EAAE,CAAC,QAC9C;AACJ,aAAO,IAAI,eAAe,EAAE,OAAO,MAAM,OAAO,SAAS,KAAK,CAAC;AAAA,IACjE;AAAA,IACA;AACE,YAAM,IAAI,MAAM,uBAAuB,IAAI,EAAE;AAAA,EACjD;AACF;AAKA,eAAsB,iBAAoD;AACxE,MAAI,QAAQ,IAAI,kBAAmB,QAAO;AAC1C,MAAI,QAAQ,IAAI,eAAgB,QAAO;AACvC,MAAI,QAAQ,IAAI,eAAgB,QAAO;AACvC,MAAI,QAAQ,IAAI,mBAAoB,QAAO;AAC3C,MAAI,QAAQ,IAAI,qBAAqB,QAAQ,IAAI,YAAa,QAAO;AACrE,MAAI,QAAQ,IAAI,kCAAkC,QAAQ,IAAI,eAAgB,QAAO;AACrF,MAAI,QAAQ,IAAI,YAAa,QAAO;AAEpC,MAAI;AACF,UAAM,OAAO,QAAQ,IAAI,eAAe;AACxC,UAAM,MAAM,MAAM,MAAM,GAAG,KAAK,QAAQ,QAAQ,EAAE,CAAC,aAAa;AAAA,MAC9D,QAAQ,YAAY,QAAQ,GAAG;AAAA,IACjC,CAAC;AACD,QAAI,IAAI,GAAI,QAAO;AAAA,EACrB,QAAQ;AAAA,EAAoB;AAE5B,SAAO;AACT;","names":[]}

package/dist/chunk-CCM2AXZG.js

@@ -1,16 +0,0 @@
-// src/providers/types.ts
-var ChatStreamError = class extends Error {
-  status;
-  retryAfter;
-  constructor(message, opts) {
-    super(message, { cause: opts?.cause });
-    this.name = "ChatStreamError";
-    this.status = opts?.status;
-    this.retryAfter = opts?.retryAfter;
-  }
-};
-
-export {
-  ChatStreamError
-};
-//# sourceMappingURL=chunk-CCM2AXZG.js.map

package/dist/chunk-CCM2AXZG.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/providers/types.ts"],"sourcesContent":["import type { ChatMessage } from \"../session/types.js\";\nimport type { ThinkingConfig } from \"../thinking/types.js\";\n\nexport interface ToolParameterProperty {\n  type: string;\n  description?: string;\n  enum?: string[];\n  default?: unknown;\n  minimum?: number;\n  maximum?: number;\n}\n\nexport interface ToolDefinition {\n  type: \"function\";\n  function: {\n    name: string;\n    description: string;\n    parameters: {\n      type: \"object\";\n      properties: Record<string, ToolParameterProperty>;\n      required?: string[];\n    };\n  };\n}\n\n// Streaming chunk types (OpenAI-compatible)\n\nexport interface ChatStreamDelta {\n  role?: \"assistant\";\n  content?: string | null;\n  thinking_content?: string | null;\n  thinking_signature?: string | null;\n  /** Opaque data payload for Anthropic redacted_thinking blocks. */\n  redacted_thinking_data?: string | null;\n  tool_calls?: Array<{\n    index: number;\n    id?: string;\n    type?: \"function\";\n    function?: {\n      name?: string;\n      arguments?: string;\n    };\n  }>;\n}\n\nexport interface ChatStreamChoice {\n  index: number;\n  delta: ChatStreamDelta;\n  finish_reason: string | null;\n}\n\nexport interface ChatStreamChunk {\n  id: string;\n  choices: ChatStreamChoice[];\n  model: string;\n  usage?: {\n    prompt_tokens: number;\n    completion_tokens: number;\n    total_tokens: number;\n    cache_read_tokens?: number;\n    cache_creation_tokens?: number;\n    thinking_tokens?: number;\n  };\n}\n\nexport interface ChatCompletionUsage {\n  prompt_tokens: number;\n  completion_tokens: number;\n  total_tokens: number;\n  cache_read_tokens?: number;\n  cache_creation_tokens?: number;\n  thinking_tokens?: number;\n}\n\n/**\n * Structured output format. When provided, the model is constrained to\n * produce a response matching the given JSON schema.\n *\n * - `json_schema`: the model must produce JSON conforming to the given schema.\n * - `json_object`: the model must produce valid JSON (no specific schema).\n */\nexport type OutputFormat = JsonSchemaOutputFormat | JsonObjectOutputFormat;\n\nexport interface JsonSchemaOutputFormat {\n  type: \"json_schema\";\n  /** JSON Schema object describing the expected output shape. */\n  schema: Record<string, unknown>;\n  /** Optional name for the schema (required by some providers). */\n  name?: string;\n  /** When true, the provider enforces strict schema adherence. */\n  strict?: boolean;\n}\n\nexport interface JsonObjectOutputFormat {\n  type: \"json_object\";\n}\n\nexport interface ChatParams {\n  model: string;\n  messages: ChatMessage[];\n  tools?: ToolDefinition[];\n  max_tokens?: number;\n  system?: string;\n  temperature?: number;\n  thinking?: ThinkingConfig;\n  /** Constrain the model to produce structured output matching this schema. */\n  outputFormat?: OutputFormat;\n  /**\n   * When true, the provider should place the cache breakpoint on the\n   * second-to-last message instead of the last. Used by subagent forks\n   * to avoid writing fork-only tails into the shared prompt cache.\n   */\n  skipCacheWrite?: boolean;\n  /** Abort signal — providers should forward this to cancel in-flight HTTP requests. */\n  signal?: AbortSignal;\n}\n\nexport interface AIProvider {\n  chat(params: ChatParams): AsyncIterable<ChatStreamChunk>;\n  /**\n   * Optional fallback model name, used when no `model` is provided at the\n   * Thread / Agent level. Consumers should pass explicit models when they\n   * can, but this lets each provider ship a sensible default without\n   * forcing Thread to hardcode provider-specific strings.\n   */\n  readonly defaultModel?: string;\n}\n\n/**\n * Extended error type that providers can throw to convey retry-relevant metadata.\n * Consumers (like the retry engine) can inspect these fields without knowing\n * provider-specific SDK error types.\n */\nexport class ChatStreamError extends Error {\n  status?: number;\n  retryAfter?: string;\n\n  constructor(\n    message: string,\n    opts?: { status?: number; retryAfter?: string; cause?: unknown },\n  ) {\n    super(message, { cause: opts?.cause });\n    this.name = \"ChatStreamError\";\n    this.status = opts?.status;\n    this.retryAfter = opts?.retryAfter;\n  }\n}\n"],"mappings":";AAqIO,IAAM,kBAAN,cAA8B,MAAM;AAAA,EACzC;AAAA,EACA;AAAA,EAEA,YACE,SACA,MACA;AACA,UAAM,SAAS,EAAE,OAAO,MAAM,MAAM,CAAC;AACrC,SAAK,OAAO;AACZ,SAAK,SAAS,MAAM;AACpB,SAAK,aAAa,MAAM;AAAA,EAC1B;AACF;","names":[]}

package/dist/chunk-DGUM43GV.js

@@ -1,11 +0,0 @@
-var __require = /* @__PURE__ */ ((x) => typeof require !== "undefined" ? require : typeof Proxy !== "undefined" ? new Proxy(x, {
-  get: (a, b) => (typeof require !== "undefined" ? require : a)[b]
-}) : x)(function(x) {
-  if (typeof require !== "undefined") return require.apply(this, arguments);
-  throw Error('Dynamic require of "' + x + '" is not supported');
-});
-
-export {
-  __require
-};
-//# sourceMappingURL=chunk-DGUM43GV.js.map

package/dist/chunk-HEQQQGK5.js

@@ -1,131 +0,0 @@
-// src/utils/context.ts
-var MODEL_CONTEXT_WINDOWS = {
-  // Anthropic (evergreen prefixes — also match dated variants via startsWith)
-  "claude-sonnet-4": 2e5,
-  "claude-opus-4": 2e5,
-  "claude-haiku-4": 2e5,
-  "claude-haiku-3-5": 2e5,
-  "claude-3-5-sonnet": 2e5,
-  "claude-3-5-haiku": 2e5,
-  // Bedrock / Vertex model ID patterns (prefix-matched)
-  "us.anthropic.claude": 2e5,
-  "eu.anthropic.claude": 2e5,
-  "ap.anthropic.claude": 2e5,
-  "anthropic.claude": 2e5,
-  // OpenAI
-  "gpt-4.1": 1047576,
-  "gpt-4.1-mini": 1047576,
-  "gpt-4.1-nano": 1047576,
-  "gpt-4o": 128e3,
-  "gpt-4o-mini": 128e3,
-  "gpt-4-turbo": 128e3,
-  "gpt-4": 8192,
-  "o1": 2e5,
-  "o1-mini": 128e3,
-  "o1-preview": 128e3,
-  "o3": 2e5,
-  "o3-mini": 2e5,
-  "o4-mini": 2e5,
-  // Google
-  "gemini-2.5-pro": 1048576,
-  "gemini-2.5-flash": 1048576,
-  "gemini-2.0-flash": 1048576,
-  "gemini-1.5-pro": 2097152,
-  "gemini-1.5-flash": 1048576
-};
-var MODEL_MAX_OUTPUT_TOKENS = {
-  "claude-sonnet-4-6": 128e3,
-  "claude-opus-4-6": 128e3,
-  "claude-sonnet-4": 64e3,
-  "claude-opus-4": 64e3,
-  "claude-haiku-4": 64e3,
-  "claude-haiku-3-5": 8192,
-  "claude-3-5-sonnet": 8192,
-  "claude-3-5-haiku": 8192,
-  "us.anthropic.claude": 64e3,
-  "eu.anthropic.claude": 64e3,
-  "ap.anthropic.claude": 64e3,
-  "anthropic.claude": 64e3,
-  "gpt-4.1": 32768,
-  "gpt-4.1-mini": 32768,
-  "gpt-4.1-nano": 32768,
-  "gpt-4o": 16384,
-  "gpt-4o-mini": 16384,
-  "gpt-4-turbo": 4096,
-  "o1": 1e5,
-  "o3": 1e5,
-  "o3-mini": 1e5,
-  "o4-mini": 1e5,
-  "gemini-2.5-pro": 65536,
-  "gemini-2.5-flash": 65536,
-  "gemini-2.0-flash": 8192
-};
-var DEFAULT_MAX_OUTPUT_TOKENS = 16384;
-var DEFAULT_CONTEXT_WINDOW = 128e3;
-var AUTOCOMPACT_BUFFER_TOKENS = 13e3;
-var MAX_OUTPUT_RESERVE = 2e4;
-var customWindows = {};
-function registerContextWindows(windows) {
-  customWindows = { ...customWindows, ...windows };
-}
-function getContextWindowForModel(model) {
-  if (customWindows[model] !== void 0) return customWindows[model];
-  if (MODEL_CONTEXT_WINDOWS[model] !== void 0)
-    return MODEL_CONTEXT_WINDOWS[model];
-  for (const [prefix, size] of Object.entries(MODEL_CONTEXT_WINDOWS)) {
-    if (model.startsWith(prefix)) return size;
-  }
-  for (const [prefix, size] of Object.entries(customWindows)) {
-    if (model.startsWith(prefix)) return size;
-  }
-  return DEFAULT_CONTEXT_WINDOW;
-}
-function getEffectiveContextWindow(model, maxOutputTokens) {
-  const window = getContextWindowForModel(model);
-  const reserve = Math.min(maxOutputTokens ?? MAX_OUTPUT_RESERVE, MAX_OUTPUT_RESERVE);
-  return window - reserve;
-}
-function getAutoCompactThreshold(model, maxOutputTokens) {
-  return getEffectiveContextWindow(model, maxOutputTokens) - AUTOCOMPACT_BUFFER_TOKENS;
-}
-function getMaxOutputTokensForModel(model) {
-  if (MODEL_MAX_OUTPUT_TOKENS[model] !== void 0)
-    return MODEL_MAX_OUTPUT_TOKENS[model];
-  for (const [prefix, size] of Object.entries(MODEL_MAX_OUTPUT_TOKENS)) {
-    if (model.startsWith(prefix)) return size;
-  }
-  return DEFAULT_MAX_OUTPUT_TOKENS;
-}
-
-// src/providers/cache.ts
-function sortToolDefinitionsForCache(tools, mcpToolNames) {
-  const byName = (a, b) => a.function.name.localeCompare(b.function.name);
-  if (!mcpToolNames || mcpToolNames.size === 0) {
-    return [...tools].sort(byName);
-  }
-  const builtIn = [];
-  const mcp = [];
-  for (const t of tools) {
-    if (mcpToolNames.has(t.function.name)) {
-      mcp.push(t);
-    } else {
-      builtIn.push(t);
-    }
-  }
-  return [...builtIn.sort(byName), ...mcp.sort(byName)];
-}
-function getMessageCacheBreakpointIndex(messages, skipCacheWrite) {
-  if (messages.length === 0) return -1;
-  return skipCacheWrite && messages.length >= 2 ? messages.length - 2 : messages.length - 1;
-}
-
-export {
-  registerContextWindows,
-  getContextWindowForModel,
-  getEffectiveContextWindow,
-  getAutoCompactThreshold,
-  getMaxOutputTokensForModel,
-  sortToolDefinitionsForCache,
-  getMessageCacheBreakpointIndex
-};
-//# sourceMappingURL=chunk-HEQQQGK5.js.map

package/dist/chunk-HEQQQGK5.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/utils/context.ts","../src/providers/cache.ts"],"sourcesContent":["/**\n * Model context window sizes and effective window calculations.\n */\n\nconst MODEL_CONTEXT_WINDOWS: Record<string, number> = {\n  // Anthropic (evergreen prefixes — also match dated variants via startsWith)\n  \"claude-sonnet-4\": 200_000,\n  \"claude-opus-4\": 200_000,\n  \"claude-haiku-4\": 200_000,\n  \"claude-haiku-3-5\": 200_000,\n  \"claude-3-5-sonnet\": 200_000,\n  \"claude-3-5-haiku\": 200_000,\n  // Bedrock / Vertex model ID patterns (prefix-matched)\n  \"us.anthropic.claude\": 200_000,\n  \"eu.anthropic.claude\": 200_000,\n  \"ap.anthropic.claude\": 200_000,\n  \"anthropic.claude\": 200_000,\n  // OpenAI\n  \"gpt-4.1\": 1_047_576,\n  \"gpt-4.1-mini\": 1_047_576,\n  \"gpt-4.1-nano\": 1_047_576,\n  \"gpt-4o\": 128_000,\n  \"gpt-4o-mini\": 128_000,\n  \"gpt-4-turbo\": 128_000,\n  \"gpt-4\": 8_192,\n  \"o1\": 200_000,\n  \"o1-mini\": 128_000,\n  \"o1-preview\": 128_000,\n  \"o3\": 200_000,\n  \"o3-mini\": 200_000,\n  \"o4-mini\": 200_000,\n  // Google\n  \"gemini-2.5-pro\": 1_048_576,\n  \"gemini-2.5-flash\": 1_048_576,\n  \"gemini-2.0-flash\": 1_048_576,\n  \"gemini-1.5-pro\": 2_097_152,\n  \"gemini-1.5-flash\": 1_048_576,\n};\n\nconst MODEL_MAX_OUTPUT_TOKENS: Record<string, number> = {\n  \"claude-sonnet-4-6\": 128_000,\n  \"claude-opus-4-6\": 128_000,\n  \"claude-sonnet-4\": 64_000,\n  \"claude-opus-4\": 64_000,\n  \"claude-haiku-4\": 64_000,\n  \"claude-haiku-3-5\": 8_192,\n  \"claude-3-5-sonnet\": 8_192,\n  \"claude-3-5-haiku\": 8_192,\n  \"us.anthropic.claude\": 64_000,\n  \"eu.anthropic.claude\": 64_000,\n  \"ap.anthropic.claude\": 64_000,\n  \"anthropic.claude\": 64_000,\n  \"gpt-4.1\": 32_768,\n  \"gpt-4.1-mini\": 32_768,\n  \"gpt-4.1-nano\": 32_768,\n  \"gpt-4o\": 16_384,\n  \"gpt-4o-mini\": 16_384,\n  \"gpt-4-turbo\": 4_096,\n  \"o1\": 100_000,\n  \"o3\": 100_000,\n  \"o3-mini\": 100_000,\n  \"o4-mini\": 100_000,\n  \"gemini-2.5-pro\": 65_536,\n  \"gemini-2.5-flash\": 65_536,\n  \"gemini-2.0-flash\": 8_192,\n};\n\nconst DEFAULT_MAX_OUTPUT_TOKENS = 16_384;\nconst DEFAULT_CONTEXT_WINDOW = 128_000;\nconst AUTOCOMPACT_BUFFER_TOKENS = 13_000;\nconst MAX_OUTPUT_RESERVE = 20_000;\n\nlet customWindows: Record<string, number> = {};\n\n/**\n * Register custom context window sizes for models not in the built-in table.\n */\nexport function registerContextWindows(\n  windows: Record<string, number>,\n): void {\n  customWindows = { ...customWindows, ...windows };\n}\n\n/**\n * Get the context window size for a model. Checks custom overrides first,\n * then built-in table, then prefix-matches, then falls back to default.\n */\nexport function getContextWindowForModel(model: string): number {\n  if (customWindows[model] !== undefined) return customWindows[model];\n  if (MODEL_CONTEXT_WINDOWS[model] !== undefined)\n    return MODEL_CONTEXT_WINDOWS[model];\n\n  for (const [prefix, size] of Object.entries(MODEL_CONTEXT_WINDOWS)) {\n    if (model.startsWith(prefix)) return size;\n  }\n  for (const [prefix, size] of Object.entries(customWindows)) {\n    if (model.startsWith(prefix)) return size;\n  }\n\n  return DEFAULT_CONTEXT_WINDOW;\n}\n\n/**\n * Effective context window = total window minus space reserved for the\n * model's output during a compaction/summary request.\n */\nexport function getEffectiveContextWindow(\n  model: string,\n  maxOutputTokens?: number,\n): number {\n  const window = getContextWindowForModel(model);\n  const reserve = Math.min(maxOutputTokens ?? MAX_OUTPUT_RESERVE, MAX_OUTPUT_RESERVE);\n  return window - reserve;\n}\n\n/**\n * Auto-compact threshold: effective window minus a buffer to ensure we\n * compact before we're at the hard limit.\n */\nexport function getAutoCompactThreshold(\n  model: string,\n  maxOutputTokens?: number,\n): number {\n  return getEffectiveContextWindow(model, maxOutputTokens) - AUTOCOMPACT_BUFFER_TOKENS;\n}\n\n/**\n * Get the maximum output tokens a model supports. Used to clamp\n * max_tokens when extended thinking budgets are added.\n */\nexport function getMaxOutputTokensForModel(model: string): number {\n  if (MODEL_MAX_OUTPUT_TOKENS[model] !== undefined)\n    return MODEL_MAX_OUTPUT_TOKENS[model];\n  for (const [prefix, size] of Object.entries(MODEL_MAX_OUTPUT_TOKENS)) {\n    if (model.startsWith(prefix)) return size;\n  }\n  return DEFAULT_MAX_OUTPUT_TOKENS;\n}\n","/**\n * Provider-agnostic prompt caching utilities.\n *\n * Stable tool ordering prevents cache invalidation when the tool set is\n * unchanged. The breakpoint index helper determines which message gets a\n * single cache_control marker per request (matching claude-code's strategy).\n */\n\nimport type { ToolDefinition } from \"./types.js\";\nimport type { ChatMessage } from \"../session/types.js\";\n\nexport type CacheScope = \"global\" | \"org\";\n\nexport interface CacheControlConfig {\n  enabled: boolean;\n  /** TTL for cached content. When set, produces `ttl: '1h'` in cache_control. */\n  ttl?: \"1h\";\n  /** Scope for shared cache across sessions/orgs. */\n  scope?: CacheScope;\n}\n\n/**\n * Sort tool definitions deterministically for prompt cache stability.\n *\n * Strategy (matching claude-code's assembleToolPool): built-in tools form a\n * contiguous prefix sorted by name, followed by MCP/external tools sorted by\n * name. Tools with `mcpInfo` on the original Tool object are treated as MCP;\n * everything else is built-in. Since ToolDefinition doesn't carry mcpInfo,\n * callers can pass an optional set of MCP tool names to partition correctly.\n */\nexport function sortToolDefinitionsForCache(\n  tools: ToolDefinition[],\n  mcpToolNames?: ReadonlySet<string>,\n): ToolDefinition[] {\n  const byName = (a: ToolDefinition, b: ToolDefinition) =>\n    a.function.name.localeCompare(b.function.name);\n\n  if (!mcpToolNames || mcpToolNames.size === 0) {\n    return [...tools].sort(byName);\n  }\n\n  const builtIn: ToolDefinition[] = [];\n  const mcp: ToolDefinition[] = [];\n\n  for (const t of tools) {\n    if (mcpToolNames.has(t.function.name)) {\n      mcp.push(t);\n    } else {\n      builtIn.push(t);\n    }\n  }\n\n  return [...builtIn.sort(byName), ...mcp.sort(byName)];\n}\n\n/**\n * Determine which message index should receive the cache_control breakpoint.\n *\n * Exactly one message per request is marked. Normally the last message;\n * for forked agents with skipCacheWrite the second-to-last so the fork\n * doesn't write its own tail into the cache.\n */\nexport function getMessageCacheBreakpointIndex(\n  messages: ChatMessage[],\n  skipCacheWrite?: boolean,\n): number {\n  if (messages.length === 0) return -1;\n  return skipCacheWrite && messages.length >= 2\n    ? messages.length - 2\n    : messages.length - 1;\n}\n"],"mappings":";AAIA,IAAM,wBAAgD;AAAA;AAAA,EAEpD,mBAAmB;AAAA,EACnB,iBAAiB;AAAA,EACjB,kBAAkB;AAAA,EAClB,oBAAoB;AAAA,EACpB,qBAAqB;AAAA,EACrB,oBAAoB;AAAA;AAAA,EAEpB,uBAAuB;AAAA,EACvB,uBAAuB;AAAA,EACvB,uBAAuB;AAAA,EACvB,oBAAoB;AAAA;AAAA,EAEpB,WAAW;AAAA,EACX,gBAAgB;AAAA,EAChB,gBAAgB;AAAA,EAChB,UAAU;AAAA,EACV,eAAe;AAAA,EACf,eAAe;AAAA,EACf,SAAS;AAAA,EACT,MAAM;AAAA,EACN,WAAW;AAAA,EACX,cAAc;AAAA,EACd,MAAM;AAAA,EACN,WAAW;AAAA,EACX,WAAW;AAAA;AAAA,EAEX,kBAAkB;AAAA,EAClB,oBAAoB;AAAA,EACpB,oBAAoB;AAAA,EACpB,kBAAkB;AAAA,EAClB,oBAAoB;AACtB;AAEA,IAAM,0BAAkD;AAAA,EACtD,qBAAqB;AAAA,EACrB,mBAAmB;AAAA,EACnB,mBAAmB;AAAA,EACnB,iBAAiB;AAAA,EACjB,kBAAkB;AAAA,EAClB,oBAAoB;AAAA,EACpB,qBAAqB;AAAA,EACrB,oBAAoB;AAAA,EACpB,uBAAuB;AAAA,EACvB,uBAAuB;AAAA,EACvB,uBAAuB;AAAA,EACvB,oBAAoB;AAAA,EACpB,WAAW;AAAA,EACX,gBAAgB;AAAA,EAChB,gBAAgB;AAAA,EAChB,UAAU;AAAA,EACV,eAAe;AAAA,EACf,eAAe;AAAA,EACf,MAAM;AAAA,EACN,MAAM;AAAA,EACN,WAAW;AAAA,EACX,WAAW;AAAA,EACX,kBAAkB;AAAA,EAClB,oBAAoB;AAAA,EACpB,oBAAoB;AACtB;AAEA,IAAM,4BAA4B;AAClC,IAAM,yBAAyB;AAC/B,IAAM,4BAA4B;AAClC,IAAM,qBAAqB;AAE3B,IAAI,gBAAwC,CAAC;AAKtC,SAAS,uBACd,SACM;AACN,kBAAgB,EAAE,GAAG,eAAe,GAAG,QAAQ;AACjD;AAMO,SAAS,yBAAyB,OAAuB;AAC9D,MAAI,cAAc,KAAK,MAAM,OAAW,QAAO,cAAc,KAAK;AAClE,MAAI,sBAAsB,KAAK,MAAM;AACnC,WAAO,sBAAsB,KAAK;AAEpC,aAAW,CAAC,QAAQ,IAAI,KAAK,OAAO,QAAQ,qBAAqB,GAAG;AAClE,QAAI,MAAM,WAAW,MAAM,EAAG,QAAO;AAAA,EACvC;AACA,aAAW,CAAC,QAAQ,IAAI,KAAK,OAAO,QAAQ,aAAa,GAAG;AAC1D,QAAI,MAAM,WAAW,MAAM,EAAG,QAAO;AAAA,EACvC;AAEA,SAAO;AACT;AAMO,SAAS,0BACd,OACA,iBACQ;AACR,QAAM,SAAS,yBAAyB,KAAK;AAC7C,QAAM,UAAU,KAAK,IAAI,mBAAmB,oBAAoB,kBAAkB;AAClF,SAAO,SAAS;AAClB;AAMO,SAAS,wBACd,OACA,iBACQ;AACR,SAAO,0BAA0B,OAAO,eAAe,IAAI;AAC7D;AAMO,SAAS,2BAA2B,OAAuB;AAChE,MAAI,wBAAwB,KAAK,MAAM;AACrC,WAAO,wBAAwB,KAAK;AACtC,aAAW,CAAC,QAAQ,IAAI,KAAK,OAAO,QAAQ,uBAAuB,GAAG;AACpE,QAAI,MAAM,WAAW,MAAM,EAAG,QAAO;AAAA,EACvC;AACA,SAAO;AACT;;;AC3GO,SAAS,4BACd,OACA,cACkB;AAClB,QAAM,SAAS,CAAC,GAAmB,MACjC,EAAE,SAAS,KAAK,cAAc,EAAE,SAAS,IAAI;AAE/C,MAAI,CAAC,gBAAgB,aAAa,SAAS,GAAG;AAC5C,WAAO,CAAC,GAAG,KAAK,EAAE,KAAK,MAAM;AAAA,EAC/B;AAEA,QAAM,UAA4B,CAAC;AACnC,QAAM,MAAwB,CAAC;AAE/B,aAAW,KAAK,OAAO;AACrB,QAAI,aAAa,IAAI,EAAE,SAAS,IAAI,GAAG;AACrC,UAAI,KAAK,CAAC;AAAA,IACZ,OAAO;AACL,cAAQ,KAAK,CAAC;AAAA,IAChB;AAAA,EACF;AAEA,SAAO,CAAC,GAAG,QAAQ,KAAK,MAAM,GAAG,GAAG,IAAI,KAAK,MAAM,CAAC;AACtD;AASO,SAAS,+BACd,UACA,gBACQ;AACR,MAAI,SAAS,WAAW,EAAG,QAAO;AAClC,SAAO,kBAAkB,SAAS,UAAU,IACxC,SAAS,SAAS,IAClB,SAAS,SAAS;AACxB;","names":[]}

package/dist/chunk-I3JTUFPK.js

@@ -1,171 +0,0 @@
-import {
-  ChatStreamError
-} from "./chunk-CCM2AXZG.js";
-
-// src/providers/openai.ts
-import OpenAI from "openai";
-var O_SERIES_PATTERN = /^o[1-9]/;
-var OpenAIProvider = class _OpenAIProvider {
-  client;
-  defaultModel;
-  compatMode;
-  constructor(opts) {
-    this.client = new OpenAI({
-      apiKey: opts.apiKey ?? "not-needed",
-      baseURL: opts.baseURL,
-      defaultHeaders: opts.defaultHeaders,
-      maxRetries: 0
-    });
-    this.defaultModel = opts.model ?? "gpt-5.4";
-    this.compatMode = opts.compatMode ?? false;
-  }
-  async *chat(params) {
-    const messages = this.buildMessages(params.system, params.messages);
-    const model = params.model ?? this.defaultModel;
-    const isOSeries = O_SERIES_PATTERN.test(model);
-    const createParams = {
-      model,
-      messages,
-      tools: params.tools?.map((t) => ({
-        type: "function",
-        function: t.function
-      })),
-      stream: true,
-      ...this.compatMode ? {} : { stream_options: { include_usage: true } }
-    };
-    if (isOSeries) {
-      createParams.max_completion_tokens = params.max_tokens ?? 16384;
-      if (params.thinking?.type === "enabled") {
-        createParams.reasoning_effort = "high";
-      }
-    } else {
-      createParams.max_tokens = params.max_tokens;
-      createParams.temperature = params.temperature;
-    }
-    if (params.outputFormat?.type === "json_schema") {
-      createParams.response_format = {
-        type: "json_schema",
-        json_schema: {
-          name: params.outputFormat.name ?? "response",
-          schema: params.outputFormat.schema,
-          strict: params.outputFormat.strict ?? false
-        }
-      };
-    } else if (params.outputFormat?.type === "json_object") {
-      createParams.response_format = {
-        type: "json_object"
-      };
-    }
-    try {
-      const stream = await this.client.chat.completions.create(createParams, {
-        ...params.signal ? { signal: params.signal } : {}
-      });
-      for await (const chunk of stream) {
-        const usage = chunk.usage;
-        let mappedUsage;
-        if (usage) {
-          const u = usage;
-          const promptDetails = u.prompt_tokens_details;
-          const completionDetails = u.completion_tokens_details;
-          mappedUsage = {
-            prompt_tokens: usage.prompt_tokens,
-            completion_tokens: usage.completion_tokens,
-            total_tokens: usage.total_tokens,
-            cache_read_tokens: promptDetails?.cached_tokens,
-            cache_creation_tokens: promptDetails?.cache_creation_tokens,
-            thinking_tokens: completionDetails?.reasoning_tokens
-          };
-        }
-        const choices = chunk.choices ?? [];
-        if (choices.length === 0 && mappedUsage) {
-          yield { id: chunk.id, model: chunk.model, choices: [], usage: mappedUsage };
-          continue;
-        }
-        yield {
-          id: chunk.id,
-          model: chunk.model,
-          choices: choices.map((c) => ({
-            index: c.index,
-            delta: {
-              role: c.delta.role,
-              content: c.delta.content,
-              thinking_content: c.delta.reasoning_content,
-              tool_calls: c.delta.tool_calls?.map((tc) => ({
-                index: tc.index,
-                id: tc.id,
-                type: tc.type,
-                function: tc.function ? {
-                  name: tc.function.name,
-                  arguments: tc.function.arguments
-                } : void 0
-              }))
-            },
-            finish_reason: c.finish_reason
-          })),
-          usage: mappedUsage
-        };
-      }
-    } catch (err) {
-      if (err instanceof ChatStreamError) throw err;
-      const apiErr = err;
-      throw new ChatStreamError(
-        err instanceof Error ? err.message : String(err),
-        {
-          status: apiErr.status,
-          retryAfter: apiErr.headers?.get?.("retry-after") ?? apiErr.headers?.["retry-after"] ?? void 0,
-          cause: err
-        }
-      );
-    }
-  }
-  static contentPartsToOpenAI(parts) {
-    return parts.map((part) => {
-      if (part.type === "text") {
-        return { type: "text", text: part.text };
-      }
-      if (part.type === "image") {
-        return {
-          type: "image_url",
-          image_url: { url: `data:${part.media_type};base64,${part.data}` }
-        };
-      }
-      return { type: "image_url", image_url: { url: part.url } };
-    });
-  }
-  buildMessages(system, messages) {
-    const result = [];
-    if (system) {
-      result.push({ role: "system", content: system });
-    }
-    for (const msg of messages) {
-      if (msg.role === "tool") {
-        const content = Array.isArray(msg.content) ? _OpenAIProvider.contentPartsToOpenAI(msg.content) : msg.content;
-        result.push({
-          role: "tool",
-          tool_call_id: msg.tool_call_id,
-          content
-        });
-      } else if (msg.role === "assistant") {
-        const entry = {
-          role: "assistant",
-          content: msg.content
-        };
-        if (msg.tool_calls) {
-          entry.tool_calls = msg.tool_calls;
-        }
-        result.push(entry);
-      } else if (msg.role === "user") {
-        const content = Array.isArray(msg.content) ? _OpenAIProvider.contentPartsToOpenAI(msg.content) : msg.content;
-        result.push({ role: "user", content });
-      } else {
-        result.push({ role: msg.role, content: msg.content });
-      }
-    }
-    return result;
-  }
-};
-
-export {
-  OpenAIProvider
-};
-//# sourceMappingURL=chunk-I3JTUFPK.js.map

package/dist/chunk-I3JTUFPK.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/providers/openai.ts"],"sourcesContent":["import OpenAI from \"openai\";\nimport type {\n  AIProvider,\n  ChatParams,\n  ChatStreamChunk,\n} from \"./types.js\";\nimport { ChatStreamError } from \"./types.js\";\nimport type { ChatMessage, ContentPart } from \"../session/types.js\";\n\nexport interface OpenAIProviderOptions {\n  apiKey?: string;\n  baseURL?: string;\n  model?: string;\n  defaultHeaders?: Record<string, string | undefined>;\n  /** When true, omits `stream_options` that some OpenAI-compatible endpoints don't support. */\n  compatMode?: boolean;\n}\n\nconst O_SERIES_PATTERN = /^o[1-9]/;\n\nexport class OpenAIProvider implements AIProvider {\n  private client: OpenAI;\n  readonly defaultModel: string;\n  protected compatMode: boolean;\n\n  constructor(opts: OpenAIProviderOptions) {\n    this.client = new OpenAI({\n      apiKey: opts.apiKey ?? \"not-needed\",\n      baseURL: opts.baseURL,\n      defaultHeaders: opts.defaultHeaders,\n      maxRetries: 0,\n    });\n    this.defaultModel = opts.model ?? \"gpt-5.4\";\n    this.compatMode = opts.compatMode ?? false;\n  }\n\n  async *chat(params: ChatParams): AsyncIterable<ChatStreamChunk> {\n    const messages = this.buildMessages(params.system, params.messages);\n    const model = params.model ?? this.defaultModel;\n    const isOSeries = O_SERIES_PATTERN.test(model);\n\n    const createParams: OpenAI.ChatCompletionCreateParamsStreaming = {\n      model,\n      messages: messages as unknown as OpenAI.ChatCompletionMessageParam[],\n      tools: params.tools?.map((t) => ({\n        type: \"function\" as const,\n        function: t.function,\n      })),\n      stream: true,\n      ...(this.compatMode ? {} : { stream_options: { include_usage: true } }),\n    };\n\n    if (isOSeries) {\n      (createParams as unknown as Record<string, unknown>).max_completion_tokens = params.max_tokens ?? 16384;\n      if (params.thinking?.type === \"enabled\") {\n        (createParams as unknown as Record<string, unknown>).reasoning_effort = \"high\";\n      }\n    } else {\n      createParams.max_tokens = params.max_tokens;\n      createParams.temperature = params.temperature;\n    }\n\n    if (params.outputFormat?.type === \"json_schema\") {\n      (createParams as unknown as Record<string, unknown>).response_format = {\n        type: \"json_schema\",\n        json_schema: {\n          name: params.outputFormat.name ?? \"response\",\n          schema: params.outputFormat.schema,\n          strict: params.outputFormat.strict ?? false,\n        },\n      };\n    } else if (params.outputFormat?.type === \"json_object\") {\n      (createParams as unknown as Record<string, unknown>).response_format = {\n        type: \"json_object\",\n      };\n    }\n\n    try {\n    const stream = await this.client.chat.completions.create(createParams, {\n      ...(params.signal ? { signal: params.signal } : {}),\n    });\n\n    for await (const chunk of stream) {\n      const usage = chunk.usage;\n      let mappedUsage: ChatStreamChunk[\"usage\"] | undefined;\n      if (usage) {\n        const u = usage as unknown as Record<string, unknown>;\n        const promptDetails = u.prompt_tokens_details as Record<string, unknown> | undefined;\n        const completionDetails = u.completion_tokens_details as Record<string, unknown> | undefined;\n        mappedUsage = {\n          prompt_tokens: usage.prompt_tokens,\n          completion_tokens: usage.completion_tokens,\n          total_tokens: usage.total_tokens,\n          cache_read_tokens: promptDetails?.cached_tokens as number | undefined,\n          cache_creation_tokens: promptDetails?.cache_creation_tokens as number | undefined,\n          thinking_tokens: completionDetails?.reasoning_tokens as number | undefined,\n        };\n      }\n\n      const choices = chunk.choices ?? [];\n      if (choices.length === 0 && mappedUsage) {\n        yield { id: chunk.id, model: chunk.model, choices: [], usage: mappedUsage };\n        continue;\n      }\n\n      yield {\n        id: chunk.id,\n        model: chunk.model,\n        choices: choices.map((c) => ({\n          index: c.index,\n          delta: {\n            role: c.delta.role as \"assistant\" | undefined,\n            content: c.delta.content,\n            thinking_content: (c.delta as Record<string, unknown>).reasoning_content as string | undefined,\n            tool_calls: c.delta.tool_calls?.map((tc) => ({\n              index: tc.index,\n              id: tc.id,\n              type: tc.type as \"function\" | undefined,\n              function: tc.function\n                ? {\n                    name: tc.function.name,\n                    arguments: tc.function.arguments,\n                  }\n                : undefined,\n            })),\n          },\n          finish_reason: c.finish_reason,\n        })),\n        usage: mappedUsage,\n      };\n    }\n    } catch (err: unknown) {\n      if (err instanceof ChatStreamError) throw err;\n      const apiErr = err as { status?: number; headers?: Record<string, string> & { get?(k: string): string | null } };\n      throw new ChatStreamError(\n        err instanceof Error ? err.message : String(err),\n        {\n          status: apiErr.status,\n          retryAfter: apiErr.headers?.get?.(\"retry-after\") ?? apiErr.headers?.[\"retry-after\"] ?? undefined,\n          cause: err,\n        },\n      );\n    }\n  }\n\n  private static contentPartsToOpenAI(\n    parts: ContentPart[],\n  ): Array<Record<string, unknown>> {\n    return parts.map((part) => {\n      if (part.type === \"text\") {\n        return { type: \"text\", text: part.text };\n      }\n      if (part.type === \"image\") {\n        return {\n          type: \"image_url\",\n          image_url: { url: `data:${part.media_type};base64,${part.data}` },\n        };\n      }\n      // image_url\n      return { type: \"image_url\", image_url: { url: part.url } };\n    });\n  }\n\n  private buildMessages(\n    system: string | undefined,\n    messages: ChatMessage[],\n  ): Array<Record<string, unknown>> {\n    const result: Array<Record<string, unknown>> = [];\n    if (system) {\n      result.push({ role: \"system\", content: system });\n    }\n    for (const msg of messages) {\n      if (msg.role === \"tool\") {\n        const content = Array.isArray(msg.content)\n          ? OpenAIProvider.contentPartsToOpenAI(msg.content as ContentPart[])\n          : msg.content;\n        result.push({\n          role: \"tool\",\n          tool_call_id: msg.tool_call_id,\n          content,\n        });\n      } else if (msg.role === \"assistant\") {\n        const entry: Record<string, unknown> = {\n          role: \"assistant\",\n          content: msg.content,\n        };\n        if (msg.tool_calls) {\n          entry.tool_calls = msg.tool_calls;\n        }\n        result.push(entry);\n      } else if (msg.role === \"user\") {\n        const content = Array.isArray(msg.content)\n          ? OpenAIProvider.contentPartsToOpenAI(msg.content as ContentPart[])\n          : msg.content;\n        result.push({ role: \"user\", content });\n      } else {\n        result.push({ role: msg.role, content: msg.content });\n      }\n    }\n    return result;\n  }\n}\n"],"mappings":";;;;;AAAA,OAAO,YAAY;AAkBnB,IAAM,mBAAmB;AAElB,IAAM,iBAAN,MAAM,gBAAqC;AAAA,EACxC;AAAA,EACC;AAAA,EACC;AAAA,EAEV,YAAY,MAA6B;AACvC,SAAK,SAAS,IAAI,OAAO;AAAA,MACvB,QAAQ,KAAK,UAAU;AAAA,MACvB,SAAS,KAAK;AAAA,MACd,gBAAgB,KAAK;AAAA,MACrB,YAAY;AAAA,IACd,CAAC;AACD,SAAK,eAAe,KAAK,SAAS;AAClC,SAAK,aAAa,KAAK,cAAc;AAAA,EACvC;AAAA,EAEA,OAAO,KAAK,QAAoD;AAC9D,UAAM,WAAW,KAAK,cAAc,OAAO,QAAQ,OAAO,QAAQ;AAClE,UAAM,QAAQ,OAAO,SAAS,KAAK;AACnC,UAAM,YAAY,iBAAiB,KAAK,KAAK;AAE7C,UAAM,eAA2D;AAAA,MAC/D;AAAA,MACA;AAAA,MACA,OAAO,OAAO,OAAO,IAAI,CAAC,OAAO;AAAA,QAC/B,MAAM;AAAA,QACN,UAAU,EAAE;AAAA,MACd,EAAE;AAAA,MACF,QAAQ;AAAA,MACR,GAAI,KAAK,aAAa,CAAC,IAAI,EAAE,gBAAgB,EAAE,eAAe,KAAK,EAAE;AAAA,IACvE;AAEA,QAAI,WAAW;AACb,MAAC,aAAoD,wBAAwB,OAAO,cAAc;AAClG,UAAI,OAAO,UAAU,SAAS,WAAW;AACvC,QAAC,aAAoD,mBAAmB;AAAA,MAC1E;AAAA,IACF,OAAO;AACL,mBAAa,aAAa,OAAO;AACjC,mBAAa,cAAc,OAAO;AAAA,IACpC;AAEA,QAAI,OAAO,cAAc,SAAS,eAAe;AAC/C,MAAC,aAAoD,kBAAkB;AAAA,QACrE,MAAM;AAAA,QACN,aAAa;AAAA,UACX,MAAM,OAAO,aAAa,QAAQ;AAAA,UAClC,QAAQ,OAAO,aAAa;AAAA,UAC5B,QAAQ,OAAO,aAAa,UAAU;AAAA,QACxC;AAAA,MACF;AAAA,IACF,WAAW,OAAO,cAAc,SAAS,eAAe;AACtD,MAAC,aAAoD,kBAAkB;AAAA,QACrE,MAAM;AAAA,MACR;AAAA,IACF;AAEA,QAAI;AACJ,YAAM,SAAS,MAAM,KAAK,OAAO,KAAK,YAAY,OAAO,cAAc;AAAA,QACrE,GAAI,OAAO,SAAS,EAAE,QAAQ,OAAO,OAAO,IAAI,CAAC;AAAA,MACnD,CAAC;AAED,uBAAiB,SAAS,QAAQ;AAChC,cAAM,QAAQ,MAAM;AACpB,YAAI;AACJ,YAAI,OAAO;AACT,gBAAM,IAAI;AACV,gBAAM,gBAAgB,EAAE;AACxB,gBAAM,oBAAoB,EAAE;AAC5B,wBAAc;AAAA,YACZ,eAAe,MAAM;AAAA,YACrB,mBAAmB,MAAM;AAAA,YACzB,cAAc,MAAM;AAAA,YACpB,mBAAmB,eAAe;AAAA,YAClC,uBAAuB,eAAe;AAAA,YACtC,iBAAiB,mBAAmB;AAAA,UACtC;AAAA,QACF;AAEA,cAAM,UAAU,MAAM,WAAW,CAAC;AAClC,YAAI,QAAQ,WAAW,KAAK,aAAa;AACvC,gBAAM,EAAE,IAAI,MAAM,IAAI,OAAO,MAAM,OAAO,SAAS,CAAC,GAAG,OAAO,YAAY;AAC1E;AAAA,QACF;AAEA,cAAM;AAAA,UACJ,IAAI,MAAM;AAAA,UACV,OAAO,MAAM;AAAA,UACb,SAAS,QAAQ,IAAI,CAAC,OAAO;AAAA,YAC3B,OAAO,EAAE;AAAA,YACT,OAAO;AAAA,cACL,MAAM,EAAE,MAAM;AAAA,cACd,SAAS,EAAE,MAAM;AAAA,cACjB,kBAAmB,EAAE,MAAkC;AAAA,cACvD,YAAY,EAAE,MAAM,YAAY,IAAI,CAAC,QAAQ;AAAA,gBAC3C,OAAO,GAAG;AAAA,gBACV,IAAI,GAAG;AAAA,gBACP,MAAM,GAAG;AAAA,gBACT,UAAU,GAAG,WACT;AAAA,kBACE,MAAM,GAAG,SAAS;AAAA,kBAClB,WAAW,GAAG,SAAS;AAAA,gBACzB,IACA;AAAA,cACN,EAAE;AAAA,YACJ;AAAA,YACA,eAAe,EAAE;AAAA,UACnB,EAAE;AAAA,UACF,OAAO;AAAA,QACT;AAAA,MACF;AAAA,IACA,SAAS,KAAc;AACrB,UAAI,eAAe,gBAAiB,OAAM;AAC1C,YAAM,SAAS;AACf,YAAM,IAAI;AAAA,QACR,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAAA,QAC/C;AAAA,UACE,QAAQ,OAAO;AAAA,UACf,YAAY,OAAO,SAAS,MAAM,aAAa,KAAK,OAAO,UAAU,aAAa,KAAK;AAAA,UACvF,OAAO;AAAA,QACT;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAAA,EAEA,OAAe,qBACb,OACgC;AAChC,WAAO,MAAM,IAAI,CAAC,SAAS;AACzB,UAAI,KAAK,SAAS,QAAQ;AACxB,eAAO,EAAE,MAAM,QAAQ,MAAM,KAAK,KAAK;AAAA,MACzC;AACA,UAAI,KAAK,SAAS,SAAS;AACzB,eAAO;AAAA,UACL,MAAM;AAAA,UACN,WAAW,EAAE,KAAK,QAAQ,KAAK,UAAU,WAAW,KAAK,IAAI,GAAG;AAAA,QAClE;AAAA,MACF;AAEA,aAAO,EAAE,MAAM,aAAa,WAAW,EAAE,KAAK,KAAK,IAAI,EAAE;AAAA,IAC3D,CAAC;AAAA,EACH;AAAA,EAEQ,cACN,QACA,UACgC;AAChC,UAAM,SAAyC,CAAC;AAChD,QAAI,QAAQ;AACV,aAAO,KAAK,EAAE,MAAM,UAAU,SAAS,OAAO,CAAC;AAAA,IACjD;AACA,eAAW,OAAO,UAAU;AAC1B,UAAI,IAAI,SAAS,QAAQ;AACvB,cAAM,UAAU,MAAM,QAAQ,IAAI,OAAO,IACrC,gBAAe,qBAAqB,IAAI,OAAwB,IAChE,IAAI;AACR,eAAO,KAAK;AAAA,UACV,MAAM;AAAA,UACN,cAAc,IAAI;AAAA,UAClB;AAAA,QACF,CAAC;AAAA,MACH,WAAW,IAAI,SAAS,aAAa;AACnC,cAAM,QAAiC;AAAA,UACrC,MAAM;AAAA,UACN,SAAS,IAAI;AAAA,QACf;AACA,YAAI,IAAI,YAAY;AAClB,gBAAM,aAAa,IAAI;AAAA,QACzB;AACA,eAAO,KAAK,KAAK;AAAA,MACnB,WAAW,IAAI,SAAS,QAAQ;AAC9B,cAAM,UAAU,MAAM,QAAQ,IAAI,OAAO,IACrC,gBAAe,qBAAqB,IAAI,OAAwB,IAChE,IAAI;AACR,eAAO,KAAK,EAAE,MAAM,QAAQ,QAAQ,CAAC;AAAA,MACvC,OAAO;AACL,eAAO,KAAK,EAAE,MAAM,IAAI,MAAM,SAAS,IAAI,QAAQ,CAAC;AAAA,MACtD;AAAA,IACF;AACA,WAAO;AAAA,EACT;AACF;","names":[]}