veryfront 0.1.263 → 0.1.265

package/src/src/provider/runtime-loader/provider-usage.ts

@@ -0,0 +1,135 @@
+import { readRecord } from "./provider-records.js";
+
+export type RuntimeUsage = {
+  inputTokens?: number;
+  outputTokens?: number;
+  totalTokens?: number;
+  cacheCreationInputTokens?: number;
+  cacheReadInputTokens?: number;
+};
+
+export function extractAnthropicUsage(payload: unknown): RuntimeUsage | undefined {
+  const record = readRecord(payload);
+  const usage = readRecord(record?.usage);
+  if (!usage) {
+    return undefined;
+  }
+
+  const inputTokens = usage.input_tokens;
+  const outputTokens = usage.output_tokens;
+  const cacheCreationInputTokens = usage.cache_creation_input_tokens;
+  const cacheReadInputTokens = usage.cache_read_input_tokens;
+
+  return {
+    inputTokens: typeof inputTokens === "number" ? inputTokens : undefined,
+    outputTokens: typeof outputTokens === "number" ? outputTokens : undefined,
+    totalTokens: typeof inputTokens === "number" || typeof outputTokens === "number"
+      ? (typeof inputTokens === "number" ? inputTokens : 0) +
+        (typeof outputTokens === "number" ? outputTokens : 0)
+      : undefined,
+    ...(typeof cacheCreationInputTokens === "number" ? { cacheCreationInputTokens } : {}),
+    ...(typeof cacheReadInputTokens === "number" ? { cacheReadInputTokens } : {}),
+  };
+}
+
+export function extractGoogleUsage(payload: unknown): RuntimeUsage | undefined {
+  const record = readRecord(payload);
+  const usage = readRecord(record?.usageMetadata);
+  if (!usage) {
+    return undefined;
+  }
+
+  const inputTokens = usage.promptTokenCount;
+  const outputTokens = usage.candidatesTokenCount;
+  const totalTokens = usage.totalTokenCount;
+  const cachedContentTokenCount = usage.cachedContentTokenCount;
+
+  return {
+    inputTokens: typeof inputTokens === "number" ? inputTokens : undefined,
+    outputTokens: typeof outputTokens === "number" ? outputTokens : undefined,
+    totalTokens: typeof totalTokens === "number" ? totalTokens : undefined,
+    ...(typeof cachedContentTokenCount === "number"
+      ? { cacheReadInputTokens: cachedContentTokenCount }
+      : {}),
+  };
+}
+
+export function extractOpenAIUsage(payload: unknown): RuntimeUsage | undefined {
+  const record = readRecord(payload);
+  const usage = readRecord(record?.usage);
+  if (!usage) {
+    return undefined;
+  }
+
+  const inputTokens = usage.prompt_tokens;
+  const outputTokens = usage.completion_tokens;
+  const totalTokens = usage.total_tokens;
+  const promptTokensDetails = readRecord(usage.prompt_tokens_details);
+  const cachedTokens = promptTokensDetails?.cached_tokens;
+
+  return {
+    inputTokens: typeof inputTokens === "number" ? inputTokens : undefined,
+    outputTokens: typeof outputTokens === "number" ? outputTokens : undefined,
+    totalTokens: typeof totalTokens === "number" ? totalTokens : undefined,
+    ...(typeof cachedTokens === "number" ? { cacheReadInputTokens: cachedTokens } : {}),
+  };
+}
+
+/**
+ * The Responses API uses `input_tokens` / `output_tokens` field names
+ * instead of Chat Completions' `prompt_tokens` / `completion_tokens`.
+ * It also nests cached input tokens under `input_tokens_details` and
+ * exposes reasoning tokens via `output_tokens_details.reasoning_tokens`.
+ */
+export function extractOpenAIResponsesUsage(payload: unknown): RuntimeUsage | undefined {
+  const record = readRecord(payload);
+  // Streaming usage lives on response.completed inside `response.usage`;
+  // non-streaming has it at the top level.
+  const responseRecord = readRecord(record?.response);
+  const usage = readRecord(responseRecord?.usage) ?? readRecord(record?.usage);
+  if (!usage) return undefined;
+
+  const inputTokens = typeof usage.input_tokens === "number" ? usage.input_tokens : undefined;
+  const outputTokens = typeof usage.output_tokens === "number" ? usage.output_tokens : undefined;
+  const totalTokens = typeof usage.total_tokens === "number"
+    ? usage.total_tokens
+    : (inputTokens !== undefined || outputTokens !== undefined
+      ? (inputTokens ?? 0) + (outputTokens ?? 0)
+      : undefined);
+  const inputDetails = readRecord(usage.input_tokens_details);
+  const cachedTokens = inputDetails?.cached_tokens;
+
+  return {
+    inputTokens,
+    outputTokens,
+    totalTokens,
+    ...(typeof cachedTokens === "number" ? { cacheReadInputTokens: cachedTokens } : {}),
+  };
+}
+
+export function mergeUsage(
+  current: RuntimeUsage | undefined,
+  next: RuntimeUsage | undefined,
+): RuntimeUsage | undefined {
+  if (!current) {
+    return next;
+  }
+
+  if (!next) {
+    return current;
+  }
+
+  const inputTokens = next.inputTokens ?? current.inputTokens;
+  const outputTokens = next.outputTokens ?? current.outputTokens;
+  const cacheCreationInputTokens = next.cacheCreationInputTokens ??
+    current.cacheCreationInputTokens;
+  const cacheReadInputTokens = next.cacheReadInputTokens ?? current.cacheReadInputTokens;
+
+  return {
+    inputTokens,
+    outputTokens,
+    totalTokens: (inputTokens ?? 0) + (outputTokens ?? 0),
+    ...(cacheCreationInputTokens !== undefined ? { cacheCreationInputTokens } : {}),
+    ...(cacheReadInputTokens !== undefined ? { cacheReadInputTokens } : {}),
+  };
+}

package/src/src/provider/runtime-loader.ts

@@ -8,11 +8,32 @@ import {
   getOpenAIEmbeddingUrl,
   getOpenAIResponsesUrl,
 } from "./runtime-loader/provider-endpoints.js";
+import {
+  extractGoogleEmbedding,
+  extractGoogleUsageTokens,
+  extractOpenAIEmbeddings,
+  extractOpenAIUsageTokens,
+} from "./runtime-loader/provider-embedding-responses.js";
+import {
+  normalizeAnthropicFinishReason,
+  normalizeGoogleFinishReason,
+  normalizeOpenAIFinishReason,
+  normalizeOpenAIResponsesFinishReason,
+} from "./runtime-loader/provider-finish-reasons.js";
 import {
   createAnthropicRequestInit,
   createGoogleRequestInit,
   createOpenAIRequestInit,
 } from "./runtime-loader/provider-request-init.js";
+import { parseSseChunk } from "./runtime-loader/provider-sse.js";
+import {
+  extractAnthropicUsage,
+  extractGoogleUsage,
+  extractOpenAIResponsesUsage,
+  extractOpenAIUsage,
+  mergeUsage,
+  type RuntimeUsage,
+} from "./runtime-loader/provider-usage.js";
 import type { ProviderKind } from "./runtime-loader/provider-http.js";
 import { requestJson, requestStream } from "./runtime-loader/provider-http.js";
 import { readRecord } from "./runtime-loader/provider-records.js";
@@ -407,66 +428,6 @@ type GoogleCompatibleRequest = {
   [key: string]: unknown;
 };
 
-function isNumberArray(value: unknown): value is number[] {
-  return Array.isArray(value) && value.every((entry) => typeof entry === "number");
-}
-
-function extractOpenAIEmbeddings(payload: unknown): number[][] {
-  const record = readRecord(payload);
-  const data = record?.data;
-  if (!Array.isArray(data)) {
-    throw new Error("Invalid OpenAI embedding response: data array missing");
-  }
-
-  const embeddings: number[][] = [];
-
-  for (const item of data) {
-    const itemRecord = readRecord(item);
-    const embedding = itemRecord?.embedding;
-    if (!isNumberArray(embedding)) {
-      throw new Error("Invalid OpenAI embedding response: embedding vector missing");
-    }
-    embeddings.push(embedding);
-  }
-
-  return embeddings;
-}
-
-function extractOpenAIUsageTokens(payload: unknown): number | undefined {
-  const record = readRecord(payload);
-  const usage = readRecord(record?.usage);
-  const totalTokens = usage?.total_tokens;
-  return typeof totalTokens === "number" ? totalTokens : undefined;
-}
-
-function extractGoogleEmbedding(payload: unknown): number[] {
-  const record = readRecord(payload);
-  const embeddings = record?.embeddings;
-
-  if (Array.isArray(embeddings) && embeddings.length > 0) {
-    const firstEmbedding = readRecord(embeddings[0]);
-    const values = firstEmbedding?.values;
-    if (isNumberArray(values)) {
-      return values;
-    }
-  }
-
-  const embedding = readRecord(record?.embedding);
-  const values = embedding?.values;
-  if (isNumberArray(values)) {
-    return values;
-  }
-
-  throw new Error("Invalid Google embedding response: embedding vector missing");
-}
-
-function extractGoogleUsageTokens(payload: unknown): number | undefined {
-  const record = readRecord(payload);
-  const usageMetadata = readRecord(record?.usageMetadata);
-  const promptTokenCount = usageMetadata?.promptTokenCount;
-  return typeof promptTokenCount === "number" ? promptTokenCount : undefined;
-}
-
 /**
  * Structured warning emitted when a provider runtime drops or rewrites a
  * caller-provided option. Mirrors the AI ecosystem convention (Vercel AI
@@ -624,85 +585,6 @@ function readProviderOptions(
   return merged;
 }
 
-function normalizeAnthropicFinishReason(
-  raw: unknown,
-): string | { unified: string; raw: string } | null {
-  if (typeof raw !== "string") {
-    return null;
-  }
-
-  switch (raw) {
-    case "tool_use":
-      return { unified: "tool-calls", raw };
-    case "end_turn":
-    case "stop_sequence":
-      return { unified: "stop", raw };
-    case "max_tokens":
-      return { unified: "length", raw };
-    default:
-      return raw;
-  }
-}
-
-type RuntimeUsage = {
-  inputTokens?: number;
-  outputTokens?: number;
-  totalTokens?: number;
-  cacheCreationInputTokens?: number;
-  cacheReadInputTokens?: number;
-};
-
-function extractAnthropicUsage(payload: unknown): RuntimeUsage | undefined {
-  const record = readRecord(payload);
-  const usage = readRecord(record?.usage);
-  if (!usage) {
-    return undefined;
-  }
-
-  const inputTokens = usage.input_tokens;
-  const outputTokens = usage.output_tokens;
-  const cacheCreationInputTokens = usage.cache_creation_input_tokens;
-  const cacheReadInputTokens = usage.cache_read_input_tokens;
-
-  return {
-    inputTokens: typeof inputTokens === "number" ? inputTokens : undefined,
-    outputTokens: typeof outputTokens === "number" ? outputTokens : undefined,
-    totalTokens: typeof inputTokens === "number" || typeof outputTokens === "number"
-      ? (typeof inputTokens === "number" ? inputTokens : 0) +
-        (typeof outputTokens === "number" ? outputTokens : 0)
-      : undefined,
-    ...(typeof cacheCreationInputTokens === "number" ? { cacheCreationInputTokens } : {}),
-    ...(typeof cacheReadInputTokens === "number" ? { cacheReadInputTokens } : {}),
-  };
-}
-
-function mergeUsage(
-  current: RuntimeUsage | undefined,
-  next: RuntimeUsage | undefined,
-): RuntimeUsage | undefined {
-  if (!current) {
-    return next;
-  }
-
-  if (!next) {
-    return current;
-  }
-
-  const inputTokens = next.inputTokens ?? current.inputTokens;
-  const outputTokens = next.outputTokens ?? current.outputTokens;
-  const cacheCreationInputTokens = next.cacheCreationInputTokens ??
-    current.cacheCreationInputTokens;
-  const cacheReadInputTokens = next.cacheReadInputTokens ?? current.cacheReadInputTokens;
-
-  return {
-    inputTokens,
-    outputTokens,
-    totalTokens: (inputTokens ?? 0) + (outputTokens ?? 0),
-    ...(cacheCreationInputTokens !== undefined ? { cacheCreationInputTokens } : {}),
-    ...(cacheReadInputTokens !== undefined ? { cacheReadInputTokens } : {}),
-  };
-}
-
 function normalizeAnthropicToolChoice(toolChoice: unknown): unknown {
   if (typeof toolChoice === "string") {
     return { type: toolChoice };
@@ -1342,36 +1224,6 @@ function buildAnthropicGenerateResult(payload: unknown): {
   };
 }
 
-function parseSseChunk(chunk: string): {
-  events: Array<unknown | "[DONE]">;
-  remainder: string;
-} {
-  const blocks = chunk.split(/\r?\n\r?\n/);
-  const remainder = blocks.pop() ?? "";
-  const events = blocks.flatMap((block) => {
-    const dataLines = block.split(/\r?\n/)
-      .filter((line) => line.startsWith("data:"))
-      .map((line) => line.slice(5).trimStart());
-
-    if (!dataLines.length) {
-      return [];
-    }
-
-    const payload = dataLines.join("\n").trim();
-    if (payload === "[DONE]") {
-      return ["[DONE]" as const];
-    }
-
-    try {
-      return [JSON.parse(payload) as unknown];
-    } catch {
-      return [];
-    }
-  });
-
-  return { events, remainder };
-}
-
 async function* streamAnthropicCompatibleParts(
   stream: ReadableStream<Uint8Array>,
 ): AsyncIterable<unknown> {
@@ -1613,45 +1465,6 @@ async function* streamAnthropicCompatibleParts(
   };
 }
 
-function normalizeOpenAIFinishReason(
-  raw: unknown,
-): string | { unified: string; raw: string } | null {
-  if (typeof raw !== "string") {
-    return null;
-  }
-
-  if (raw === "tool_calls") {
-    return { unified: "tool-calls", raw };
-  }
-
-  if (raw === "content_filter") {
-    return { unified: "content-filter", raw };
-  }
-
-  return raw;
-}
-
-function extractOpenAIUsage(payload: unknown): RuntimeUsage | undefined {
-  const record = readRecord(payload);
-  const usage = readRecord(record?.usage);
-  if (!usage) {
-    return undefined;
-  }
-
-  const inputTokens = usage.prompt_tokens;
-  const outputTokens = usage.completion_tokens;
-  const totalTokens = usage.total_tokens;
-  const promptTokensDetails = readRecord(usage.prompt_tokens_details);
-  const cachedTokens = promptTokensDetails?.cached_tokens;
-
-  return {
-    inputTokens: typeof inputTokens === "number" ? inputTokens : undefined,
-    outputTokens: typeof outputTokens === "number" ? outputTokens : undefined,
-    totalTokens: typeof totalTokens === "number" ? totalTokens : undefined,
-    ...(typeof cachedTokens === "number" ? { cacheReadInputTokens: cachedTokens } : {}),
-  };
-}
-
 function extractOpenAIContentText(content: unknown): string {
   if (typeof content === "string") {
     return content;
@@ -1874,48 +1687,6 @@ function buildOpenAIChatRequest(
   return body;
 }
 
-function normalizeGoogleFinishReason(
-  raw: unknown,
-): string | { unified: string; raw: string } | null {
-  if (typeof raw !== "string") {
-    return null;
-  }
-
-  switch (raw) {
-    case "STOP":
-      return { unified: "stop", raw };
-    case "MAX_TOKENS":
-      return { unified: "length", raw };
-    case "SAFETY":
-    case "RECITATION":
-      return { unified: "content-filter", raw };
-    default:
-      return raw.toLowerCase();
-  }
-}
-
-function extractGoogleUsage(payload: unknown): RuntimeUsage | undefined {
-  const record = readRecord(payload);
-  const usage = readRecord(record?.usageMetadata);
-  if (!usage) {
-    return undefined;
-  }
-
-  const inputTokens = usage.promptTokenCount;
-  const outputTokens = usage.candidatesTokenCount;
-  const totalTokens = usage.totalTokenCount;
-  const cachedContentTokenCount = usage.cachedContentTokenCount;
-
-  return {
-    inputTokens: typeof inputTokens === "number" ? inputTokens : undefined,
-    outputTokens: typeof outputTokens === "number" ? outputTokens : undefined,
-    totalTokens: typeof totalTokens === "number" ? totalTokens : undefined,
-    ...(typeof cachedContentTokenCount === "number"
-      ? { cacheReadInputTokens: cachedContentTokenCount }
-      : {}),
-  };
-}
-
 function toGoogleContents(
   prompt: RuntimePromptMessage[],
 ): {
@@ -2941,56 +2712,6 @@ function buildOpenAIResponsesRequest(
   return body;
 }
 
-/**
- * The Responses API uses `input_tokens` / `output_tokens` field names
- * instead of Chat Completions' `prompt_tokens` / `completion_tokens`.
- * It also nests cached input tokens under `input_tokens_details` and
- * exposes reasoning tokens via `output_tokens_details.reasoning_tokens`.
- */
-function extractOpenAIResponsesUsage(payload: unknown): RuntimeUsage | undefined {
-  const record = readRecord(payload);
-  // Streaming usage lives on response.completed inside `response.usage`;
-  // non-streaming has it at the top level.
-  const responseRecord = readRecord(record?.response);
-  const usage = readRecord(responseRecord?.usage) ?? readRecord(record?.usage);
-  if (!usage) return undefined;
-
-  const inputTokens = typeof usage.input_tokens === "number" ? usage.input_tokens : undefined;
-  const outputTokens = typeof usage.output_tokens === "number" ? usage.output_tokens : undefined;
-  const totalTokens = typeof usage.total_tokens === "number"
-    ? usage.total_tokens
-    : (inputTokens !== undefined || outputTokens !== undefined
-      ? (inputTokens ?? 0) + (outputTokens ?? 0)
-      : undefined);
-  const inputDetails = readRecord(usage.input_tokens_details);
-  const cachedTokens = inputDetails?.cached_tokens;
-
-  return {
-    inputTokens,
-    outputTokens,
-    totalTokens,
-    ...(typeof cachedTokens === "number" ? { cacheReadInputTokens: cachedTokens } : {}),
-  };
-}
-
-function normalizeOpenAIResponsesFinishReason(
-  raw: unknown,
-): string | { unified: string; raw: string } | null {
-  if (typeof raw !== "string") return null;
-  switch (raw) {
-    case "completed":
-      return { unified: "stop", raw };
-    case "incomplete":
-      return { unified: "length", raw };
-    case "failed":
-      return { unified: "error", raw };
-    case "in_progress":
-      return null;
-    default:
-      return raw;
-  }
-}
-
 type OpenAIResponsesContentPart =
   | { type: "text"; text: string }
   | {

package/src/src/utils/version-constant.ts

@@ -1,3 +1,3 @@
 // Keep in sync with deno.json version.
 // scripts/release.ts updates this constant during releases.
-export const VERSION = "0.1.263";
+export const VERSION = "0.1.265";