@tool-bridge/server 0.21.1 → 0.22.1

package/dist/perplexity-NVOBYN24.js

@@ -0,0 +1,220 @@
+import {
+  createAuthedClient
+} from "./chunk-RTW5VRTK.js";
+import "./chunk-BNMLA2RC.js";
+import "./chunk-JXHMDBAY.js";
+import {
+  createProviderPlugin,
+  upstreamError
+} from "./chunk-X7ZUTKHR.js";
+import "./chunk-7ISGFRC6.js";
+import "./chunk-UY6ANCOY.js";
+import {
+  TBError
+} from "./chunk-HA4BGL6C.js";
+import "./chunk-U67V476Y.js";
+
+// ../plugins/src/perplexity/api.ts
+var SERVICE = "perplexity";
+var API_BASE = "https://api.perplexity.ai";
+var http = createAuthedClient({
+  baseUrl: API_BASE,
+  service: SERVICE,
+  auth: { kind: "bearer" },
+  headers: { accept: "application/json" },
+  errorMessage: {
+    keys: ["error.message", "error.detail", "message", "detail"],
+    fallback: (status) => `perplexity request failed with ${status}`
+  }
+});
+var EMBEDDING_MIN_DIMENSIONS = 128;
+var EMBEDDING_MAX_DIMENSIONS = { "pplx-embed-v1-0.6b": 1024 };
+var EMBEDDING_MAX_DIMENSIONS_FALLBACK = 2560;
+async function request(ctx, method, path, body) {
+  const { data } = await http.request(ctx, {
+    path,
+    method,
+    // 上游先过一道 `compactObject`;JSON.stringify 本来就丢 undefined 值,故不再重复一遍。
+    ...method === "POST" ? { json: body ?? {} } : {},
+    invalidJsonMessage: "Perplexity \u8FD4\u56DE\u4E86\u975E JSON \u54CD\u5E94",
+    // 消息里带着本次的 method/path,只能逐请求给,不能挪到 client 级。
+    mapTransportError: ({ message }) => upstreamError(
+      502,
+      `perplexity ${method} ${path} failed before receiving response: ${message ?? "unknown network error"}`
+    )
+  });
+  return data ?? null;
+}
+function assertStreamingDisabled(input) {
+  if (input.stream === true) {
+    throw new TBError("invalid_argument", "stream=true is not supported by connector actions");
+  }
+}
+function assertEmbeddingDimensions(input) {
+  const dimensions = input.dimensions;
+  if (dimensions === void 0) return;
+  const max = EMBEDDING_MAX_DIMENSIONS[input.model] ?? EMBEDDING_MAX_DIMENSIONS_FALLBACK;
+  if (dimensions < EMBEDDING_MIN_DIMENSIONS || dimensions > max) {
+    throw new TBError(
+      "invalid_argument",
+      `${input.model} dimensions must be between ${EMBEDDING_MIN_DIMENSIONS} and ${max}`
+    );
+  }
+}
+async function listModels(_input, ctx) {
+  return request(ctx, "GET", "/v1/models");
+}
+async function search(input, ctx) {
+  return request(ctx, "POST", "/search", input);
+}
+async function createChatCompletion(input, ctx) {
+  assertStreamingDisabled(input);
+  return request(ctx, "POST", "/v1/sonar", input);
+}
+async function createEmbeddings(input, ctx) {
+  assertEmbeddingDimensions(input);
+  return request(ctx, "POST", "/v1/embeddings", input);
+}
+
+// ../plugins/src/perplexity/schema.ts
+import { z } from "zod/v4";
+var listModelsInput = z.strictObject({}).describe("No input parameters are required for this action.");
+var listModelsOutput = z.looseObject({
+  object: z.string().describe("The top-level object type.").optional(),
+  data: z.array(z.looseObject({
+    id: z.string().describe("The model identifier."),
+    object: z.string().describe("The object type returned by the API.").optional(),
+    created: z.int().describe("The Unix timestamp when the model metadata was created.").optional(),
+    owned_by: z.string().describe("The organization or owner that provides the model.").optional(),
+    type: z.string().describe("The model family or type.").optional()
+  }).describe("A Perplexity model entry.")).describe("The list of available models.")
+}).describe("The response payload for listing Perplexity models.");
+var searchInput = z.strictObject({
+  query: z.union([z.string().min(1).describe("One or more raw web search queries."), z.array(z.string().min(1).describe("One raw web search query.")).min(1).describe("One raw web search query.")]).describe("One or more raw web search queries."),
+  country: z.string().min(2).max(2).describe("The ISO 3166-1 alpha-2 country code used to localize search results.").optional(),
+  max_results: z.int().min(1).max(20).describe("The maximum number of search results to return.").optional(),
+  search_after_date: z.string().describe("Only return content published after this date, for example 01/15/2024.").optional(),
+  search_before_date: z.string().describe("Only return content published before this date, for example 12/31/2024.").optional(),
+  max_tokens_per_page: z.int().min(1).describe("The maximum number of tokens to retrieve from each webpage.").optional(),
+  search_domain_filter: z.array(z.string().min(1).describe("A domain or URL prefix used to filter results.")).max(20).describe("Only return search results from these domains or URL prefixes.").optional()
+}).describe("The input payload for the raw Perplexity web search action.");
+var searchOutput = z.looseObject({
+  results: z.array(z.looseObject({
+    title: z.string().describe("The search result title.").optional(),
+    url: z.string().describe("The canonical URL of the result.").optional(),
+    snippet: z.string().describe("A short snippet extracted from the result page.").optional(),
+    date: z.string().describe("The published date reported for the result.").optional(),
+    last_updated: z.string().describe("The last updated date reported for the result.").optional()
+  }).describe("A raw search result returned by Perplexity.")).describe("The ranked raw search results.").optional()
+}).describe("The response payload for the raw search action.");
+var createChatCompletionInput = z.strictObject({
+  model: z.string().describe("The Sonar model to use for the chat completion."),
+  messages: z.array(z.strictObject({
+    role: z.enum(["system", "user", "assistant"]).describe("The role of the message author.").optional(),
+    content: z.string().describe("The plain-text message content.").optional()
+  }).describe("A chat message.")).min(1).describe("The ordered conversation messages."),
+  max_tokens: z.int().min(1).describe("The maximum number of tokens to generate.").optional(),
+  temperature: z.number().min(0).max(2).describe("The sampling temperature for generation.").optional(),
+  top_p: z.number().min(0).max(1).describe("The nucleus sampling threshold.").optional(),
+  top_k: z.int().min(0).describe("The number of top tokens to consider per step.").optional(),
+  stream: z.boolean().describe("Whether to stream the response. This connector only accepts false or an omitted value.").optional(),
+  return_images: z.boolean().describe("Whether to include image references in the response.").optional(),
+  return_citations: z.boolean().describe("Whether to include citations in the response.").optional(),
+  disable_search: z.boolean().describe("Whether to disable web search grounding for this request.").optional(),
+  search_domain_filter: z.array(z.string().min(1).describe("A domain or URL prefix used to limit web search.")).describe("Only search within these domains or URL prefixes.").optional(),
+  search_recency_filter: z.string().describe("A recency filter such as day, week, month, or year.").optional(),
+  presence_penalty: z.number().min(-2).max(2).describe("The penalty applied to tokens that have already appeared.").optional(),
+  frequency_penalty: z.number().min(0).max(2).describe("The penalty applied based on token frequency.").optional()
+}).describe("The input payload for the Perplexity chat completion action.");
+var createChatCompletionOutput = z.looseObject({
+  id: z.string().describe("The completion identifier.").optional(),
+  model: z.string().describe("The model that generated the completion.").optional(),
+  created: z.int().describe("The Unix timestamp when the completion was created.").optional(),
+  choices: z.array(z.looseObject({
+    index: z.int().describe("The choice index.").optional(),
+    finish_reason: z.string().describe("The reason why generation stopped.").optional(),
+    message: z.looseObject({
+      role: z.string().describe("The role of the generated message.").optional(),
+      content: z.string().describe("The generated message content.").optional()
+    }).describe("The generated message.").optional()
+  }).describe("A single generated completion choice.")).describe("The completion choices."),
+  citations: z.array(z.string().min(1).describe("A citation URL returned with the response.")).describe("The citation URLs referenced by the answer.").optional(),
+  search_results: z.array(z.looseObject({
+    title: z.string().describe("The search result title.").optional(),
+    url: z.string().describe("The canonical URL of the result.").optional(),
+    snippet: z.string().describe("A short snippet extracted from the result page.").optional(),
+    date: z.string().describe("The published date reported for the result.").optional(),
+    last_updated: z.string().describe("The last updated date reported for the result.").optional()
+  }).describe("A raw search result returned by Perplexity.")).describe("The search results used to ground the response.").optional(),
+  usage: z.looseObject({
+    prompt_tokens: z.int().describe("The number of prompt tokens consumed.").optional(),
+    completion_tokens: z.int().describe("The number of completion tokens generated.").optional(),
+    total_tokens: z.int().describe("The total number of tokens consumed.").optional()
+  }).describe("Token usage metadata for the completion.").optional()
+}).describe("The response payload for the chat completion action.");
+var createEmbeddingsInput = z.strictObject({
+  model: z.enum(["pplx-embed-v1-0.6b", "pplx-embed-v1-4b"]).describe("The embedding model identifier."),
+  input: z.union([z.string().min(1).describe("One or more input strings to embed."), z.array(z.string().min(1).describe("One input string to embed.")).min(1).max(512).describe("One input string to embed.")]).describe("One or more input strings to embed."),
+  dimensions: z.int().describe("The target embedding dimension count.").optional(),
+  encoding_format: z.enum(["base64_int8", "base64_binary"]).describe("The output encoding format for embeddings.").optional()
+}).describe("The input payload for the embeddings action.");
+var createEmbeddingsOutput = z.looseObject({
+  object: z.string().describe("The top-level object type.").optional(),
+  model: z.string().describe("The embedding model that generated the output.").optional(),
+  data: z.array(z.looseObject({
+    object: z.string().describe("The object type for the embedding item.").optional(),
+    index: z.int().describe("The zero-based embedding index.").optional(),
+    embedding: z.union([z.array(z.number().describe("One embedding vector value.")).describe("The numeric embedding vector."), z.string().describe("The encoded embedding value.")]).describe("The embedding payload.").optional()
+  }).describe("A single embedding item.")).describe("The generated embeddings."),
+  usage: z.looseObject({
+    prompt_tokens: z.int().describe("The number of prompt tokens consumed.").optional(),
+    total_tokens: z.int().describe("The total number of tokens consumed.").optional()
+  }).describe("Usage information for the embeddings request.").optional()
+}).describe("The response payload for the embeddings action.");
+var perplexityActions = {
+  list_models: {
+    description: "List the models currently available from Perplexity.",
+    effect: "read",
+    inputSchema: listModelsInput,
+    outputSchema: z.toJSONSchema(listModelsOutput, { io: "output", unrepresentable: "any" })
+  },
+  search: {
+    description: "Search the web and return ranked raw results from Perplexity without LLM synthesis.",
+    effect: "write",
+    inputSchema: searchInput,
+    outputSchema: z.toJSONSchema(searchOutput, { io: "output", unrepresentable: "any" })
+  },
+  create_chat_completion: {
+    description: "Create a Perplexity Sonar chat completion grounded by web search when enabled.",
+    effect: "write",
+    inputSchema: createChatCompletionInput,
+    outputSchema: z.toJSONSchema(createChatCompletionOutput, { io: "output", unrepresentable: "any" })
+  },
+  create_embeddings: {
+    description: "Generate vector embeddings for one or more input strings with Perplexity.",
+    effect: "write",
+    inputSchema: createEmbeddingsInput,
+    outputSchema: z.toJSONSchema(createEmbeddingsOutput, { io: "output", unrepresentable: "any" })
+  }
+};
+
+// ../plugins/src/perplexity/index.ts
+function createPerplexityPlugin() {
+  return createProviderPlugin({
+    description: "Perplexity",
+    actions: perplexityActions,
+    // 上游 credentialValidators 也打 /v1/models —— 只读、无必填入参,正好当挂载时的凭证探针。
+    credentialProbe: "list_models",
+    handlers: {
+      list_models: listModels,
+      search,
+      create_chat_completion: createChatCompletion,
+      create_embeddings: createEmbeddings
+    }
+  });
+}
+var perplexity_default = createPerplexityPlugin();
+export {
+  createPerplexityPlugin,
+  perplexity_default as default
+};