kopia lustrzana https://github.com/transitive-bullshit/chatgpt-api
pull/643/head^2
Travis Fischer
rodzic
ca9254d567
commit
7731ca09bc
|
|
@ -1,135 +1,160 @@
|
|||
import defaultKy, { type KyInstance } from 'ky'
|
||||
import { z } from 'zod'
|
||||
|
||||
import { aiFunction, AIFunctionsProvider } from '../fns.js'
|
||||
import { assert, getEnv } from '../utils.js'
|
||||
|
||||
export namespace exa {
|
||||
/**
|
||||
* Search options for performing a search query.
|
||||
*/
|
||||
export type BaseSearchOptions = {
|
||||
/** Number of search results to return. Default 10. Max 10 for basic plans. */
|
||||
numResults?: number
|
||||
/** List of domains to include in the search. */
|
||||
includeDomains?: string[]
|
||||
/** List of domains to exclude in the search. */
|
||||
excludeDomains?: string[]
|
||||
/** Start date for results based on crawl date. */
|
||||
startCrawlDate?: string
|
||||
/** End date for results based on crawl date. */
|
||||
endCrawlDate?: string
|
||||
/** Start date for results based on published date. */
|
||||
startPublishedDate?: string
|
||||
/** End date for results based on published date. */
|
||||
endPublishedDate?: string
|
||||
/** A data category to focus on, with higher comprehensivity and data cleanliness. Currently, the only category is company. */
|
||||
category?: string
|
||||
}
|
||||
export const TextContentsOptionsSchema = z.object({
|
||||
maxCharacters: z
|
||||
.number()
|
||||
.optional()
|
||||
.describe('The maximum number of characters to return.'),
|
||||
includeHtmlTags: z
|
||||
.boolean()
|
||||
.optional()
|
||||
.describe('If true, includes HTML tags in the returned text.')
|
||||
})
|
||||
export type TextContentsOptions = z.infer<typeof TextContentsOptionsSchema>
|
||||
|
||||
/**
|
||||
* Search options for performing a search query.
|
||||
*/
|
||||
export type RegularSearchOptions = BaseSearchOptions & {
|
||||
/** If true, converts query to a Metaphor query. */
|
||||
useAutoprompt?: boolean
|
||||
/** Type of search, 'keyword' or 'neural'. */
|
||||
type?: string
|
||||
}
|
||||
|
||||
/**
|
||||
* Options for finding similar links.
|
||||
*/
|
||||
export type FindSimilarOptions = BaseSearchOptions & {
|
||||
/** If true, excludes links from the base domain of the input. */
|
||||
excludeSourceDomain?: boolean
|
||||
}
|
||||
|
||||
/**
|
||||
* Search options for performing a search query.
|
||||
*/
|
||||
export type ContentsOptions = {
|
||||
/** Options for retrieving text contents. */
|
||||
text?: TextContentsOptions | true
|
||||
/** Options for retrieving highlights. */
|
||||
highlights?: HighlightsContentsOptions | true
|
||||
}
|
||||
|
||||
/**
|
||||
* Options for retrieving text from page.
|
||||
*/
|
||||
export type TextContentsOptions = {
|
||||
/** The maximum number of characters to return. */
|
||||
maxCharacters?: number
|
||||
/** If true, includes HTML tags in the returned text. Default: false */
|
||||
includeHtmlTags?: boolean
|
||||
}
|
||||
|
||||
/**
|
||||
* Options for retrieving highlights from page.
|
||||
* @typedef {Object} HighlightsContentsOptions
|
||||
*/
|
||||
export type HighlightsContentsOptions = {
|
||||
/** The query string to use for highlights search. */
|
||||
query?: string
|
||||
/** The number of sentences to return for each highlight. */
|
||||
numSentences?: number
|
||||
/** The number of highlights to return for each URL. */
|
||||
highlightsPerUrl?: number
|
||||
}
|
||||
|
||||
export type TextResponse = {
|
||||
/** Text from page */
|
||||
text: string
|
||||
}
|
||||
|
||||
export type HighlightsResponse = {
|
||||
/** The highlights as an array of strings. */
|
||||
highlights: string[]
|
||||
/** The corresponding scores as an array of floats, 0 to 1 */
|
||||
highlightScores: number[]
|
||||
}
|
||||
|
||||
export type Default<T extends {}, U> = [keyof T] extends [never] ? U : T
|
||||
|
||||
/**
|
||||
* Depending on 'ContentsOptions', this yields either a 'TextResponse',
|
||||
* a 'HighlightsResponse', both, or an empty object.
|
||||
*/
|
||||
export type ContentsResultComponent<T extends ContentsOptions> = Default<
|
||||
(T['text'] extends object | true ? TextResponse : {}) &
|
||||
(T['highlights'] extends object | true ? HighlightsResponse : {}),
|
||||
TextResponse
|
||||
export const HighlightsContentsOptionsSchema = z.object({
|
||||
query: z
|
||||
.string()
|
||||
.optional()
|
||||
.describe('The query string to use for highlights search.'),
|
||||
numSentences: z
|
||||
.number()
|
||||
.optional()
|
||||
.describe('The number of sentences to return for each highlight.'),
|
||||
highlightsPerUrl: z
|
||||
.number()
|
||||
.optional()
|
||||
.describe('The number of highlights to return for each URL.')
|
||||
})
|
||||
export type HighlightsContentsOptions = z.infer<
|
||||
typeof HighlightsContentsOptionsSchema
|
||||
>
|
||||
|
||||
export const ContentsOptionsSchema = z.object({
|
||||
text: z.union([TextContentsOptionsSchema, z.literal(true)]).optional(),
|
||||
highlights: z
|
||||
.union([HighlightsContentsOptionsSchema, z.literal(true)])
|
||||
.optional()
|
||||
})
|
||||
export type ContentsOptions = z.infer<typeof ContentsOptionsSchema>
|
||||
|
||||
export const BaseSearchOptionsSchema = z.object({
|
||||
numResults: z
|
||||
.number()
|
||||
.optional()
|
||||
.describe('Number of search results to return.'),
|
||||
includeDomains: z
|
||||
.array(z.string())
|
||||
.optional()
|
||||
.describe('List of domains to include in the search.'),
|
||||
excludeDomains: z
|
||||
.array(z.string())
|
||||
.optional()
|
||||
.describe('List of domains to exclude from the search.'),
|
||||
startCrawlDate: z
|
||||
.string()
|
||||
.optional()
|
||||
.describe('Start date for results based on crawl date.'),
|
||||
endCrawlDate: z
|
||||
.string()
|
||||
.optional()
|
||||
.describe('End date for results based on crawl date.'),
|
||||
startPublishedDate: z
|
||||
.string()
|
||||
.optional()
|
||||
.describe('Start date for results based on published date.'),
|
||||
endPublishedDate: z
|
||||
.string()
|
||||
.optional()
|
||||
.describe('End date for results based on published date.'),
|
||||
category: z
|
||||
.string()
|
||||
.optional()
|
||||
.describe(
|
||||
'A data category to focus on, with higher comprehensivity and data cleanliness. Currently, the only category is company.'
|
||||
),
|
||||
contents: ContentsOptionsSchema.optional().describe(
|
||||
'Whether to include the contents of the search results.'
|
||||
)
|
||||
})
|
||||
export type BaseSearchOptions = z.infer<typeof BaseSearchOptionsSchema>
|
||||
|
||||
export const RegularSearchOptionsSchema = BaseSearchOptionsSchema.extend({
|
||||
query: z.string().describe('search query'),
|
||||
useAutoprompt: z.boolean().optional(),
|
||||
type: z.enum(['keyword', 'neural', 'magic']).optional()
|
||||
})
|
||||
export type RegularSearchOptions = z.infer<typeof RegularSearchOptionsSchema>
|
||||
|
||||
export const FindSimilarOptionsSchema = BaseSearchOptionsSchema.extend({
|
||||
url: z
|
||||
.string()
|
||||
.describe('The url for which you would like to find similar links'),
|
||||
excludeSourceDomain: z
|
||||
.boolean()
|
||||
.optional()
|
||||
.describe('If true, excludes links from the base domain of the input.')
|
||||
})
|
||||
export type FindSimilarOptions = z.infer<typeof FindSimilarOptionsSchema>
|
||||
|
||||
export const GetContentsOptionsSchema = ContentsOptionsSchema.extend({
|
||||
ids: z
|
||||
.array(z.string())
|
||||
.nonempty()
|
||||
.describe('Exa IDs of the documents to retrieve.')
|
||||
})
|
||||
export type GetContentsOptions = z.infer<typeof GetContentsOptionsSchema>
|
||||
|
||||
/**
|
||||
* Represents a search result object.
|
||||
*/
|
||||
export type SearchResult<T extends ContentsOptions = ContentsOptions> = {
|
||||
export type SearchResult = {
|
||||
/** The title of the search result. */
|
||||
title: string | null
|
||||
|
||||
/** The URL of the search result. */
|
||||
url: string
|
||||
|
||||
/** The estimated creation date of the content. */
|
||||
publishedDate?: string
|
||||
|
||||
/** The author of the content, if available. */
|
||||
author?: string
|
||||
|
||||
/** Similarity score between the query/url and the result. */
|
||||
score?: number
|
||||
/** The temporary ID for the document. */
|
||||
|
||||
/** The temporary Exa ID for the document. */
|
||||
id: string
|
||||
} & ContentsResultComponent<T>
|
||||
|
||||
/** Text from page */
|
||||
text?: string
|
||||
|
||||
/** The highlights as an array of strings. */
|
||||
highlights?: string[]
|
||||
|
||||
/** The corresponding scores as an array of floats, 0 to 1 */
|
||||
highlightScores?: number[]
|
||||
}
|
||||
|
||||
/**
|
||||
* Represents a search response object.
|
||||
*/
|
||||
export type SearchResponse<T extends ContentsOptions = ContentsOptions> = {
|
||||
export type SearchResponse = {
|
||||
/** The list of search results. */
|
||||
results: SearchResult<T>[]
|
||||
results: SearchResult[]
|
||||
|
||||
/** The autoprompt string, if applicable. */
|
||||
autopromptString?: string
|
||||
}
|
||||
}
|
||||
|
||||
export class ExaClient {
|
||||
export class ExaClient extends AIFunctionsProvider {
|
||||
readonly apiKey: string
|
||||
readonly apiBaseUrl: string
|
||||
readonly ky: KyInstance
|
||||
|
|
@ -147,6 +172,7 @@ export class ExaClient {
|
|||
apiKey,
|
||||
'ExaClient missing required "apiKey" (defaults to "EXA_API_KEY")'
|
||||
)
|
||||
super()
|
||||
|
||||
this.apiKey = apiKey
|
||||
this.apiBaseUrl = apiBaseUrl
|
||||
|
|
@ -162,103 +188,52 @@ export class ExaClient {
|
|||
/**
|
||||
* Performs an Exa search for the given query.
|
||||
*/
|
||||
async search(opts: { query: string } & exa.RegularSearchOptions) {
|
||||
return this.ky.get('search', { json: opts }).json<exa.SearchResponse>()
|
||||
}
|
||||
@aiFunction({
|
||||
name: 'exa_search',
|
||||
description: 'Search the web for the given query.',
|
||||
inputSchema: exa.RegularSearchOptionsSchema
|
||||
})
|
||||
async search(queryOrOpts: string | exa.RegularSearchOptions) {
|
||||
const json =
|
||||
typeof queryOrOpts === 'string' ? { query: queryOrOpts } : queryOrOpts
|
||||
|
||||
/**
|
||||
* Performs a search with a Exa prompt-engineered query and returns the
|
||||
* contents of the documents.
|
||||
*/
|
||||
async searchAndContents<T extends exa.ContentsOptions = exa.ContentsOptions>({
|
||||
query,
|
||||
text,
|
||||
highlights,
|
||||
...rest
|
||||
}: { query: string } & exa.RegularSearchOptions & T) {
|
||||
return this.ky
|
||||
.post('search', {
|
||||
json: {
|
||||
query,
|
||||
contents:
|
||||
!text && !highlights
|
||||
? { text: true }
|
||||
: {
|
||||
...(text ? { text } : {}),
|
||||
...(highlights ? { highlights } : {})
|
||||
},
|
||||
...rest
|
||||
}
|
||||
})
|
||||
.json<exa.SearchResponse<T>>()
|
||||
return this.ky.post('search', { json }).json<exa.SearchResponse>()
|
||||
}
|
||||
|
||||
/**
|
||||
* Finds similar links to the provided URL.
|
||||
*/
|
||||
async findSimilar(opts: { url: string } & exa.FindSimilarOptions) {
|
||||
@aiFunction({
|
||||
name: 'exa_find_similar',
|
||||
description: 'Find similar links to the provided URL.',
|
||||
inputSchema: exa.FindSimilarOptionsSchema
|
||||
})
|
||||
async findSimilar(opts: exa.FindSimilarOptions) {
|
||||
return this.ky
|
||||
.post('findSimilar', { json: opts })
|
||||
.json<exa.SearchResponse>()
|
||||
}
|
||||
|
||||
/**
|
||||
* Finds similar links to the provided URL and returns the contents of the
|
||||
* documents.
|
||||
* Retrieves contents of documents based on a list of Exa document IDs.
|
||||
*/
|
||||
async findSimilarAndContents<
|
||||
T extends exa.ContentsOptions = exa.ContentsOptions
|
||||
>({
|
||||
url,
|
||||
text,
|
||||
highlights,
|
||||
...rest
|
||||
}: { url: string } & exa.FindSimilarOptions & T) {
|
||||
return this.ky
|
||||
.post('findSimilar', {
|
||||
json: {
|
||||
url,
|
||||
contents:
|
||||
!text && !highlights
|
||||
? { text: true }
|
||||
: {
|
||||
...(text ? { text } : {}),
|
||||
...(highlights ? { highlights } : {})
|
||||
},
|
||||
...rest
|
||||
}
|
||||
})
|
||||
.json<exa.SearchResponse<T>>()
|
||||
}
|
||||
|
||||
/**
|
||||
* Retrieves contents of documents based on a list of document IDs.
|
||||
*/
|
||||
async getContents<T extends exa.ContentsOptions = exa.ContentsOptions>({
|
||||
ids,
|
||||
...opts
|
||||
}: { ids: string | string[] | exa.SearchResult[] } & T) {
|
||||
let requestIds: string[]
|
||||
|
||||
if (typeof ids === 'string') {
|
||||
requestIds = [ids]
|
||||
} else if (typeof ids[0] === 'string') {
|
||||
requestIds = ids as string[]
|
||||
} else {
|
||||
requestIds = (ids as exa.SearchResult[]).map((result) => result.id)
|
||||
}
|
||||
|
||||
if (ids.length === 0) {
|
||||
throw new Error('Must provide at least one ID')
|
||||
}
|
||||
@aiFunction({
|
||||
name: 'exa_get_contents',
|
||||
description:
|
||||
'Retrieve contents of documents based on a list of Exa document IDs.',
|
||||
inputSchema: exa.GetContentsOptionsSchema
|
||||
})
|
||||
async getContents({ ids, ...opts }: exa.GetContentsOptions) {
|
||||
const documentIDs = Array.isArray(ids) ? ids : [ids]
|
||||
assert(documentIDs.length, 'Must provide at least one document ID')
|
||||
|
||||
return this.ky
|
||||
.post('contents', {
|
||||
json: {
|
||||
...opts,
|
||||
ids: requestIds
|
||||
ids: documentIDs
|
||||
}
|
||||
})
|
||||
.json<exa.SearchResponse<T>>()
|
||||
.json<exa.SearchResponse>()
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -5,7 +5,8 @@ import { aiFunction, AIFunctionsProvider } from '../fns.js'
|
|||
import { assert, delay, getEnv } from '../utils.js'
|
||||
import { zodToJsonSchema } from '../zod-to-json-schema.js'
|
||||
|
||||
// TODO: Deprioritizing this client for now because the API doesn't seem to be stable.
|
||||
// TODO: Deprioritizing this client for now because the API doesn't seem to be
|
||||
// stable.
|
||||
|
||||
export namespace firecrawl {
|
||||
/**
|
||||
|
|
|
|||
Ładowanie…
Reference in New Issue