@ai-sdk/otel 1.0.0-beta.3 → 1.0.0-beta.30

package/src/gen-ai-format-messages.ts

@@ -0,0 +1,573 @@
+import {
+  LanguageModelV4CustomPart,
+  LanguageModelV4FilePart,
+  LanguageModelV4Message,
+  LanguageModelV4Prompt,
+  LanguageModelV4ReasoningFilePart,
+  LanguageModelV4ReasoningPart,
+  LanguageModelV4TextPart,
+  LanguageModelV4ToolApprovalResponsePart,
+  LanguageModelV4ToolCallPart,
+  LanguageModelV4ToolResultPart,
+} from '@ai-sdk/provider';
+import type { ModelMessage } from '@ai-sdk/provider-utils';
+import { convertDataContentToBase64String } from 'ai';
+
+type LanguageModelV4ContentPart =
+  | LanguageModelV4TextPart
+  | LanguageModelV4FilePart
+  | LanguageModelV4CustomPart
+  | LanguageModelV4ReasoningPart
+  | LanguageModelV4ReasoningFilePart
+  | LanguageModelV4ToolCallPart
+  | LanguageModelV4ToolResultPart
+  | LanguageModelV4ToolApprovalResponsePart;
+
+type SemConvPart =
+  | { type: 'text'; content: string }
+  | { type: 'reasoning'; content: string }
+  | {
+      type: 'tool_call';
+      id: string | null;
+      name: string;
+      arguments?: unknown;
+    }
+  | {
+      type: 'tool_call_response';
+      id: string | null;
+      response: unknown;
+    }
+  | {
+      type: 'blob';
+      modality: string;
+      mime_type: string | null;
+      content: string;
+    }
+  | { type: string; [key: string]: unknown };
+
+interface SemConvInputMessage {
+  role: string;
+  parts: SemConvPart[];
+}
+
+interface SemConvOutputMessage {
+  role: string;
+  parts: SemConvPart[];
+  finish_reason: string;
+}
+
+interface SemConvSystemInstruction {
+  type: string;
+  content?: string;
+  [key: string]: unknown;
+}
+
+/**
+ * Maps an AI SDK provider string (e.g. "anthropic.messages", "openai.chat")
+ * to a well-known gen_ai.provider.name value per the OTel GenAI SemConv.
+ *
+ * Provider strings come in formats like:
+ *   "openai.chat", "google.generative-ai", "google.vertex.chat",
+ *   "google-vertex", "amazon-bedrock.chat", "azure.chat"
+ *
+ * We match against longest-prefix-first to handle multi-segment prefixes
+ * like "google.vertex" and "google.generative-ai" before the single-segment "google".
+ */
+export function mapProviderName(provider: string): string {
+  const lower = provider.toLowerCase();
+
+  const wellKnownPrefixes: Array<[string, string]> = [
+    ['google.vertex', 'gcp.vertex_ai'],
+    ['google.generative-ai', 'gcp.gemini'],
+    ['google-vertex', 'gcp.vertex_ai'],
+    ['amazon-bedrock', 'aws.bedrock'],
+    ['azure-openai', 'azure.ai.openai'],
+    ['anthropic', 'anthropic'],
+    ['openai', 'openai'],
+    ['azure', 'azure.ai.inference'],
+    ['google', 'gcp.gemini'],
+    ['mistral', 'mistral_ai'],
+    ['cohere', 'cohere'],
+    ['bedrock', 'aws.bedrock'],
+    ['groq', 'groq'],
+    ['deepseek', 'deepseek'],
+    ['perplexity', 'perplexity'],
+    ['xai', 'x_ai'],
+  ];
+
+  for (const [prefix, mapped] of wellKnownPrefixes) {
+    if (
+      lower === prefix ||
+      lower.startsWith(prefix + '.') ||
+      lower.startsWith(prefix + '-')
+    ) {
+      return mapped;
+    }
+  }
+
+  return provider;
+}
+
+/**
+ * Maps an AI SDK operationId to a gen_ai.operation.name value.
+ */
+export function mapOperationName(operationId: string): string {
+  const mapping: Record<string, string> = {
+    'ai.generateText': 'invoke_agent',
+    'ai.streamText': 'invoke_agent',
+    'ai.generateObject': 'invoke_agent',
+    'ai.streamObject': 'invoke_agent',
+    'ai.embed': 'embeddings',
+    'ai.embedMany': 'embeddings',
+    'ai.rerank': 'rerank',
+  };
+  return mapping[operationId] ?? operationId;
+}
+
+/**
+ * Converts a system value to the gen_ai.system_instructions SemConv format.
+ * Accepts a plain string, a single SystemModelMessage, or an array of them.
+ * Schema: array of parts, each with at least { type, content }.
+ */
+export function formatSystemInstructions(
+  system:
+    | string
+    | { role: 'system'; content: string }
+    | Array<{ role: 'system'; content: string }>,
+): SemConvSystemInstruction[] {
+  if (typeof system === 'string') {
+    return [{ type: 'text', content: system }];
+  }
+  if (Array.isArray(system)) {
+    return system.map(msg => ({ type: 'text', content: msg.content }));
+  }
+  return [{ type: 'text', content: system.content }];
+}
+
+function convertMessagePartToSemConv(
+  part: LanguageModelV4ContentPart,
+): SemConvPart {
+  switch (part.type) {
+    case 'text':
+      return { type: 'text', content: part.text };
+
+    case 'reasoning':
+      return { type: 'reasoning', content: part.text };
+
+    case 'tool-call':
+      return {
+        type: 'tool_call',
+        id: part.toolCallId ?? null,
+        name: part.toolName,
+        arguments: part.input,
+      };
+
+    case 'tool-result': {
+      const output = part.output;
+      let response: unknown;
+      if (output) {
+        if (output.type === 'text' || output.type === 'error-text') {
+          response = output.value;
+        } else if (output.type === 'json' || output.type === 'error-json') {
+          response = output.value;
+        } else if (output.type === 'execution-denied') {
+          response = { denied: true, reason: output.reason };
+        } else {
+          response = output;
+        }
+      }
+      return {
+        type: 'tool_call_response',
+        id: part.toolCallId ?? null,
+        response,
+      };
+    }
+
+    case 'file': {
+      const data = part.data;
+      let content: string;
+      if (data instanceof Uint8Array) {
+        content = convertDataContentToBase64String(data);
+      } else if (typeof data === 'string') {
+        if (data.startsWith('http://') || data.startsWith('https://')) {
+          return {
+            type: 'uri',
+            modality: getModality(part.mediaType),
+            mime_type: part.mediaType ?? null,
+            uri: data,
+          };
+        }
+        content = data;
+      } else {
+        content = String(data);
+      }
+      return {
+        type: 'blob',
+        modality: getModality(part.mediaType),
+        mime_type: part.mediaType ?? null,
+        content,
+      };
+    }
+
+    case 'tool-approval-response':
+      return {
+        type: 'tool_approval_response',
+        approval_id: part.approvalId,
+        approved: part.approved,
+        reason: part.reason,
+      };
+
+    case 'custom':
+      return { type: 'custom', kind: part.kind };
+
+    case 'reasoning-file':
+      return { type: String(part.type) };
+
+    default: {
+      const _exhaustive: never = part;
+      return { type: String((_exhaustive as { type: string }).type) };
+    }
+  }
+}
+
+function getModality(mediaType: string | undefined): string {
+  if (!mediaType) return 'image';
+  if (mediaType.startsWith('image/')) return 'image';
+  if (mediaType.startsWith('video/')) return 'video';
+  if (mediaType.startsWith('audio/')) return 'audio';
+  return 'image';
+}
+
+/**
+ * Converts a LanguageModelV4Prompt to the gen_ai.input.messages SemConv format.
+ * System messages are excluded (they go into gen_ai.system_instructions).
+ */
+export function formatInputMessages(
+  prompt: LanguageModelV4Prompt,
+): SemConvInputMessage[] {
+  return prompt
+    .filter(msg => msg.role !== 'system')
+    .map((message: LanguageModelV4Message) => {
+      if (message.role === 'system') {
+        return {
+          role: 'system',
+          parts: [{ type: 'text', content: message.content }],
+        };
+      }
+
+      const parts = message.content.map(convertMessagePartToSemConv);
+      return { role: message.role, parts };
+    });
+}
+
+/**
+ * Converts user-facing ModelMessage[] (and optional prompt string) to the
+ * gen_ai.input.messages SemConv format. System messages are excluded
+ * (they belong in gen_ai.system_instructions).
+ */
+export function formatModelMessages({
+  prompt,
+  messages,
+}: {
+  prompt: string | Array<ModelMessage> | undefined;
+  messages: Array<ModelMessage> | undefined;
+}): SemConvInputMessage[] {
+  const result: SemConvInputMessage[] = [];
+
+  if (typeof prompt === 'string') {
+    result.push({
+      role: 'user',
+      parts: [{ type: 'text', content: prompt }],
+    });
+  } else if (Array.isArray(prompt)) {
+    for (const msg of prompt) {
+      const converted = convertModelMessageToSemConv(msg);
+      if (converted) result.push(converted);
+    }
+  }
+
+  if (messages) {
+    for (const msg of messages) {
+      const converted = convertModelMessageToSemConv(msg);
+      if (converted) result.push(converted);
+    }
+  }
+
+  return result;
+}
+
+function convertModelMessageToSemConv(
+  msg: ModelMessage,
+): SemConvInputMessage | undefined {
+  if (msg.role === 'system') return undefined;
+
+  if (msg.role === 'user') {
+    if (typeof msg.content === 'string') {
+      return {
+        role: 'user',
+        parts: [{ type: 'text', content: msg.content }],
+      };
+    }
+    const parts: SemConvPart[] = msg.content.map(part => {
+      switch (part.type) {
+        case 'text':
+          return { type: 'text' as const, content: part.text };
+        case 'image': {
+          const data = part.image;
+          if (data instanceof URL) {
+            return {
+              type: 'uri' as const,
+              modality: 'image',
+              mime_type: part.mediaType ?? null,
+              uri: data.toString(),
+            };
+          }
+          if (typeof data === 'string') {
+            if (data.startsWith('http://') || data.startsWith('https://')) {
+              return {
+                type: 'uri' as const,
+                modality: 'image',
+                mime_type: part.mediaType ?? null,
+                uri: data,
+              };
+            }
+            return {
+              type: 'blob' as const,
+              modality: 'image',
+              mime_type: part.mediaType ?? null,
+              content: data,
+            };
+          }
+          return {
+            type: 'blob' as const,
+            modality: 'image',
+            mime_type: part.mediaType ?? null,
+            content: convertDataContentToBase64String(data as Uint8Array),
+          };
+        }
+        case 'file': {
+          const data = part.data;
+          if (data instanceof URL) {
+            return {
+              type: 'uri' as const,
+              modality: getModality(part.mediaType),
+              mime_type: part.mediaType ?? null,
+              uri: data.toString(),
+            };
+          }
+          if (typeof data === 'string') {
+            if (data.startsWith('http://') || data.startsWith('https://')) {
+              return {
+                type: 'uri' as const,
+                modality: getModality(part.mediaType),
+                mime_type: part.mediaType ?? null,
+                uri: data,
+              };
+            }
+            return {
+              type: 'blob' as const,
+              modality: getModality(part.mediaType),
+              mime_type: part.mediaType ?? null,
+              content: data,
+            };
+          }
+          return {
+            type: 'blob' as const,
+            modality: getModality(part.mediaType),
+            mime_type: part.mediaType ?? null,
+            content: convertDataContentToBase64String(data as Uint8Array),
+          };
+        }
+        default:
+          return { type: String((part as { type: string }).type) };
+      }
+    });
+    return { role: 'user', parts };
+  }
+
+  if (msg.role === 'assistant') {
+    if (typeof msg.content === 'string') {
+      return {
+        role: 'assistant',
+        parts: [{ type: 'text', content: msg.content }],
+      };
+    }
+    const parts: SemConvPart[] = msg.content.map(part => {
+      switch (part.type) {
+        case 'text':
+          return { type: 'text' as const, content: part.text };
+        case 'reasoning':
+          return { type: 'reasoning' as const, content: part.text };
+        case 'tool-call':
+          return {
+            type: 'tool_call' as const,
+            id: part.toolCallId ?? null,
+            name: part.toolName,
+            arguments: part.input,
+          };
+        case 'tool-result': {
+          const output = part.output;
+          let response: unknown;
+          if (output) {
+            if (output.type === 'text' || output.type === 'error-text') {
+              response = output.value;
+            } else if (output.type === 'json' || output.type === 'error-json') {
+              response = output.value;
+            } else if (output.type === 'execution-denied') {
+              response = { denied: true, reason: output.reason };
+            } else {
+              response = output;
+            }
+          }
+          return {
+            type: 'tool_call_response' as const,
+            id: part.toolCallId ?? null,
+            response,
+          };
+        }
+        default:
+          return { type: String((part as { type: string }).type) };
+      }
+    });
+    return { role: 'assistant', parts };
+  }
+
+  if (msg.role === 'tool') {
+    const parts: SemConvPart[] = msg.content.map(part => {
+      if (part.type === 'tool-result') {
+        const output = part.output;
+        let response: unknown;
+        if (output) {
+          if (output.type === 'text' || output.type === 'error-text') {
+            response = output.value;
+          } else if (output.type === 'json' || output.type === 'error-json') {
+            response = output.value;
+          } else if (output.type === 'execution-denied') {
+            response = { denied: true, reason: output.reason };
+          } else {
+            response = output;
+          }
+        }
+        return {
+          type: 'tool_call_response' as const,
+          id: part.toolCallId ?? null,
+          response,
+        };
+      }
+      return { type: String((part as { type: string }).type) };
+    });
+    return { role: 'tool', parts };
+  }
+
+  return undefined;
+}
+
+/**
+ * Extracts the system instruction from a LanguageModelV4Prompt if present.
+ */
+export function extractSystemFromPrompt(
+  prompt: LanguageModelV4Prompt,
+): string | undefined {
+  const systemMsg = prompt.find(msg => msg.role === 'system');
+  if (systemMsg && systemMsg.role === 'system') {
+    return systemMsg.content;
+  }
+  return undefined;
+}
+
+/**
+ * Converts step result data to the gen_ai.output.messages SemConv format.
+ */
+export function formatOutputMessages({
+  text,
+  reasoning,
+  toolCalls,
+  files,
+  finishReason,
+}: {
+  text?: string;
+  reasoning?: ReadonlyArray<{ text?: string }>;
+  toolCalls?: ReadonlyArray<{
+    toolCallId: string;
+    toolName: string;
+    input: unknown;
+  }>;
+  files?: ReadonlyArray<{ mediaType: string; base64: string }>;
+  finishReason: string;
+}): SemConvOutputMessage[] {
+  const parts: SemConvPart[] = [];
+
+  if (reasoning) {
+    for (const r of reasoning) {
+      if ('text' in r && r.text) {
+        parts.push({ type: 'reasoning', content: r.text });
+      }
+    }
+  }
+
+  if (text != null && text.length > 0) {
+    parts.push({ type: 'text', content: text });
+  }
+
+  if (toolCalls) {
+    for (const tc of toolCalls) {
+      parts.push({
+        type: 'tool_call',
+        id: tc.toolCallId,
+        name: tc.toolName,
+        arguments: tc.input,
+      });
+    }
+  }
+
+  if (files) {
+    for (const file of files) {
+      parts.push({
+        type: 'blob',
+        modality: getModality(file.mediaType),
+        mime_type: file.mediaType,
+        content: file.base64,
+      });
+    }
+  }
+
+  return [
+    {
+      role: 'assistant',
+      parts,
+      finish_reason: mapFinishReason(finishReason),
+    },
+  ];
+}
+
+/**
+ * Converts generateObject result to the gen_ai.output.messages SemConv format.
+ */
+export function formatObjectOutputMessages({
+  objectText,
+  finishReason,
+}: {
+  objectText: string;
+  finishReason: string;
+}): SemConvOutputMessage[] {
+  return [
+    {
+      role: 'assistant',
+      parts: [{ type: 'text', content: objectText }],
+      finish_reason: mapFinishReason(finishReason),
+    },
+  ];
+}
+
+function mapFinishReason(reason: string): string {
+  const mapping: Record<string, string> = {
+    stop: 'stop',
+    length: 'length',
+    'content-filter': 'content_filter',
+    'tool-calls': 'tool_call',
+    error: 'error',
+    other: 'stop',
+    unknown: 'stop',
+  };
+  return mapping[reason] ?? reason;
+}