@codehz/ai 0.1.8 → 0.2.1

package/src/helpers/usage-mapping.ts

@@ -1,7 +1,7 @@
 /**
  * Provider usage → canonical Usage 映射
  *
- * best-effort 提取 reasoning / cache / billable 等扩展字段。
+ * best-effort 提取 reasoning / cache 等扩展字段。
  */
 
 import type { Usage } from "../types/index.js";
@@ -20,27 +20,6 @@ function record(obj: Record<string, number | undefined>): Partial<Usage> {
   return out;
 }
 
-function billableFromOpenAIStyle(
-  inputTokens: number | undefined,
-  outputTokens: number | undefined,
-  cachedInputTokens: number | undefined,
-  reasoningTokens: number | undefined,
-): Pick<Usage, "billableInputTokens" | "billableOutputTokens"> {
-  let billableInputTokens: number | undefined;
-  if (inputTokens !== undefined) {
-    billableInputTokens =
-      cachedInputTokens !== undefined ? Math.max(0, inputTokens - cachedInputTokens) : inputTokens;
-  }
-
-  let billableOutputTokens: number | undefined;
-  if (outputTokens !== undefined) {
-    billableOutputTokens =
-      reasoningTokens !== undefined ? Math.max(0, outputTokens - reasoningTokens) : outputTokens;
-  }
-
-  return record({ billableInputTokens, billableOutputTokens });
-}
-
 /** OpenAI Chat Completions `usage` */
 export function usageFromChatCompletions(raw: {
   prompt_tokens?: number;
@@ -63,7 +42,6 @@ export function usageFromChatCompletions(raw: {
     totalTokens,
     cachedInputTokens,
     reasoningTokens,
-    ...billableFromOpenAIStyle(inputTokens, outputTokens, cachedInputTokens, reasoningTokens),
   });
 }
 
@@ -90,7 +68,6 @@ export function usageFromOpenAIResponses(raw: {
     totalTokens,
     cachedInputTokens,
     reasoningTokens,
-    ...billableFromOpenAIStyle(inputTokens, outputTokens, cachedInputTokens, reasoningTokens),
   });
 }
 
@@ -111,13 +88,7 @@ export function usageFromAnthropicMessages(raw: {
     (n): n is number => n !== undefined,
   );
   const inputTokens = inputParts.length > 0 ? inputParts.reduce((sum, n) => sum + n, 0) : undefined;
-  const totalTokens =
-    inputTokens !== undefined && outputTokens !== undefined ? inputTokens + outputTokens : undefined;
-
-  let billableInputTokens: number | undefined;
-  if (uncachedInputTokens !== undefined || cacheWriteInputTokens !== undefined) {
-    billableInputTokens = (uncachedInputTokens ?? 0) + (cacheWriteInputTokens ?? 0);
-  }
+  const totalTokens = inputTokens !== undefined && outputTokens !== undefined ? inputTokens + outputTokens : undefined;
 
   return record({
     inputTokens,
@@ -125,26 +96,18 @@ export function usageFromAnthropicMessages(raw: {
     totalTokens,
     cachedInputTokens,
     cacheWriteInputTokens,
-    billableInputTokens,
-    billableOutputTokens: outputTokens,
   });
 }
 
-/** Ollama 流式 chunk（无 cache / reasoning 细分时仅填基础与 billable 镜像） */
-export function usageFromOllama(raw: {
-  prompt_eval_count?: number;
-  eval_count?: number;
-}): Partial<Usage> {
+/** Ollama 流式 chunk */
+export function usageFromOllama(raw: { prompt_eval_count?: number; eval_count?: number }): Partial<Usage> {
   const inputTokens = num(raw.prompt_eval_count);
   const outputTokens = num(raw.eval_count);
-  const totalTokens =
-    inputTokens !== undefined && outputTokens !== undefined ? inputTokens + outputTokens : undefined;
+  const totalTokens = inputTokens !== undefined && outputTokens !== undefined ? inputTokens + outputTokens : undefined;
 
   return record({
     inputTokens,
     outputTokens,
     totalTokens,
-    billableInputTokens: inputTokens,
-    billableOutputTokens: outputTokens,
   });
-}
+}

package/src/types/adapter.ts

@@ -21,9 +21,20 @@ export type NormalizedRequest = AIRequest & {
 
 // ── Adapter 接口 ──────────────────────────────────────────────
 
+export type StreamingCapability = "native" | "synthetic" | "none";
+
+export type AdapterCapabilities = {
+  readonly textStreaming: StreamingCapability;
+  readonly reasoningStreaming: StreamingCapability;
+  readonly toolCallStreaming: StreamingCapability;
+  readonly replay: "canonical" | "opaque" | "none";
+  readonly usage: "stream" | "final" | "none";
+  readonly toolResultOutcomes: ReadonlyArray<"success" | "error" | "rejected">;
+};
+
 export interface BackendAdapter {
   readonly kind: "chat-completions" | "messages" | "responses" | "ollama" | "mock";
-  readonly nativeStreaming: boolean;
+  readonly capabilities: AdapterCapabilities;
   stream(request: NormalizedRequest): AsyncIterable<AIStreamEvent>;
 }
 

package/src/types/events.ts

@@ -7,8 +7,8 @@
  */
 
 import type { ContentBlock } from "./content.js";
-import type { MessageItem, ReasoningItem, ToolCallItem } from "./items.js";
-import type { AIResponse, Usage, BillingInfo, AuxiliaryInfo } from "./response.js";
+import type { Usage, BillingInfo, AuxiliaryInfo, BackendTrace, StopReason } from "./response.js";
+import type { OpaqueItem, ReplayItem } from "./items.js";
 
 // ── 事件基类 ──────────────────────────────────────────────────
 
@@ -45,7 +45,14 @@ export type ResponseAuxiliaryEvent = StreamEventBase & {
 
 export type ResponseCompletedEvent = StreamEventBase & {
   type: "response.completed";
-  response: AIResponse;
+  replay: ReplayItem[];
+  stopReason?: StopReason;
+  usage?: Usage;
+  billing?: BillingInfo;
+  auxiliary?: AuxiliaryInfo;
+  warnings?: string[];
+  opaqueOutput?: OpaqueItem[];
+  trace?: Partial<BackendTrace>;
 };
 
 // ── 消息流事件 ────────────────────────────────────────────────
@@ -61,15 +68,12 @@ export type MessageStartedEvent = StreamEventBase & {
 export type MessageDeltaEvent = StreamEventBase & {
   type: "message.delta";
   itemId: string;
-  delta: {
-    type: "text";
-    text: string;
-  };
+  delta: ContentBlock;
 };
 
 export type MessageCompletedEvent = StreamEventBase & {
   type: "message.completed";
-  item: MessageItem;
+  itemId: string;
 };
 
 // ── 思维链流事件 ──────────────────────────────────────────────
@@ -90,7 +94,7 @@ export type ReasoningDeltaEvent = StreamEventBase & {
 
 export type ReasoningCompletedEvent = StreamEventBase & {
   type: "reasoning.completed";
-  item: ReasoningItem;
+  itemId: string;
 };
 
 // ── 工具调用流事件 ────────────────────────────────────────────
@@ -113,7 +117,7 @@ export type ToolCallDeltaEvent = StreamEventBase & {
 
 export type ToolCallCompletedEvent = StreamEventBase & {
   type: "tool_call.completed";
-  item: ToolCallItem;
+  itemId: string;
 };
 
 // ── 统一事件联合 ──────────────────────────────────────────────

package/src/types/index.ts

@@ -46,4 +46,12 @@ export type {
 } from "./events.js";
 
 // Adapter 协议和 client 类型
-export type { BackendAdapter, FetchFn, NormalizedRequest, CreateAIClientOptions, AIClient } from "./adapter.js";
+export type {
+  AdapterCapabilities,
+  StreamingCapability,
+  BackendAdapter,
+  FetchFn,
+  NormalizedRequest,
+  CreateAIClientOptions,
+  AIClient,
+} from "./adapter.js";

package/src/types/response.ts

@@ -24,10 +24,6 @@ export type Usage = {
   cachedInputTokens?: number;
   /** Tokens written to prompt cache (Anthropic cache_creation_input_tokens) */
   cacheWriteInputTokens?: number;
-  /** Best-effort billable input (full-rate input; excludes discounted cache reads where known) */
-  billableInputTokens?: number;
-  /** Best-effort billable output (non-reasoning slice when provider gives reasoning breakdown) */
-  billableOutputTokens?: number;
 };
 
 export type BillingInfo = {