@yourgpt/llm-sdk 0.1.0 → 0.1.1

package/dist/providers/ollama/index.mjs

@@ -0,0 +1,232 @@
+import { generateMessageId } from '@yourgpt/copilot-sdk/core';
+
+// src/adapters/ollama.ts
+
+// src/adapters/base.ts
+function formatMessages(messages, systemPrompt) {
+  const formatted = [];
+  if (systemPrompt) {
+    formatted.push({ role: "system", content: systemPrompt });
+  }
+  for (const msg of messages) {
+    formatted.push({
+      role: msg.role,
+      content: msg.content ?? ""
+    });
+  }
+  return formatted;
+}
+
+// src/adapters/ollama.ts
+var OllamaAdapter = class {
+  constructor(config = {}) {
+    this.provider = "ollama";
+    this.config = config;
+    this.model = config.model || "llama3";
+    this.baseUrl = config.baseUrl || "http://localhost:11434";
+  }
+  async *stream(request) {
+    const messages = formatMessages(request.messages, request.systemPrompt);
+    const messageId = generateMessageId();
+    yield { type: "message:start", id: messageId };
+    try {
+      const response = await fetch(`${this.baseUrl}/api/chat`, {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json"
+        },
+        body: JSON.stringify({
+          model: request.config?.model || this.model,
+          messages,
+          stream: true,
+          options: {
+            temperature: request.config?.temperature ?? this.config.temperature,
+            num_predict: request.config?.maxTokens ?? this.config.maxTokens
+          }
+        }),
+        signal: request.signal
+      });
+      if (!response.ok) {
+        throw new Error(`Ollama API error: ${response.status}`);
+      }
+      if (!response.body) {
+        throw new Error("No response body");
+      }
+      const reader = response.body.getReader();
+      const decoder = new TextDecoder();
+      let buffer = "";
+      while (true) {
+        const { done, value } = await reader.read();
+        if (done) break;
+        buffer += decoder.decode(value, { stream: true });
+        const lines = buffer.split("\n");
+        buffer = lines.pop() || "";
+        for (const line of lines) {
+          if (!line.trim()) continue;
+          try {
+            const chunk = JSON.parse(line);
+            if (chunk.message?.content) {
+              yield { type: "message:delta", content: chunk.message.content };
+            }
+            if (chunk.done) {
+              break;
+            }
+          } catch {
+          }
+        }
+      }
+      yield { type: "message:end" };
+      yield { type: "done" };
+    } catch (error) {
+      if (error.name === "AbortError") {
+        yield { type: "done" };
+      } else {
+        yield {
+          type: "error",
+          message: error instanceof Error ? error.message : "Unknown error",
+          code: "OLLAMA_ERROR"
+        };
+      }
+    }
+  }
+};
+function createOllamaAdapter(config) {
+  return new OllamaAdapter(config);
+}
+
+// src/providers/ollama/index.ts
+var OLLAMA_MODELS = {
+  // Llama series
+  llama3: {
+    vision: false,
+    tools: true,
+    maxTokens: 8192
+  },
+  "llama3:70b": {
+    vision: false,
+    tools: true,
+    maxTokens: 8192
+  },
+  "llama3.2": {
+    vision: false,
+    tools: true,
+    maxTokens: 8192
+  },
+  "llama3.2-vision": {
+    vision: true,
+    tools: true,
+    maxTokens: 8192
+  },
+  // Mistral series
+  mistral: {
+    vision: false,
+    tools: true,
+    maxTokens: 8192
+  },
+  "mistral-nemo": {
+    vision: false,
+    tools: true,
+    maxTokens: 128e3
+  },
+  mixtral: {
+    vision: false,
+    tools: true,
+    maxTokens: 32768
+  },
+  // CodeLlama
+  codellama: {
+    vision: false,
+    tools: false,
+    maxTokens: 16384
+  },
+  // Phi series
+  phi3: {
+    vision: false,
+    tools: true,
+    maxTokens: 4096
+  },
+  "phi3:medium": {
+    vision: false,
+    tools: true,
+    maxTokens: 4096
+  },
+  // Gemma series
+  gemma2: {
+    vision: false,
+    tools: false,
+    maxTokens: 8192
+  },
+  "gemma2:27b": {
+    vision: false,
+    tools: false,
+    maxTokens: 8192
+  },
+  // Qwen series
+  qwen2: {
+    vision: false,
+    tools: true,
+    maxTokens: 32768
+  },
+  "qwen2.5-coder": {
+    vision: false,
+    tools: true,
+    maxTokens: 32768
+  },
+  // LLaVA (vision)
+  llava: {
+    vision: true,
+    tools: false,
+    maxTokens: 4096
+  },
+  // DeepSeek
+  deepseek: {
+    vision: false,
+    tools: true,
+    maxTokens: 16384
+  },
+  "deepseek-coder": {
+    vision: false,
+    tools: false,
+    maxTokens: 16384
+  }
+};
+var DEFAULT_MODEL_CAPS = {
+  vision: false,
+  tools: false,
+  maxTokens: 4096
+};
+function createOllama(config = {}) {
+  const baseUrl = config.baseUrl ?? "http://localhost:11434";
+  return {
+    name: "ollama",
+    supportedModels: Object.keys(OLLAMA_MODELS),
+    languageModel(modelId) {
+      return createOllamaAdapter({
+        model: modelId,
+        baseUrl
+      });
+    },
+    getCapabilities(modelId) {
+      const baseModelName = modelId.split(":")[0];
+      const model = OLLAMA_MODELS[modelId] ?? OLLAMA_MODELS[baseModelName] ?? DEFAULT_MODEL_CAPS;
+      return {
+        supportsVision: model.vision,
+        supportsTools: model.tools,
+        supportsThinking: false,
+        supportsStreaming: true,
+        supportsPDF: false,
+        supportsAudio: false,
+        supportsVideo: false,
+        maxTokens: model.maxTokens,
+        supportedImageTypes: model.vision ? ["image/png", "image/jpeg", "image/gif"] : [],
+        supportsJsonMode: false,
+        supportsSystemMessages: true
+      };
+    }
+  };
+}
+var createOllamaProvider = createOllama;
+
+export { createOllama, createOllamaProvider };
+//# sourceMappingURL=index.mjs.map
+//# sourceMappingURL=index.mjs.map

package/dist/providers/ollama/index.mjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../../src/adapters/base.ts","../../../src/adapters/ollama.ts","../../../src/providers/ollama/index.ts"],"names":[],"mappings":";;;;;AAyEO,SAAS,cAAA,CACd,UACA,YAAA,EAC0C;AAC1C,EAAA,MAAM,YAAsD,EAAC;AAG7D,EAAA,IAAI,YAAA,EAAc;AAChB,IAAA,SAAA,CAAU,KAAK,EAAE,IAAA,EAAM,QAAA,EAAU,OAAA,EAAS,cAAc,CAAA;AAAA,EAC1D;AAGA,EAAA,KAAA,MAAW,OAAO,QAAA,EAAU;AAC1B,IAAA,SAAA,CAAU,IAAA,CAAK;AAAA,MACb,MAAM,GAAA,CAAI,IAAA;AAAA,MACV,OAAA,EAAS,IAAI,OAAA,IAAW;AAAA,KACzB,CAAA;AAAA,EACH;AAEA,EAAA,OAAO,SAAA;AACT;;;AC3EO,IAAM,gBAAN,MAA0C;AAAA,EAO/C,WAAA,CAAY,MAAA,GAA8B,EAAC,EAAG;AAN9C,IAAA,IAAA,CAAS,QAAA,GAAW,QAAA;AAOlB,IAAA,IAAA,CAAK,MAAA,GAAS,MAAA;AACd,IAAA,IAAA,CAAK,KAAA,GAAQ,OAAO,KAAA,IAAS,QAAA;AAC7B,IAAA,IAAA,CAAK,OAAA,GAAU,OAAO,OAAA,IAAW,wBAAA;AAAA,EACnC;AAAA,EAEA,OAAO,OAAO,OAAA,EAA6D;AACzE,IAAA,MAAM,QAAA,GAAW,cAAA,CAAe,OAAA,CAAQ,QAAA,EAAU,QAAQ,YAAY,CAAA;AAEtE,IAAA,MAAM,YAAY,iBAAA,EAAkB;AAGpC,IAAA,MAAM,EAAE,IAAA,EAAM,eAAA,EAAiB,EAAA,EAAI,SAAA,EAAU;AAE7C,IAAA,IAAI;AACF,MAAA,MAAM,WAAW,MAAM,KAAA,CAAM,CAAA,EAAG,IAAA,CAAK,OAAO,CAAA,SAAA,CAAA,EAAa;AAAA,QACvD,MAAA,EAAQ,MAAA;AAAA,QACR,OAAA,EAAS;AAAA,UACP,cAAA,EAAgB;AAAA,SAClB;AAAA,QACA,IAAA,EAAM,KAAK,SAAA,CAAU;AAAA,UACnB,KAAA,EAAO,OAAA,CAAQ,MAAA,EAAQ,KAAA,IAAS,IAAA,CAAK,KAAA;AAAA,UACrC,QAAA;AAAA,UACA,MAAA,EAAQ,IAAA;AAAA,UACR,OAAA,EAAS;AAAA,YACP,WAAA,EAAa,OAAA,CAAQ,MAAA,EAAQ,WAAA,IAAe,KAAK,MAAA,CAAO,WAAA;AAAA,YACxD,WAAA,EAAa,OAAA,CAAQ,MAAA,EAAQ,SAAA,IAAa,KAAK,MAAA,CAAO;AAAA;AACxD,SACD,CAAA;AAAA,QACD,QAAQ,OAAA,CAAQ;AAAA,OACjB,CAAA;AAED,MAAA,IAAI,CAAC,SAAS,EAAA,EAAI;AAChB,QAAA,MAAM,IAAI,KAAA,CAAM,CAAA,kBAAA,EAAqB,QAAA,CAAS,MAAM,CAAA,CAAE,CAAA;AAAA,MACxD;AAEA,MAAA,IAAI,CAAC,SAAS,IAAA,EAAM;AAClB,QAAA,MAAM,IAAI,MAAM,kBAAkB,CAAA;AAAA,MACpC;AAEA,MAAA,MAAM,MAAA,GAAS,QAAA,CAAS,IAAA,CAAK,SAAA,EAAU;AACvC,MAAA,MAAM,OAAA,GAAU,IAAI,WAAA,EAAY;AAChC,MAAA,IAAI,MAAA,GAAS,EAAA;AAEb,MAAA,OAAO,IAAA,EAAM;AACX,QAAA,MAAM,EAAE,IAAA,EAAM,KAAA,EAAM,GAAI,MAAM,OAAO,IAAA,EAAK;AAC1C,QAAA,IAAI,IAAA,EAAM;AAEV,QAAA,MAAA,IAAU,QAAQ,MAAA,CAAO,KAAA,EAAO,EAAE,MAAA,EAAQ,MAAM,CAAA;AAChD,QAAA,MAAM,KAAA,GAAQ,MAAA,CAAO,KAAA,CAAM,IAAI,CAAA;AAC/B,QAAA,MAAA,GAAS,KAAA,CAAM,KAAI,IAAK,EAAA;AAExB,QAAA,KAAA,MAAW,QAAQ,KAAA,EAAO;AACxB,UAAA,IAAI,CAAC,IAAA,CAAK,IAAA,EAAK,EAAG;AAElB,UAAA,IAAI;AACF,YAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,KAAA,CAAM,IAAI,CAAA;AAE7B,YAAA,IAAI,KAAA,CAAM,SAAS,OAAA,EAAS;AAC1B,cAAA,MAAM,EAAE,IAAA,EAAM,eAAA,EAAiB,OAAA,EAAS,KAAA,CAAM,QAAQ,OAAA,EAAQ;AAAA,YAChE;AAGA,YAAA,IAAI,MAAM,IAAA,EAAM;AACd,cAAA;AAAA,YACF;AAAA,UACF,CAAA,CAAA,MAAQ;AAAA,UAER;AAAA,QACF;AAAA,MACF;AAEA,MAAA,MAAM,EAAE,MAAM,aAAA,EAAc;AAC5B,MAAA,MAAM,EAAE,MAAM,MAAA,EAAO;AAAA,IACvB,SAAS,KAAA,EAAO;AACd,MAAA,IAAK,KAAA,CAAgB,SAAS,YAAA,EAAc;AAC1C,QAAA,MAAM,EAAE,MAAM,MAAA,EAAO;AAAA,MACvB,CAAA,MAAO;AACL,QAAA,MAAM;AAAA,UACJ,IAAA,EAAM,OAAA;AAAA,UACN,OAAA,EAAS,KAAA,YAAiB,KAAA,GAAQ,KAAA,CAAM,OAAA,GAAU,eAAA;AAAA,UAClD,IAAA,EAAM;AAAA,SACR;AAAA,MACF;AAAA,IACF;AAAA,EACF;AACF,CAAA;AAKO,SAAS,oBACd,MAAA,EACe;AACf,EAAA,OAAO,IAAI,cAAc,MAAM,CAAA;AACjC;;;AC/FA,IAAM,aAAA,GAAmD;AAAA;AAAA,EAEvD,MAAA,EAAQ;AAAA,IACN,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA,EACA,YAAA,EAAc;AAAA,IACZ,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA,EACA,UAAA,EAAY;AAAA,IACV,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA,EACA,iBAAA,EAAmB;AAAA,IACjB,MAAA,EAAQ,IAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA;AAAA,EAGA,OAAA,EAAS;AAAA,IACP,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA,EACA,cAAA,EAAgB;AAAA,IACd,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA,EACA,OAAA,EAAS;AAAA,IACP,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA;AAAA,EAGA,SAAA,EAAW;AAAA,IACT,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,KAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA;AAAA,EAGA,IAAA,EAAM;AAAA,IACJ,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA,EACA,aAAA,EAAe;AAAA,IACb,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA;AAAA,EAGA,MAAA,EAAQ;AAAA,IACN,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,KAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA,EACA,YAAA,EAAc;AAAA,IACZ,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,KAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA;AAAA,EAGA,KAAA,EAAO;AAAA,IACL,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA,EACA,eAAA,EAAiB;AAAA,IACf,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA;AAAA,EAGA,KAAA,EAAO;AAAA,IACL,MAAA,EAAQ,IAAA;AAAA,IACR,KAAA,EAAO,KAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA;AAAA,EAGA,QAAA,EAAU;AAAA,IACR,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,IAAA;AAAA,IACP,SAAA,EAAW;AAAA,GACb;AAAA,EACA,gBAAA,EAAkB;AAAA,IAChB,MAAA,EAAQ,KAAA;AAAA,IACR,KAAA,EAAO,KAAA;AAAA,IACP,SAAA,EAAW;AAAA;AAEf,CAAA;AAGA,IAAM,kBAAA,GAAwC;AAAA,EAC5C,MAAA,EAAQ,KAAA;AAAA,EACR,KAAA,EAAO,KAAA;AAAA,EACP,SAAA,EAAW;AACb,CAAA;AAeO,SAAS,YAAA,CAAa,MAAA,GAA+B,EAAC,EAAe;AAC1E,EAAA,MAAM,OAAA,GAAU,OAAO,OAAA,IAAW,wBAAA;AAElC,EAAA,OAAO;AAAA,IACL,IAAA,EAAM,QAAA;AAAA,IACN,eAAA,EAAiB,MAAA,CAAO,IAAA,CAAK,aAAa,CAAA;AAAA,IAE1C,cAAc,OAAA,EAAiB;AAC7B,MAAA,OAAO,mBAAA,CAAoB;AAAA,QACzB,KAAA,EAAO,OAAA;AAAA,QACP;AAAA,OACD,CAAA;AAAA,IACH,CAAA;AAAA,IAEA,gBAAgB,OAAA,EAAuC;AAErD,MAAA,MAAM,aAAA,GAAgB,OAAA,CAAQ,KAAA,CAAM,GAAG,EAAE,CAAC,CAAA;AAC1C,MAAA,MAAM,QACJ,aAAA,CAAc,OAAO,CAAA,IACrB,aAAA,CAAc,aAAa,CAAA,IAC3B,kBAAA;AAEF,MAAA,OAAO;AAAA,QACL,gBAAgB,KAAA,CAAM,MAAA;AAAA,QACtB,eAAe,KAAA,CAAM,KAAA;AAAA,QACrB,gBAAA,EAAkB,KAAA;AAAA,QAClB,iBAAA,EAAmB,IAAA;AAAA,QACnB,WAAA,EAAa,KAAA;AAAA,QACb,aAAA,EAAe,KAAA;AAAA,QACf,aAAA,EAAe,KAAA;AAAA,QACf,WAAW,KAAA,CAAM,SAAA;AAAA,QACjB,mBAAA,EAAqB,MAAM,MAAA,GACvB,CAAC,aAAa,YAAA,EAAc,WAAW,IACvC,EAAC;AAAA,QACL,gBAAA,EAAkB,KAAA;AAAA,QAClB,sBAAA,EAAwB;AAAA,OAC1B;AAAA,IACF;AAAA,GACF;AACF;AAGO,IAAM,oBAAA,GAAuB","file":"index.mjs","sourcesContent":["import type {\n  Message,\n  MessageAttachment,\n  ActionDefinition,\n  StreamEvent,\n  LLMConfig,\n} from \"@yourgpt/copilot-sdk/core\";\n\n/**\n * Chat completion request\n */\nexport interface ChatCompletionRequest {\n  /** Conversation messages */\n  messages: Message[];\n  /**\n   * Raw provider-formatted messages (for agent loop with tool calls)\n   * When provided, these are used instead of converting from Message[]\n   * This allows passing messages with tool_calls and tool role\n   */\n  rawMessages?: Array<Record<string, unknown>>;\n  /** Available actions/tools */\n  actions?: ActionDefinition[];\n  /** System prompt */\n  systemPrompt?: string;\n  /** LLM configuration overrides */\n  config?: Partial<LLMConfig>;\n  /** Abort signal for cancellation */\n  signal?: AbortSignal;\n}\n\n/**\n * Non-streaming completion result\n */\nexport interface CompletionResult {\n  /** Text content */\n  content: string;\n  /** Tool calls */\n  toolCalls: Array<{ id: string; name: string; args: Record<string, unknown> }>;\n  /** Thinking content (if extended thinking enabled) */\n  thinking?: string;\n  /** Raw provider response for debugging */\n  rawResponse: Record<string, unknown>;\n}\n\n/**\n * Base LLM adapter interface\n */\nexport interface LLMAdapter {\n  /** Provider name */\n  readonly provider: string;\n\n  /** Model name */\n  readonly model: string;\n\n  /**\n   * Stream a chat completion\n   */\n  stream(request: ChatCompletionRequest): AsyncGenerator<StreamEvent>;\n\n  /**\n   * Non-streaming chat completion (for debugging/comparison)\n   */\n  complete?(request: ChatCompletionRequest): Promise<CompletionResult>;\n}\n\n/**\n * Adapter factory function type\n */\nexport type AdapterFactory = (config: LLMConfig) => LLMAdapter;\n\n/**\n * Convert messages to provider format (simple text only)\n */\nexport function formatMessages(\n  messages: Message[],\n  systemPrompt?: string,\n): Array<{ role: string; content: string }> {\n  const formatted: Array<{ role: string; content: string }> = [];\n\n  // Add system prompt if provided\n  if (systemPrompt) {\n    formatted.push({ role: \"system\", content: systemPrompt });\n  }\n\n  // Add conversation messages\n  for (const msg of messages) {\n    formatted.push({\n      role: msg.role,\n      content: msg.content ?? \"\",\n    });\n  }\n\n  return formatted;\n}\n\n/**\n * Convert ActionParameter to JSON Schema format recursively\n */\nfunction parameterToJsonSchema(param: {\n  type: string;\n  description?: string;\n  enum?: string[];\n  items?: unknown;\n  properties?: Record<string, unknown>;\n}): Record<string, unknown> {\n  const schema: Record<string, unknown> = {\n    type: param.type,\n  };\n\n  if (param.description) {\n    schema.description = param.description;\n  }\n\n  if (param.enum) {\n    schema.enum = param.enum;\n  }\n\n  // Handle array items\n  if (param.type === \"array\" && param.items) {\n    schema.items = parameterToJsonSchema(\n      param.items as {\n        type: string;\n        description?: string;\n        enum?: string[];\n        items?: unknown;\n        properties?: Record<string, unknown>;\n      },\n    );\n  }\n\n  // Handle nested object properties\n  if (param.type === \"object\" && param.properties) {\n    schema.properties = Object.fromEntries(\n      Object.entries(param.properties).map(([key, prop]) => [\n        key,\n        parameterToJsonSchema(\n          prop as {\n            type: string;\n            description?: string;\n            enum?: string[];\n            items?: unknown;\n            properties?: Record<string, unknown>;\n          },\n        ),\n      ]),\n    );\n  }\n\n  return schema;\n}\n\n/**\n * Convert actions to OpenAI tool format\n */\nexport function formatTools(actions: ActionDefinition[]): Array<{\n  type: \"function\";\n  function: {\n    name: string;\n    description: string;\n    parameters: object;\n  };\n}> {\n  return actions.map((action) => ({\n    type: \"function\" as const,\n    function: {\n      name: action.name,\n      description: action.description,\n      parameters: {\n        type: \"object\",\n        properties: action.parameters\n          ? Object.fromEntries(\n              Object.entries(action.parameters).map(([key, param]) => [\n                key,\n                parameterToJsonSchema(param),\n              ]),\n            )\n          : {},\n        required: action.parameters\n          ? Object.entries(action.parameters)\n              .filter(([, param]) => param.required)\n              .map(([key]) => key)\n          : [],\n      },\n    },\n  }));\n}\n\n// ============================================\n// Vision/Multimodal Support\n// ============================================\n\n/**\n * Content block types for multimodal messages\n */\nexport type AnthropicContentBlock =\n  | { type: \"text\"; text: string }\n  | {\n      type: \"image\";\n      source:\n        | {\n            type: \"base64\";\n            media_type: string;\n            data: string;\n          }\n        | {\n            type: \"url\";\n            url: string;\n          };\n    }\n  | {\n      type: \"document\";\n      source:\n        | {\n            type: \"base64\";\n            media_type: string;\n            data: string;\n          }\n        | {\n            type: \"url\";\n            url: string;\n          };\n    };\n\nexport type OpenAIContentBlock =\n  | { type: \"text\"; text: string }\n  | {\n      type: \"image_url\";\n      image_url: {\n        url: string;\n        detail?: \"low\" | \"high\" | \"auto\";\n      };\n    };\n\n/**\n * Check if a message has image attachments\n * Supports both new format (metadata.attachments) and legacy (attachments)\n */\nexport function hasImageAttachments(message: Message): boolean {\n  const attachments = message.metadata?.attachments;\n  return attachments?.some((a) => a.type === \"image\") ?? false;\n}\n\n/**\n * Check if a message has media attachments (images or PDFs)\n */\nexport function hasMediaAttachments(message: Message): boolean {\n  const attachments = message.metadata?.attachments;\n  return (\n    attachments?.some(\n      (a) =>\n        a.type === \"image\" ||\n        (a.type === \"file\" && a.mimeType === \"application/pdf\"),\n    ) ?? false\n  );\n}\n\n/**\n * Convert MessageAttachment to Anthropic image content block\n *\n * Anthropic format:\n * {\n *   type: \"image\",\n *   source: {\n *     type: \"base64\",\n *     media_type: \"image/png\",\n *     data: \"base64data...\"\n *   }\n * }\n */\nexport function attachmentToAnthropicImage(\n  attachment: MessageAttachment,\n): AnthropicContentBlock | null {\n  if (attachment.type !== \"image\") return null;\n\n  // Use URL if available (cloud storage)\n  if (attachment.url) {\n    return {\n      type: \"image\",\n      source: {\n        type: \"url\",\n        url: attachment.url,\n      },\n    };\n  }\n\n  // Fall back to base64 data\n  if (!attachment.data) return null;\n\n  // Extract base64 data (remove data URI prefix if present)\n  let base64Data = attachment.data;\n  if (base64Data.startsWith(\"data:\")) {\n    const commaIndex = base64Data.indexOf(\",\");\n    if (commaIndex !== -1) {\n      base64Data = base64Data.slice(commaIndex + 1);\n    }\n  }\n\n  return {\n    type: \"image\",\n    source: {\n      type: \"base64\",\n      media_type: attachment.mimeType || \"image/png\",\n      data: base64Data,\n    },\n  };\n}\n\n/**\n * Convert MessageAttachment to OpenAI image_url content block\n *\n * OpenAI format:\n * {\n *   type: \"image_url\",\n *   image_url: {\n *     url: \"data:image/png;base64,...\"\n *   }\n * }\n */\nexport function attachmentToOpenAIImage(\n  attachment: MessageAttachment,\n): OpenAIContentBlock | null {\n  if (attachment.type !== \"image\") return null;\n\n  let imageUrl: string;\n\n  // Use URL if available (cloud storage)\n  if (attachment.url) {\n    imageUrl = attachment.url;\n  } else if (attachment.data) {\n    // Build data URI if not already one\n    imageUrl = attachment.data.startsWith(\"data:\")\n      ? attachment.data\n      : `data:${attachment.mimeType || \"image/png\"};base64,${attachment.data}`;\n  } else {\n    return null;\n  }\n\n  return {\n    type: \"image_url\",\n    image_url: {\n      url: imageUrl,\n      detail: \"auto\",\n    },\n  };\n}\n\n/**\n * Convert MessageAttachment (PDF) to Anthropic document content block\n *\n * Anthropic format:\n * {\n *   type: \"document\",\n *   source: {\n *     type: \"base64\",\n *     media_type: \"application/pdf\",\n *     data: \"base64data...\"\n *   }\n * }\n */\nexport function attachmentToAnthropicDocument(\n  attachment: MessageAttachment,\n): AnthropicContentBlock | null {\n  // Only handle PDF files\n  if (attachment.type !== \"file\" || attachment.mimeType !== \"application/pdf\") {\n    return null;\n  }\n\n  // Use URL if available (cloud storage)\n  if (attachment.url) {\n    return {\n      type: \"document\",\n      source: {\n        type: \"url\",\n        url: attachment.url,\n      },\n    };\n  }\n\n  // Fall back to base64 data\n  if (!attachment.data) return null;\n\n  // Extract base64 data (remove data URI prefix if present)\n  let base64Data = attachment.data;\n  if (base64Data.startsWith(\"data:\")) {\n    const commaIndex = base64Data.indexOf(\",\");\n    if (commaIndex !== -1) {\n      base64Data = base64Data.slice(commaIndex + 1);\n    }\n  }\n\n  return {\n    type: \"document\",\n    source: {\n      type: \"base64\",\n      media_type: \"application/pdf\",\n      data: base64Data,\n    },\n  };\n}\n\n/**\n * Convert a Message to Anthropic multimodal content blocks\n */\nexport function messageToAnthropicContent(\n  message: Message,\n): string | AnthropicContentBlock[] {\n  const attachments = message.metadata?.attachments;\n  const content = message.content ?? \"\";\n\n  // If no media attachments (images or PDFs), return simple string\n  if (!hasMediaAttachments(message)) {\n    return content;\n  }\n\n  // Build content blocks array\n  const blocks: AnthropicContentBlock[] = [];\n\n  // Add media attachments first (Claude recommends media before text)\n  if (attachments) {\n    for (const attachment of attachments) {\n      // Try image first\n      const imageBlock = attachmentToAnthropicImage(attachment);\n      if (imageBlock) {\n        blocks.push(imageBlock);\n        continue;\n      }\n      // Try document (PDF)\n      const docBlock = attachmentToAnthropicDocument(attachment);\n      if (docBlock) {\n        blocks.push(docBlock);\n      }\n    }\n  }\n\n  // Add text content\n  if (content) {\n    blocks.push({ type: \"text\", text: content });\n  }\n\n  return blocks;\n}\n\n/**\n * Convert a Message to OpenAI multimodal content blocks\n */\nexport function messageToOpenAIContent(\n  message: Message,\n): string | OpenAIContentBlock[] {\n  const attachments = message.metadata?.attachments;\n  const content = message.content ?? \"\";\n\n  // If no image attachments, return simple string\n  if (!hasImageAttachments(message)) {\n    return content;\n  }\n\n  // Build content blocks array\n  const blocks: OpenAIContentBlock[] = [];\n\n  // Add text content first\n  if (content) {\n    blocks.push({ type: \"text\", text: content });\n  }\n\n  // Add image attachments\n  if (attachments) {\n    for (const attachment of attachments) {\n      const imageBlock = attachmentToOpenAIImage(attachment);\n      if (imageBlock) {\n        blocks.push(imageBlock);\n      }\n    }\n  }\n\n  return blocks;\n}\n\n/**\n * Anthropic content block types (extended for tools)\n */\nexport type AnthropicToolUseBlock = {\n  type: \"tool_use\";\n  id: string;\n  name: string;\n  input: Record<string, unknown>;\n};\n\nexport type AnthropicToolResultBlock = {\n  type: \"tool_result\";\n  tool_use_id: string;\n  content: string;\n};\n\nexport type AnthropicMessageContent =\n  | string\n  | Array<\n      AnthropicContentBlock | AnthropicToolUseBlock | AnthropicToolResultBlock\n    >;\n\n/**\n * Format messages for Anthropic with full tool support\n * Handles: text, images, tool_use, and tool_result\n *\n * Key differences from OpenAI:\n * - tool_calls become tool_use blocks in assistant content\n * - tool results become tool_result blocks in user content\n */\nexport function formatMessagesForAnthropic(\n  messages: Message[],\n  systemPrompt?: string,\n): {\n  system: string;\n  messages: Array<{\n    role: \"user\" | \"assistant\";\n    content: AnthropicMessageContent;\n  }>;\n} {\n  const formatted: Array<{\n    role: \"user\" | \"assistant\";\n    content: AnthropicMessageContent;\n  }> = [];\n\n  for (let i = 0; i < messages.length; i++) {\n    const msg = messages[i];\n\n    if (msg.role === \"system\") continue; // System handled separately\n\n    if (msg.role === \"assistant\") {\n      // Build content array for assistant\n      const content: Array<AnthropicContentBlock | AnthropicToolUseBlock> = [];\n\n      // Add text content if present\n      if (msg.content) {\n        content.push({ type: \"text\", text: msg.content });\n      }\n\n      // Convert tool_calls to tool_use blocks\n      if (msg.tool_calls && msg.tool_calls.length > 0) {\n        for (const tc of msg.tool_calls) {\n          content.push({\n            type: \"tool_use\",\n            id: tc.id,\n            name: tc.function.name,\n            input: JSON.parse(tc.function.arguments),\n          });\n        }\n      }\n\n      formatted.push({\n        role: \"assistant\",\n        content:\n          content.length === 1 && content[0].type === \"text\"\n            ? (content[0] as { type: \"text\"; text: string }).text\n            : content,\n      });\n    } else if (msg.role === \"tool\" && msg.tool_call_id) {\n      // Tool results go in user message as tool_result blocks\n      // Group consecutive tool messages together\n      const toolResults: AnthropicToolResultBlock[] = [\n        {\n          type: \"tool_result\",\n          tool_use_id: msg.tool_call_id,\n          content: msg.content ?? \"\",\n        },\n      ];\n\n      // Look ahead for more consecutive tool messages\n      while (i + 1 < messages.length && messages[i + 1].role === \"tool\") {\n        i++;\n        const nextTool = messages[i];\n        if (nextTool.tool_call_id) {\n          toolResults.push({\n            type: \"tool_result\",\n            tool_use_id: nextTool.tool_call_id,\n            content: nextTool.content ?? \"\",\n          });\n        }\n      }\n\n      formatted.push({\n        role: \"user\",\n        content: toolResults,\n      });\n    } else if (msg.role === \"user\") {\n      formatted.push({\n        role: \"user\",\n        content: messageToAnthropicContent(msg),\n      });\n    }\n  }\n\n  return {\n    system: systemPrompt || \"\",\n    messages: formatted,\n  };\n}\n\n/**\n * OpenAI message format with tool support\n */\nexport type OpenAIMessage =\n  | { role: \"system\"; content: string }\n  | { role: \"user\"; content: string | OpenAIContentBlock[] }\n  | {\n      role: \"assistant\";\n      content: string | null;\n      tool_calls?: Array<{\n        id: string;\n        type: \"function\";\n        function: { name: string; arguments: string };\n      }>;\n    }\n  | { role: \"tool\"; content: string; tool_call_id: string };\n\n/**\n * Format messages for OpenAI with full tool support\n * Handles: text, images, tool_calls, and tool results\n */\nexport function formatMessagesForOpenAI(\n  messages: Message[],\n  systemPrompt?: string,\n): OpenAIMessage[] {\n  const formatted: OpenAIMessage[] = [];\n\n  // Add system prompt if provided\n  if (systemPrompt) {\n    formatted.push({ role: \"system\", content: systemPrompt });\n  }\n\n  for (const msg of messages) {\n    if (msg.role === \"system\") {\n      formatted.push({ role: \"system\", content: msg.content ?? \"\" });\n    } else if (msg.role === \"user\") {\n      formatted.push({\n        role: \"user\",\n        content: messageToOpenAIContent(msg),\n      });\n    } else if (msg.role === \"assistant\") {\n      const assistantMsg: OpenAIMessage = {\n        role: \"assistant\",\n        content: msg.content,\n      };\n      if (msg.tool_calls && msg.tool_calls.length > 0) {\n        (assistantMsg as { tool_calls: typeof msg.tool_calls }).tool_calls =\n          msg.tool_calls;\n      }\n      formatted.push(assistantMsg);\n    } else if (msg.role === \"tool\" && msg.tool_call_id) {\n      formatted.push({\n        role: \"tool\",\n        content: msg.content ?? \"\",\n        tool_call_id: msg.tool_call_id,\n      });\n    }\n  }\n\n  return formatted;\n}\n","import type { LLMConfig, StreamEvent } from \"@yourgpt/copilot-sdk/core\";\nimport { generateMessageId } from \"@yourgpt/copilot-sdk/core\";\nimport type { LLMAdapter, ChatCompletionRequest } from \"./base\";\nimport { formatMessages } from \"./base\";\n\n/**\n * Ollama adapter configuration\n */\nexport interface OllamaAdapterConfig extends Partial<LLMConfig> {\n  model?: string;\n  baseUrl?: string;\n}\n\n/**\n * Ollama LLM Adapter (Local models)\n *\n * Supports: Llama 3, Mistral, Phi, Gemma, CodeLlama, etc.\n */\nexport class OllamaAdapter implements LLMAdapter {\n  readonly provider = \"ollama\";\n  readonly model: string;\n\n  private baseUrl: string;\n  private config: OllamaAdapterConfig;\n\n  constructor(config: OllamaAdapterConfig = {}) {\n    this.config = config;\n    this.model = config.model || \"llama3\";\n    this.baseUrl = config.baseUrl || \"http://localhost:11434\";\n  }\n\n  async *stream(request: ChatCompletionRequest): AsyncGenerator<StreamEvent> {\n    const messages = formatMessages(request.messages, request.systemPrompt);\n\n    const messageId = generateMessageId();\n\n    // Emit message start\n    yield { type: \"message:start\", id: messageId };\n\n    try {\n      const response = await fetch(`${this.baseUrl}/api/chat`, {\n        method: \"POST\",\n        headers: {\n          \"Content-Type\": \"application/json\",\n        },\n        body: JSON.stringify({\n          model: request.config?.model || this.model,\n          messages,\n          stream: true,\n          options: {\n            temperature: request.config?.temperature ?? this.config.temperature,\n            num_predict: request.config?.maxTokens ?? this.config.maxTokens,\n          },\n        }),\n        signal: request.signal,\n      });\n\n      if (!response.ok) {\n        throw new Error(`Ollama API error: ${response.status}`);\n      }\n\n      if (!response.body) {\n        throw new Error(\"No response body\");\n      }\n\n      const reader = response.body.getReader();\n      const decoder = new TextDecoder();\n      let buffer = \"\";\n\n      while (true) {\n        const { done, value } = await reader.read();\n        if (done) break;\n\n        buffer += decoder.decode(value, { stream: true });\n        const lines = buffer.split(\"\\n\");\n        buffer = lines.pop() || \"\";\n\n        for (const line of lines) {\n          if (!line.trim()) continue;\n\n          try {\n            const chunk = JSON.parse(line);\n\n            if (chunk.message?.content) {\n              yield { type: \"message:delta\", content: chunk.message.content };\n            }\n\n            // Ollama indicates completion with done: true\n            if (chunk.done) {\n              break;\n            }\n          } catch {\n            // Skip invalid JSON\n          }\n        }\n      }\n\n      yield { type: \"message:end\" };\n      yield { type: \"done\" };\n    } catch (error) {\n      if ((error as Error).name === \"AbortError\") {\n        yield { type: \"done\" };\n      } else {\n        yield {\n          type: \"error\",\n          message: error instanceof Error ? error.message : \"Unknown error\",\n          code: \"OLLAMA_ERROR\",\n        };\n      }\n    }\n  }\n}\n\n/**\n * Create Ollama adapter\n */\nexport function createOllamaAdapter(\n  config?: OllamaAdapterConfig,\n): OllamaAdapter {\n  return new OllamaAdapter(config);\n}\n","/**\n * Ollama Provider\n *\n * Wraps the existing OllamaAdapter with provider interface.\n * Ollama runs models locally on your machine.\n */\n\nimport { createOllamaAdapter } from \"../../adapters/ollama\";\nimport type {\n  AIProvider,\n  ProviderCapabilities,\n  OllamaProviderConfig,\n} from \"../types\";\n\n// ============================================\n// Model Definitions\n// ============================================\n\ninterface ModelCapabilities {\n  vision: boolean;\n  tools: boolean;\n  maxTokens: number;\n}\n\n// Common Ollama models - users can run any model\nconst OLLAMA_MODELS: Record<string, ModelCapabilities> = {\n  // Llama series\n  llama3: {\n    vision: false,\n    tools: true,\n    maxTokens: 8192,\n  },\n  \"llama3:70b\": {\n    vision: false,\n    tools: true,\n    maxTokens: 8192,\n  },\n  \"llama3.2\": {\n    vision: false,\n    tools: true,\n    maxTokens: 8192,\n  },\n  \"llama3.2-vision\": {\n    vision: true,\n    tools: true,\n    maxTokens: 8192,\n  },\n\n  // Mistral series\n  mistral: {\n    vision: false,\n    tools: true,\n    maxTokens: 8192,\n  },\n  \"mistral-nemo\": {\n    vision: false,\n    tools: true,\n    maxTokens: 128000,\n  },\n  mixtral: {\n    vision: false,\n    tools: true,\n    maxTokens: 32768,\n  },\n\n  // CodeLlama\n  codellama: {\n    vision: false,\n    tools: false,\n    maxTokens: 16384,\n  },\n\n  // Phi series\n  phi3: {\n    vision: false,\n    tools: true,\n    maxTokens: 4096,\n  },\n  \"phi3:medium\": {\n    vision: false,\n    tools: true,\n    maxTokens: 4096,\n  },\n\n  // Gemma series\n  gemma2: {\n    vision: false,\n    tools: false,\n    maxTokens: 8192,\n  },\n  \"gemma2:27b\": {\n    vision: false,\n    tools: false,\n    maxTokens: 8192,\n  },\n\n  // Qwen series\n  qwen2: {\n    vision: false,\n    tools: true,\n    maxTokens: 32768,\n  },\n  \"qwen2.5-coder\": {\n    vision: false,\n    tools: true,\n    maxTokens: 32768,\n  },\n\n  // LLaVA (vision)\n  llava: {\n    vision: true,\n    tools: false,\n    maxTokens: 4096,\n  },\n\n  // DeepSeek\n  deepseek: {\n    vision: false,\n    tools: true,\n    maxTokens: 16384,\n  },\n  \"deepseek-coder\": {\n    vision: false,\n    tools: false,\n    maxTokens: 16384,\n  },\n};\n\n// Default capabilities for unknown models\nconst DEFAULT_MODEL_CAPS: ModelCapabilities = {\n  vision: false,\n  tools: false,\n  maxTokens: 4096,\n};\n\n// ============================================\n// Provider Implementation\n// ============================================\n\n/**\n * Create an Ollama provider\n *\n * @example\n * ```typescript\n * const ollama = createOllama({ baseUrl: 'http://localhost:11434' });\n * const adapter = ollama.languageModel('llama3');\n * ```\n */\nexport function createOllama(config: OllamaProviderConfig = {}): AIProvider {\n  const baseUrl = config.baseUrl ?? \"http://localhost:11434\";\n\n  return {\n    name: \"ollama\",\n    supportedModels: Object.keys(OLLAMA_MODELS),\n\n    languageModel(modelId: string) {\n      return createOllamaAdapter({\n        model: modelId,\n        baseUrl,\n      });\n    },\n\n    getCapabilities(modelId: string): ProviderCapabilities {\n      // Try exact match first, then try base model name\n      const baseModelName = modelId.split(\":\")[0];\n      const model =\n        OLLAMA_MODELS[modelId] ??\n        OLLAMA_MODELS[baseModelName] ??\n        DEFAULT_MODEL_CAPS;\n\n      return {\n        supportsVision: model.vision,\n        supportsTools: model.tools,\n        supportsThinking: false,\n        supportsStreaming: true,\n        supportsPDF: false,\n        supportsAudio: false,\n        supportsVideo: false,\n        maxTokens: model.maxTokens,\n        supportedImageTypes: model.vision\n          ? [\"image/png\", \"image/jpeg\", \"image/gif\"]\n          : [],\n        supportsJsonMode: false,\n        supportsSystemMessages: true,\n      };\n    },\n  };\n}\n\n// Alias for consistency\nexport const createOllamaProvider = createOllama;\n"]}

package/dist/providers/openai/index.d.mts

@@ -0,0 +1,82 @@
+import { L as LanguageModel } from '../../types-CdORv1Yu.mjs';
+import { b as OpenAIProviderConfig, A as AIProvider } from '../../types-BBCZ3Fxy.mjs';
+import 'zod';
+import '@yourgpt/copilot-sdk/core';
+import '../../base-D_FyHFKj.mjs';
+
+/**
+ * OpenAI Provider - Modern Pattern
+ *
+ * Returns a LanguageModel instance that can be used directly with
+ * generateText() and streamText().
+ *
+ * @example
+ * ```ts
+ * import { openai } from '@yourgpt/llm-sdk/openai';
+ * import { generateText } from '@yourgpt/llm-sdk';
+ *
+ * const result = await generateText({
+ *   model: openai('gpt-4o'),
+ *   prompt: 'Hello!',
+ * });
+ * ```
+ */
+
+interface OpenAIProviderOptions {
+    /** API key (defaults to OPENAI_API_KEY env var) */
+    apiKey?: string;
+    /** Base URL for API (defaults to https://api.openai.com/v1) */
+    baseURL?: string;
+    /** Organization ID */
+    organization?: string;
+    /** Default headers */
+    headers?: Record<string, string>;
+}
+/**
+ * Create an OpenAI language model
+ *
+ * @param modelId - Model ID (e.g., 'gpt-4o', 'gpt-4o-mini')
+ * @param options - Provider options
+ * @returns LanguageModel instance
+ *
+ * @example
+ * ```ts
+ * // Basic usage
+ * const model = openai('gpt-4o');
+ *
+ * // With custom options
+ * const model = openai('gpt-4o', {
+ *   apiKey: 'sk-...',
+ *   baseURL: 'https://custom-endpoint.com/v1',
+ * });
+ * ```
+ */
+declare function openai(modelId: string, options?: OpenAIProviderOptions): LanguageModel;
+
+/**
+ * OpenAI Provider
+ *
+ * Modern pattern: openai('gpt-4o') returns a LanguageModel
+ * Legacy pattern: createOpenAI({ apiKey }) returns an AIProvider
+ */
+
+/**
+ * Create an OpenAI provider (Legacy API)
+ *
+ * @deprecated Use `import { openai } from '@yourgpt/llm-sdk/openai'` instead.
+ *
+ * @example
+ * ```typescript
+ * // OLD (deprecated):
+ * const provider = createOpenAI({ apiKey: '...' });
+ * const adapter = provider.languageModel('gpt-4o');
+ *
+ * // NEW (recommended):
+ * import { openai } from '@yourgpt/llm-sdk/openai';
+ * const model = openai('gpt-4o');
+ * ```
+ */
+declare function createOpenAI(config?: OpenAIProviderConfig): AIProvider;
+declare const createOpenAIProvider: typeof createOpenAI;
+
+export { type OpenAIProviderOptions, createOpenAI, openai as createOpenAIModel, createOpenAIProvider, openai };

package/dist/providers/openai/index.d.ts

@@ -0,0 +1,82 @@
+import { L as LanguageModel } from '../../types-CdORv1Yu.js';
+import { b as OpenAIProviderConfig, A as AIProvider } from '../../types-DcoCaVVC.js';
+import 'zod';
+import '@yourgpt/copilot-sdk/core';
+import '../../base-D_FyHFKj.js';
+
+/**
+ * OpenAI Provider - Modern Pattern
+ *
+ * Returns a LanguageModel instance that can be used directly with
+ * generateText() and streamText().
+ *
+ * @example
+ * ```ts
+ * import { openai } from '@yourgpt/llm-sdk/openai';
+ * import { generateText } from '@yourgpt/llm-sdk';
+ *
+ * const result = await generateText({
+ *   model: openai('gpt-4o'),
+ *   prompt: 'Hello!',
+ * });
+ * ```
+ */
+
+interface OpenAIProviderOptions {
+    /** API key (defaults to OPENAI_API_KEY env var) */
+    apiKey?: string;
+    /** Base URL for API (defaults to https://api.openai.com/v1) */
+    baseURL?: string;
+    /** Organization ID */
+    organization?: string;
+    /** Default headers */
+    headers?: Record<string, string>;
+}
+/**
+ * Create an OpenAI language model
+ *
+ * @param modelId - Model ID (e.g., 'gpt-4o', 'gpt-4o-mini')
+ * @param options - Provider options
+ * @returns LanguageModel instance
+ *
+ * @example
+ * ```ts
+ * // Basic usage
+ * const model = openai('gpt-4o');
+ *
+ * // With custom options
+ * const model = openai('gpt-4o', {
+ *   apiKey: 'sk-...',
+ *   baseURL: 'https://custom-endpoint.com/v1',
+ * });
+ * ```
+ */
+declare function openai(modelId: string, options?: OpenAIProviderOptions): LanguageModel;
+
+/**
+ * OpenAI Provider
+ *
+ * Modern pattern: openai('gpt-4o') returns a LanguageModel
+ * Legacy pattern: createOpenAI({ apiKey }) returns an AIProvider
+ */
+
+/**
+ * Create an OpenAI provider (Legacy API)
+ *
+ * @deprecated Use `import { openai } from '@yourgpt/llm-sdk/openai'` instead.
+ *
+ * @example
+ * ```typescript
+ * // OLD (deprecated):
+ * const provider = createOpenAI({ apiKey: '...' });
+ * const adapter = provider.languageModel('gpt-4o');
+ *
+ * // NEW (recommended):
+ * import { openai } from '@yourgpt/llm-sdk/openai';
+ * const model = openai('gpt-4o');
+ * ```
+ */
+declare function createOpenAI(config?: OpenAIProviderConfig): AIProvider;
+declare const createOpenAIProvider: typeof createOpenAI;
+
+export { type OpenAIProviderOptions, createOpenAI, openai as createOpenAIModel, createOpenAIProvider, openai };