@juvio15/pi-cliproxyapi-provider 0.0.0-stage → 0.1.0

package/src/web-search/search.ts

@@ -0,0 +1,571 @@
+import { calculateCost } from "@earendil-works/pi-ai";
+import type { Api, Model, Usage } from "@earendil-works/pi-ai";
+
+import { resolveResponsesEndpoint } from "./endpoint.ts";
+import { consumeSseStream } from "./sse.ts";
+import type { SseEvent } from "./sse.ts";
+
+export interface WebSearchSource {
+  url: string;
+  title?: string;
+}
+
+export type WebSearchProgressPhase =
+  | "starting"
+  | "searching"
+  | "analyzing"
+  | "generating"
+  | "done";
+
+export interface WebSearchProgressEvent {
+  type: WebSearchProgressPhase;
+  phase: WebSearchProgressPhase;
+  message: string;
+  model: string;
+  query: string;
+  searchQuery?: string;
+  chars?: number;
+  sourcesCount?: number;
+  durationMs?: number;
+}
+
+export interface RunWebSearchOptions {
+  /** CLIProxyAPI root, `/v1` base URL, or full `/v1/responses` URL. */
+  responsesUrl: string;
+  apiKey?: string;
+  /** A model id, or a pi-ai Model when usage cost should be priced. */
+  model: string | Model<Api>;
+  query: string;
+  signal?: AbortSignal;
+  onProgress?: (event: WebSearchProgressEvent) => void;
+  referenceDate?: Date | string;
+  fetchFn?: typeof fetch;
+}
+
+export interface WebSearchResult {
+  answer: string;
+  sources: WebSearchSource[];
+  usage?: Usage;
+}
+
+interface ResponsesSourcePayload {
+  title?: string;
+  url?: string;
+}
+
+interface ResponsesContentPayload {
+  annotations?: ResponsesSourcePayload[];
+  text?: string;
+  type?: string;
+}
+
+interface ResponsesItemPayload {
+  action?: {
+    query?: string;
+    sources?: ResponsesSourcePayload[];
+  };
+  content?: ResponsesContentPayload[];
+  type?: string;
+}
+
+interface ResponsesUsagePayload {
+  input_tokens?: number;
+  input_tokens_details?: {
+    cache_write_tokens?: number;
+    cached_tokens?: number;
+  };
+  output_tokens?: number;
+  output_tokens_details?: {
+    reasoning_tokens?: number;
+  };
+  total_tokens?: number;
+}
+
+interface ResponsesPayload {
+  error?: {
+    message?: string;
+  };
+  message?: string;
+  output?: ResponsesItemPayload[];
+  usage?: ResponsesUsagePayload;
+}
+
+interface ResponsesEventPayload extends ResponsesPayload {
+  delta?: string;
+  error?: {
+    message?: string;
+  };
+  item?: ResponsesItemPayload;
+  message?: string;
+  response?: ResponsesPayload;
+  text?: string;
+  type?: string;
+}
+
+interface SearchStreamState {
+  completedMessageText: string;
+  lastDeltaUpdateTime: number;
+  lastPhase: string;
+  rawUsage?: ResponsesUsagePayload;
+  sources: Map<string, string>;
+  streamedMessageText: string;
+  streamedText: string;
+}
+
+type ProgressEmitter = (
+  phase: WebSearchProgressPhase,
+  message: string,
+  extra?: Partial<WebSearchProgressEvent>
+) => void;
+
+export const SEARCH_INSTRUCTIONS =
+  "You are a web search assistant. Search the web to find accurate, up-to-date information for the user's query. Provide a comprehensive summary with all relevant details, numbers, dates, and sources.";
+
+export const buildSearchInstructions = (
+  referenceDate: Date | string = new Date()
+): string => {
+  const date =
+    referenceDate instanceof Date ? referenceDate : new Date(referenceDate);
+
+  const dateStr = Number.isNaN(date.getTime())
+    ? String(referenceDate).slice(0, 10)
+    : date.toISOString().slice(0, 10);
+
+  return `${SEARCH_INSTRUCTIONS}\nCurrent reference date: ${dateStr}. Use this reference date to ground relative time expressions (such as today, yesterday, recent, latest, or this year) and ensure search queries and results remain up to date.`;
+};
+
+const extractErrorMessage = (
+  payload: ResponsesEventPayload,
+  defaultMessage: string
+): string => {
+  if (payload.error?.message?.trim()) {
+    return payload.error.message.trim();
+  }
+
+  if (payload.message?.trim()) {
+    return payload.message.trim();
+  }
+
+  return defaultMessage;
+};
+
+const isTokenCount = (value: number | undefined): value is number =>
+  value !== undefined && Number.isSafeInteger(value) && value >= 0;
+
+const toUsage = (
+  raw: ResponsesUsagePayload | undefined,
+  model: Model<Api> | undefined
+): Usage | undefined => {
+  if (!raw) {
+    return undefined;
+  }
+
+  const input = raw.input_tokens;
+  const output = raw.output_tokens;
+  const cacheRead = raw.input_tokens_details?.cached_tokens ?? 0;
+  const cacheWrite = raw.input_tokens_details?.cache_write_tokens ?? 0;
+  const reasoning = raw.output_tokens_details?.reasoning_tokens ?? 0;
+
+  const totalTokens =
+    raw.total_tokens ??
+    (isTokenCount(input) && isTokenCount(output) ? input + output : undefined);
+
+  if (
+    !isTokenCount(input) ||
+    !isTokenCount(output) ||
+    !isTokenCount(cacheRead) ||
+    !isTokenCount(cacheWrite) ||
+    !isTokenCount(reasoning) ||
+    !isTokenCount(totalTokens)
+  ) {
+    return undefined;
+  }
+
+  const usage: Usage = {
+    cacheRead,
+    cacheWrite,
+    cost: { cacheRead: 0, cacheWrite: 0, input: 0, output: 0, total: 0 },
+    input: Math.max(0, input - cacheRead - cacheWrite),
+    output,
+    reasoning,
+    totalTokens,
+  };
+
+  if (model) {
+    calculateCost(model, usage);
+  }
+
+  return usage;
+};
+
+const addSource = (
+  sources: Map<string, string>,
+  urlValue: string | undefined,
+  titleValue: string | undefined
+): void => {
+  if (!urlValue?.trim()) {
+    return;
+  }
+
+  const url = urlValue.trim();
+  const title = titleValue?.trim() ?? "";
+
+  if (!sources.has(url) || (!sources.get(url) && title)) {
+    sources.set(url, title);
+  }
+};
+
+const extractSourcesFromItem = (
+  item: ResponsesItemPayload,
+  sources: Map<string, string>
+): void => {
+  if (item.type === "web_search_call") {
+    for (const source of item.action?.sources ?? []) {
+      addSource(sources, source.url, source.title);
+    }
+  }
+
+  if (item.type === "message") {
+    for (const part of item.content ?? []) {
+      for (const annotation of part.annotations ?? []) {
+        addSource(sources, annotation.url, annotation.title);
+      }
+    }
+  }
+};
+
+const messageTextFromItem = (item: ResponsesItemPayload): string => {
+  if (item.type !== "message") {
+    return "";
+  }
+
+  const parts: string[] = [];
+
+  for (const part of item.content ?? []) {
+    if ((part.type === "output_text" || part.type === "text") && part.text) {
+      parts.push(part.text);
+    }
+  }
+
+  return parts.join("");
+};
+
+const lastMessageText = (output: ResponsesItemPayload[]): string => {
+  let text = "";
+
+  for (const item of output) {
+    const candidate = messageTextFromItem(item);
+
+    if (candidate.trim()) {
+      text = candidate;
+    }
+  }
+
+  return text;
+};
+
+const handleLifecycleEvent = (
+  eventType: string,
+  state: SearchStreamState,
+  emit: ProgressEmitter,
+  query: string
+): boolean => {
+  if (
+    eventType === "response.web_search_call.searching" ||
+    eventType === "response.web_search_call.in_progress"
+  ) {
+    if (state.lastPhase !== "searching") {
+      state.lastPhase = "searching";
+      emit("searching", `Searching the web for ${query}...`);
+    }
+
+    return true;
+  }
+
+  if (eventType === "response.web_search_call.completed") {
+    if (state.lastPhase !== "analyzing") {
+      state.lastPhase = "analyzing";
+      emit("analyzing", "Analyzing search results...");
+    }
+
+    return true;
+  }
+
+  return false;
+};
+
+const handleTextEvent = (
+  eventType: string,
+  payload: ResponsesEventPayload,
+  state: SearchStreamState,
+  emit: ProgressEmitter
+): boolean => {
+  if (eventType === "response.output_text.delta") {
+    if (payload.delta) {
+      state.streamedText += payload.delta;
+      const now = Date.now();
+
+      if (now - state.lastDeltaUpdateTime >= 100) {
+        state.lastDeltaUpdateTime = now;
+        emit(
+          "generating",
+          `Summarizing results (${state.streamedText.length} chars)...`,
+          {
+            chars: state.streamedText.length,
+          }
+        );
+      }
+    }
+
+    return true;
+  }
+
+  if (eventType === "response.output_text.done") {
+    if (payload.text && !state.streamedText) {
+      state.streamedText = payload.text;
+    }
+
+    return true;
+  }
+
+  return false;
+};
+
+const handleItemEvent = (
+  eventType: string,
+  payload: ResponsesEventPayload,
+  state: SearchStreamState,
+  emit: ProgressEmitter
+): boolean => {
+  if (
+    eventType !== "response.output_item.done" &&
+    eventType !== "response.output_item.added"
+  ) {
+    return false;
+  }
+
+  if (!payload.item) {
+    return true;
+  }
+
+  extractSourcesFromItem(payload.item, state.sources);
+
+  const searchQuery = payload.item.action?.query;
+
+  if (
+    payload.item.type === "web_search_call" &&
+    searchQuery &&
+    state.lastPhase !== "searching"
+  ) {
+    state.lastPhase = "searching";
+    emit("searching", `Searching the web for ${searchQuery}...`, {
+      searchQuery,
+    });
+  }
+
+  const text = messageTextFromItem(payload.item);
+
+  if (text.trim()) {
+    state.streamedMessageText = text;
+  }
+
+  return true;
+};
+
+const handleCompletedEvent = (
+  eventType: string,
+  payload: ResponsesEventPayload,
+  state: SearchStreamState
+): boolean => {
+  if (eventType !== "response.completed") {
+    return false;
+  }
+
+  const completed = payload.response ?? payload;
+  state.rawUsage = completed.usage;
+
+  for (const item of completed.output ?? []) {
+    extractSourcesFromItem(item, state.sources);
+  }
+
+  if (completed.output) {
+    // Fact 2: Kimi and Muse put the answer in a later message item.
+    state.completedMessageText = lastMessageText(completed.output);
+  }
+
+  return true;
+};
+
+const handleEvent = (
+  { event, data }: SseEvent,
+  state: SearchStreamState,
+  emit: ProgressEmitter,
+  query: string
+): void => {
+  // SAFETY: SSE JSON uses the Responses wire shape. Every payload field is
+  // optional in the local payload types and is checked before it is used.
+  const payload = data as ResponsesEventPayload;
+  const eventType = payload.type ?? event;
+
+  if (eventType === "response.failed" || eventType === "error") {
+    throw new Error(extractErrorMessage(payload, "Web search response failed"));
+  }
+
+  if (handleLifecycleEvent(eventType, state, emit, query)) {
+    return;
+  }
+
+  if (handleTextEvent(eventType, payload, state, emit)) {
+    return;
+  }
+
+  if (handleItemEvent(eventType, payload, state, emit)) {
+    return;
+  }
+
+  handleCompletedEvent(eventType, payload, state);
+};
+
+export const runWebSearch = async (
+  options: RunWebSearchOptions
+): Promise<WebSearchResult> => {
+  const {
+    responsesUrl,
+    apiKey,
+    model,
+    query,
+    signal,
+    onProgress,
+    referenceDate = new Date(),
+    fetchFn = fetch,
+  } = options;
+
+  const trimmedQuery = query.trim();
+
+  if (!trimmedQuery) {
+    throw new Error("web_search requires a non-empty query");
+  }
+
+  // oxlint-disable-next-line anti-slop/no-runtime-typeof -- The typed union is a model id string or a pi-ai Model, and typeof is the intended way to tell them apart.
+  const modelObject = typeof model === "object" ? model : undefined;
+
+  const modelId = modelObject ? modelObject.id : String(model).trim();
+
+  if (!modelId) {
+    throw new Error("web_search requires a model");
+  }
+
+  signal?.throwIfAborted();
+
+  const startTime = Date.now();
+
+  const emit: ProgressEmitter = (phase, message, extra = {}) => {
+    onProgress?.({
+      message,
+      model: modelId,
+      phase,
+      query: trimmedQuery,
+      type: phase,
+      ...extra,
+    });
+  };
+
+  emit("starting", `Starting web search (${modelId})...`);
+
+  const baseHeaders = {
+    accept: "text/event-stream",
+    "content-type": "application/json",
+  };
+
+  const headers = apiKey
+    ? { ...baseHeaders, authorization: `Bearer ${apiKey}` }
+    : baseHeaders;
+
+  const response = await fetchFn(resolveResponsesEndpoint(responsesUrl), {
+    body: JSON.stringify({
+      input: [
+        {
+          content: [{ text: trimmedQuery, type: "input_text" }],
+          role: "user",
+        },
+      ],
+      instructions: buildSearchInstructions(referenceDate),
+      model: modelId,
+      store: false,
+      stream: true,
+      tools: [{ type: "web_search" }],
+    }),
+    headers,
+    method: "POST",
+    signal,
+  });
+
+  if (!response.ok) {
+    let errorText = `HTTP ${response.status}`;
+
+    try {
+      const parsed: ResponsesEventPayload = JSON.parse(await response.text());
+      errorText = extractErrorMessage(parsed, errorText);
+    } catch {
+      // Keep the HTTP status fallback when the error body is not JSON.
+    }
+
+    throw new Error(`Web search failed: ${errorText}`);
+  }
+
+  if (!response.body) {
+    throw new Error("Web search streaming response has no body");
+  }
+
+  const state: SearchStreamState = {
+    completedMessageText: "",
+    lastDeltaUpdateTime: 0,
+    lastPhase: "",
+    sources: new Map<string, string>(),
+    streamedMessageText: "",
+    streamedText: "",
+  };
+
+  await consumeSseStream(
+    response.body,
+    (event) => handleEvent(event, state, emit, trimmedQuery),
+    signal
+  );
+  signal?.throwIfAborted();
+
+  const answer = (
+    state.completedMessageText ||
+    state.streamedMessageText ||
+    state.streamedText
+  ).trim();
+
+  if (!answer) {
+    throw new Error("Web search completed without any search result text");
+  }
+
+  const usage = toUsage(state.rawUsage, modelObject);
+  const sourceList: WebSearchSource[] = [];
+
+  for (const [url, title] of state.sources) {
+    const source: WebSearchSource = { url };
+
+    if (title) {
+      source.title = title;
+    }
+
+    sourceList.push(source);
+  }
+
+  emit("done", "Web search complete.", {
+    durationMs: Date.now() - startTime,
+    sourcesCount: sourceList.length,
+  });
+
+  const result: WebSearchResult = { answer, sources: sourceList };
+
+  if (usage) {
+    result.usage = usage;
+  }
+
+  return result;
+};

package/src/web-search/sse.ts

@@ -0,0 +1,144 @@
+export interface SseEvent {
+  event: string;
+  data: unknown;
+}
+
+export interface SseParser {
+  push: (chunk: string) => void;
+  end: () => void;
+}
+
+export const createSseParser = (
+  onEvent: (event: SseEvent) => void
+): SseParser => {
+  let buffer = "";
+  let eventName = "";
+  let dataLines: string[] = [];
+
+  const dispatch = () => {
+    if (dataLines.length === 0) {
+      eventName = "";
+
+      return;
+    }
+
+    const text = dataLines.join("\n");
+    let data: unknown = text;
+
+    if (text !== "[DONE]") {
+      try {
+        data = JSON.parse(text);
+      } catch {
+        data = text;
+      }
+    }
+
+    onEvent({ data, event: eventName || "message" });
+    eventName = "";
+    dataLines = [];
+  };
+
+  const processLine = (line: string) => {
+    if (line === "") {
+      dispatch();
+
+      return;
+    }
+
+    if (line.startsWith(":")) {
+      return;
+    }
+
+    const separator = line.indexOf(":");
+    const field = separator === -1 ? line : line.slice(0, separator);
+    let value = separator === -1 ? "" : line.slice(separator + 1);
+
+    if (value.startsWith(" ")) {
+      value = value.slice(1);
+    }
+
+    if (field === "event") {
+      eventName = value;
+    }
+
+    if (field === "data") {
+      dataLines.push(value);
+    }
+  };
+
+  return {
+    end: () => {
+      if (buffer.length > 0) {
+        let line = buffer;
+        buffer = "";
+
+        if (line.endsWith("\r")) {
+          line = line.slice(0, -1);
+        }
+
+        processLine(line);
+      }
+
+      dispatch();
+    },
+    push: (chunk: string) => {
+      buffer += chunk;
+      let newline = buffer.indexOf("\n");
+
+      while (newline !== -1) {
+        let line = buffer.slice(0, newline);
+        buffer = buffer.slice(newline + 1);
+
+        if (line.endsWith("\r")) {
+          line = line.slice(0, -1);
+        }
+
+        processLine(line);
+        newline = buffer.indexOf("\n");
+      }
+    },
+  };
+};
+
+export const consumeSseStream = async (
+  stream: ReadableStream<Uint8Array>,
+  onEvent: (event: SseEvent) => void,
+  signal?: AbortSignal
+): Promise<void> => {
+  const reader = stream.getReader();
+  const decoder = new TextDecoder();
+  const parser = createSseParser(onEvent);
+
+  const onAbort = () => {
+    void reader.cancel().catch(() => null);
+  };
+
+  signal?.throwIfAborted();
+  signal?.addEventListener("abort", onAbort, { once: true });
+
+  const readNext = async (): Promise<void> => {
+    signal?.throwIfAborted();
+
+    const { done, value } = await reader.read();
+
+    if (done) {
+      return;
+    }
+
+    if (value) {
+      parser.push(decoder.decode(value, { stream: true }));
+    }
+
+    await readNext();
+  };
+
+  try {
+    await readNext();
+    signal?.throwIfAborted();
+    parser.push(decoder.decode());
+    parser.end();
+  } finally {
+    signal?.removeEventListener("abort", onAbort);
+    reader.releaseLock();
+  }
+};

package/README.md

@@ -1,3 +0,0 @@
-# Temporary Holding Version
-
-This version is a temporary placeholder for this package. An operational version to replace this has been submitted for review and is awaiting a staged release.