Merge branch 'main' of https://github.com/NotePlan/plugins

helpers/general.js

@@ -329,11 +329,11 @@ export function createPrettyOpenNoteLink(linkText: string, titleOrFilename: stri
  * @param {string} linkText - the text to display for the link
  * @param {string} pluginID - the plugin's ID
  * @param {boolean} command - the "name" field of the plugin command to run
- * @param {Array<string> | string} args - heading inside of note (optional)
+ * @param {any} args - arguments to pass (optional)
  * @returns {string} arguments as strings (or single argument string) to send through to plugin
  * @tests available
  */
-export function createPrettyRunPluginLink(linkText: string, pluginID: string, command: string, args: Array<string> | string): string {
+export function createPrettyRunPluginLink(linkText: string, pluginID: string, command: string, args: any): string {
   return `[${linkText}](${createRunPluginCallbackUrl(pluginID, command, args)})`
 }
 

helpers/openAI.js

@@ -0,0 +1,353 @@
+// @flow
+
+const pluginJson = '@helpers/openAI.js'
+import { log, logError, logWarn, logDebug, timer, clo, JSP } from '@helpers/dev'
+import { getSettings } from '@helpers/NPConfiguration'
+import { getInput, showMessage } from '@helpers/userInput'
+
+// @flow
+
+/****************************************************************************************************************************
+ *                             CONSTANTS
+ ****************************************************************************************************************************/
+
+const BASE_URL = 'https://api.openai.com/v1'
+const TOKEN_LIMIT = 3000 // tokens per request (actually 3072)
+const MAX_RETRIES = 5 // number of times to retry a request if it fails
+const CHAT_COMPONENT = 'chat/completions'
+const MODEL_COST = { 'gpt-4': { inputCost: 0.03 / 1000, outputCost: 0.06 / 1000 }, 'gpt-3.5-turbo': { inputCost: 0.0015 / 1000, outputCost: 0.002 / 1000 } }
+
+/****************************************************************************************************************************
+ *                             TYPES
+ ****************************************************************************************************************************/
+
+// export type DallERequestOptions = { prompt?: string, n?: number, size?: string, response_format?: string, user?: string }
+export type CompletionsRequest = { model: string, prompt?: string, max_tokens?: number, user?: string, suffix?: string, temperature?: string, top_p?: string, n?: number }
+export type ResearchListResult = { initialQuery: string, currentQuery: string, selection?: string, options?: [string] }
+
+export type JSONClickData = { unclickedLinks: Array<string>, clickedLinks: Array<string>, initialSubject: string, remixes: Array<string>, totalTokensUsed: number }
+
+export type ChatMode = 'insert' | 'new_document' | 'return'
+
+export type ChatReturn = { question: string, prompt: string, answer: string }
+
+export type ChatResponse = {
+  error?: {
+    message: string,
+    type: string,
+    param: string,
+    code: string,
+  },
+  id: string,
+  object: string,
+  created: number,
+  model: string,
+  usage: {
+    prompt_tokens: number,
+    completion_tokens: number,
+    total_tokens: number,
+  },
+  choices: Array<{
+    message: {
+      role: string,
+      content: string,
+    },
+    finish_reason: string,
+    index: number,
+  }>,
+}
+
+// https://platform.openai.com/docs/api-reference/completions
+export type ChatObject = {
+  model: string /* currently only gpt-3.5-turbo is supported */,
+  messages: Array<{
+    role: 'system' | 'user' | 'assistant',
+    content: string,
+  }>,
+  suffix?: string,
+  temperature?: number,
+  max_tokens?: number,
+  top_p?: number,
+  presence_penalty?: number,
+  frequency_penalty?: number,
+  best_of?: number,
+  n?: number,
+  stream?: boolean,
+  logprobs?: number,
+  echo?: boolean,
+}
+
+/****************************************************************************************************************************
+ *                             LOCAL FUNCTIONS
+ ****************************************************************************************************************************/
+
+/**
+ * Make a request to the GPT API
+ * @param {string} method - GET, POST, PUT, etc.
+ * @param {any} body - JSON or null
+ * @returns {any|null} JSON results or null
+ */
+export async function getFetchRequestObj(method: string = 'GET', body: any = null): any {
+  const apiKey = await getOpenAIKey()
+  if (apiKey?.length) {
+    const obj = {
+      method,
+      headers: {
+        'Content-Type': 'application/json',
+        Authorization: `Bearer ${apiKey}`,
+      },
+    }
+    if (body && method !== 'GET') {
+      // logDebug(pluginJson, `getFetchRequestObj body type was: ${typeof body}`)
+      // $FlowFixMe
+      obj.body = typeof body === 'object' ? JSON.stringify(body) : body
+    }
+    // clo(obj, 'getFetchRequestObj request object is:')
+    return obj
+  } else {
+    showMessage('Please set your API key in the plugin settings')
+    logError(pluginJson, 'No API Key found')
+    return null
+  }
+}
+
+/**
+ * Get a human-readable error string to display to the user
+ * @param {Object} resultJSON - the JSON returned by the API
+ * @returns a string to display to the user
+ */
+function getErrorStringToDisplay(resultJSON: any): string {
+  const open = `OpenAI sent back an error message:\n"${resultJSON?.error?.message || ''}"`
+  let middle = ''
+  switch (resultJSON?.error?.code) {
+    case 'insufficient_quota':
+      middle = `\n\nDo you have a current credit card on your OpenAI account? If not, you will need to add one. Using OpenAI (chatGPT/DALL-E etc.) is quite inexpensive, but you do need a credit card on file.`
+      break
+    case 'too_many_requests':
+      middle = `\n\nYou have may have made too many API calls in a short period of time or their servers are overloaded.`
+      break
+    case 'invalid_api_key':
+      middle = `\n\nYou need to put a valid OpenAI API key in the plugin preferences for these commands to work properly. Please check your API key on OpenAI's website or create a new one.`
+      break
+    case 'invalid_request_error':
+      middle = `\n\nThe request sent to OpenAI was invalid. This may be a bug in the plugin. Please report it.`
+      break
+  }
+  const close = `\n\nPlease correct the error and try again.`
+  return `${open}${middle}${close}`
+}
+
+/**
+ * Count the number of tokens in a string (words + newlines)
+ * @param {string} inputString
+ * @returns {number} number of tokens in the string
+ */
+export function countTokens(inputString: string): number {
+  const words = inputString.trim().split(' ')
+  let count = 0
+
+  words.forEach((word) => {
+    count += word.split('\n').length
+  })
+
+  return count
+}
+
+/**
+ * Test if a string is too long for the API
+ * @param {string} string
+ * @param {boolean} shouldShowMessage - show a message if the string is too long (default: false)
+ * @returns {boolean} true if the string is too long, false if it is not
+ */
+export async function testTOKEN_LIMIT(str: string, shouldShowMessage: boolean = false): Promise<boolean> {
+  const tokens = countTokens(str)
+  if (tokens > TOKEN_LIMIT) {
+    if (shouldShowMessage) {
+      const message = `The string you entered is ${tokens} tokens long (including history). OpenAI's API has an approx limit of ${TOKEN_LIMIT} tokens. It may get rejected, but we will try it and see.`
+      await showMessage(message, 'error')
+    }
+    return false
+  }
+  return true
+}
+
+function logUsage(request: ChatObject, response: ChatResponse) {
+  const { model } = request
+  const { usage } = response
+  if (MODEL_COST[model]) {
+    const { prompt_tokens, completion_tokens } = usage
+    const { inputCost, outputCost } = MODEL_COST[model]
+    const inputCostStr = (inputCost * prompt_tokens).toFixed(4)
+    const outputCostStr = (outputCost * completion_tokens).toFixed(4)
+    const totalCostStr = (inputCost * prompt_tokens + outputCost * completion_tokens).toFixed(4)
+    logDebug(pluginJson, `outputUsage: inputCost=$${inputCostStr} outputCost=$${outputCostStr} totalCost=$${totalCostStr}`)
+  }
+}
+
+/****************************************************************************************************************************
+ *                             EXPORTED FUNCTIONS
+ ****************************************************************************************************************************/
+
+/**
+ * Create an initial chat request object with the starting system message
+ * (contains no input from the user)
+ * @param {string|null} model - the model to use (default: gpt-3.5-turbo)
+ * @param {string} systemMessage - the initial system message (e.g. "You are a helpful assistant")
+ * @returns
+ */
+export function createInitialChatObject(model: string = 'gpt-3.5-turbo', systemMessage: string): ChatObject {
+  const { chatModel } = DataStore.settings
+  return {
+    model: chatModel || model,
+    messages: [{ role: 'system', content: systemMessage }],
+  }
+}
+
+/**
+ * Create a new chat request object with the starting system message and the user's initial request text
+ * Uses the model specified in the plugin settings (DataStore.settings.chatModel)
+ * @param {string} systemMessage  - the initial system message (e.g. "You are a helpful assistant")
+ * @param {string} initialUserPrompt - the user's initial request text
+ * @param {string} modelName - the model to use (otherwise pulls from DataStore.settings.chatModel)
+ */
+export function newChatObject(systemMessage: string, initialUserPrompt: string, modelName?: string): ChatObject | null {
+  const chatModel = modelName ?? DataStore.settings.chatModel
+
+  if (!chatModel) {
+    logError(pluginJson, `newChatObject: chatModel is null. Stopping`)
+    return null
+  }
+  const request: ChatObject = createInitialChatObject(chatModel, systemMessage)
+  if (request) {
+    request.messages.push({ role: 'user', content: initialUserPrompt })
+  }
+  return request
+}
+
+/**
+ * Look in various places for the OpenAI API key. If we don't find it, ask the user for it.
+ * @returns {string} api key
+ */
+export async function getOpenAIKey(): Promise<string | null> {
+  // first check if key is in preferences
+  let key = DataStore.preference('openAIKey')
+  if (!key) {
+    logDebug(pluginJson, `No OpenAI key in DataStore.preference`)
+    // next check the running plugin's settings
+    key = DataStore.settings?.apiKey
+
+    if (!key) {
+      logDebug(pluginJson, `No OpenAI key in this plugin's settings`)
+      // next check AI plugin
+      const settings = await getSettings('shared.AI', null)
+      if (settings) {
+        key = settings.apiKey
+      }
+      if (!key) {
+        logDebug(pluginJson, `No OpenAI key in shared.AI plugin`)
+        // finally, ask user
+        key = await getInput('OpenAI API Key', 'OK', 'Enter Key', '')
+      }
+      if (!key) {
+        logError(pluginJson, `Tried 3x to get API Key but was set to null`)
+      }
+    }
+  }
+  if (key && key.length) {
+    // validate key?
+    DataStore.setPreference('openAIKey', key)
+  }
+  return key ? String(key).trim() : null
+}
+
+/**
+ * Make a one-shot chat request to the ChatGPT API
+ * Use for one-off requests, or to begin a conversation
+ * Returns object of the request and the response (so it can be used and saved for caching)
+ * @returns {Promise<{request: ChatObject, response: ChatResponse | null}} API result JSON response or null
+ */
+export async function makeOneShotChatRequest(SYSTEM_MESSAGE: string, userPrompt: string, model?: string): Promise<{ request: ChatObject, response: ChatResponse | null }> {
+  const request = newChatObject(SYSTEM_MESSAGE, userPrompt, model)
+  // clo(request, `makeOneShotChatObject: request=`)
+  const response = await makeRequest(CHAT_COMPONENT, 'POST', request)
+  return { request, response }
+}
+
+/**
+ * Make a request to the GPT API
+ * @param {string} component - the last part of the URL (after the base URL), e.g. "models" or "images/generations"
+ * @param {string} requestType - GET, POST, PUT, etc.
+ * @param {string} data - body of a POST/PUT request - can be an object (it will get stringified)
+ * @param {number} retry - number of times through the retry loop (you don't need to set this)
+ * @returns {any|null} JSON results or null
+ */
+export async function makeRequest(component: string, requestType: string = 'GET', requestBody: any = null, retry: number = 0): any | null {
+  const timesRetried = retry + 1
+  const url = `${BASE_URL}/${component}`
+  // logDebug(pluginJson, `makeRequest: about to fetch ${url}`)
+  // clo(data, `makeRequest() about to send to: "${url}" data=`)
+  await testTOKEN_LIMIT(JSON.stringify(requestBody), true)
+  const requestObj = await getFetchRequestObj(requestType, requestBody)
+  if (!requestObj) {
+    await showMessage('There was an error getting the request object. Check the plugin log and please report this issue.')
+    return null
+  }
+  // clo(requestObj, `makeRequest() about to fetch ${url} requestObj=`)
+  const result = await fetch(url, requestObj)
+  if (result) {
+    // clo(result, `makeRequest() result of fetch to: "${url}" response is type: ${typeof result} and value:`)
+    const resultJSON = JSON.parse(result)
+    if (!resultJSON || resultJSON?.error) {
+      const msg = resultJSON ? getErrorStringToDisplay(resultJSON) : `No response from OpenAI. Check log.`
+      await showMessage(msg)
+      clo(resultJSON?.error || {}, `askNewQuestion: Error:`)
+      return { error: { message: msg } }
+    }
+    // clo(resultJSON, `makeRequest() result of fetch to: "${url}" response is type: ${typeof resultJSON} and value:`)
+    if (resultJSON?.choices && resultJSON.choices[0] && resultJSON.choices[0]['finish_reason'] === 'length') {
+      resultJSON.choices[0].message += '... [ChatGPT truncated due to length, consider following up with "please continue"]'
+      logWarn(pluginJson, `makeRequest: ChatGPT truncated due to length, consider following up with "please continue"`)
+    }
+    logUsage(requestBody, resultJSON)
+    return resultJSON
+  } else {
+    // must have timed out/failed, let's try again
+    logWarn(pluginJson, `makeRequest failed on try: ${timesRetried}. Will retry.`)
+    while (timesRetried < MAX_RETRIES) {
+      const result = await makeRequest(component, requestType, requestBody, timesRetried)
+      if (result) {
+        return result
+      }
+    }
+    const failMsg = `Call to OpenAI failed after ${MAX_RETRIES} attempts. This may be a temporary problem (sometimes their servers are overloaded, or maybe you're offline?). Please try again, but report the problem if it persists.`
+    await showMessage(failMsg)
+    throw failMsg
+  }
+}
+
+/****************************************************************************************************************************
+ *                             DEBUGGING
+ ****************************************************************************************************************************/
+
+/**
+ * If the user has enabled saving responses, save the response to the DataStore
+ * Only saves if there is a setting in the running plugin's settings { saveResponses: true }
+ * @param {string} folderName
+ * @param {string} filename - the note filename (which will be based on the question but perhaps shortened by NP)
+ * @param {ChatObject} request
+ * @param {ChatResponse} chatResponse
+ * @example const chatResponse = await makeRequest(CHAT_COMPONENT, 'POST', request)
+            saveDebugResponse('summarizeNote', `summarize_${note.filename || ''}`, request, chatResponse)
+ */
+export function saveDebugResponse(folderName: string, filename: string, request: ChatObject, chatResponse: ChatResponse | null) {
+  if (chatResponse) {
+    const { saveResponses } = DataStore.settings
+    if (saveResponses) {
+      const fa = filename.split('/')
+      const fname = fa[fa.length - 1].replace(/\.md$|\.txt$/g, '').substring(0, 100) + String(new Date())
+      // logDebug(pluginJson, `saveDebugResponse fa=${fa.toString()} fname=${fname}`)
+      DataStore.saveJSON(chatResponse, `${folderName}/${fname}.${String(request.messages.length / 2)}.json`)
+      // clo(chatResponse, `chatResponse/${filename}.${String(request.messages.length / 2)}.json`)
+    }
+  }
+}