initial checkin

.eslintignore

@@ -0,0 +1 @@
+test/*

.eslintrc.json

@@ -0,0 +1,126 @@
+{
+  "env": {
+    "node": true,
+    "es6": true  
+  },
+  "parserOptions": {
+    "ecmaFeatures": {
+      "jsx": false,
+      "modules": false
+    },
+    "ecmaVersion": 2020
+  },
+  "plugins": ["promise"],
+  "rules": {
+    "promise/always-return": "error",
+    "promise/no-return-wrap": "error",
+    "promise/param-names": "error",
+    "promise/catch-or-return": "error",
+    "promise/no-native": "off",
+    "promise/no-nesting": "warn",
+    "promise/no-promise-in-callback": "warn",
+    "promise/no-callback-in-promise": "warn",
+    "promise/no-return-in-finally": "warn",
+
+    // Possible Errors
+    // http://eslint.org/docs/rules/#possible-errors
+    "comma-dangle": [2, "only-multiline"],
+    "no-control-regex": 2,
+    "no-debugger": 2,
+    "no-dupe-args": 2,
+    "no-dupe-keys": 2,
+    "no-duplicate-case": 2,
+    "no-empty-character-class": 2,
+    "no-ex-assign": 2,
+    "no-extra-boolean-cast" : 2,
+    "no-extra-parens": [2, "functions"],
+    "no-extra-semi": 2,
+    "no-func-assign": 2,
+    "no-invalid-regexp": 2,
+    "no-irregular-whitespace": 2,
+    "no-negated-in-lhs": 2,
+    "no-obj-calls": 2,
+    "no-proto": 2,
+    "no-unexpected-multiline": 2,
+    "no-unreachable": 2,
+    "use-isnan": 2,
+    "valid-typeof": 2,
+
+    // Best Practices
+    // http://eslint.org/docs/rules/#best-practices
+    "no-fallthrough": 2,
+    "no-octal": 2,
+    "no-redeclare": 2,
+    "no-self-assign": 2,
+    "no-unused-labels": 2,
+
+    // Strict Mode
+    // http://eslint.org/docs/rules/#strict-mode
+    "strict": [2, "never"],
+
+    // Variables
+    // http://eslint.org/docs/rules/#variables
+    "no-delete-var": 2,
+    "no-undef": 2,
+    "no-unused-vars": [2, {"args": "none"}],
+
+    // Node.js and CommonJS
+    // http://eslint.org/docs/rules/#nodejs-and-commonjs
+    "no-mixed-requires": 2,
+    "no-new-require": 2,
+    "no-path-concat": 2,
+    "no-restricted-modules": [2, "sys", "_linklist"],
+
+    // Stylistic Issues
+    // http://eslint.org/docs/rules/#stylistic-issues
+    "comma-spacing": 2,
+    "eol-last": 2,
+    "indent": [2, 2, {"SwitchCase": 1}],
+    "keyword-spacing": 2,
+    "max-len": [2, 120, 2],
+    "new-parens": 2,
+    "no-mixed-spaces-and-tabs": 2,
+    "no-multiple-empty-lines": [2, {"max": 2}],
+    "no-trailing-spaces": [2, {"skipBlankLines": false }],
+    "quotes": [2, "single", "avoid-escape"],
+    "semi": 2,
+    "space-before-blocks": [2, "always"],
+    "space-before-function-paren": [2, "never"],
+    "space-in-parens": [2, "never"],
+    "space-infix-ops": 2,
+    "space-unary-ops": 2,
+
+    // ECMAScript 6
+    // http://eslint.org/docs/rules/#ecmascript-6
+    "arrow-parens": [2, "always"],
+    "arrow-spacing": [2, {"before": true, "after": true}],
+    "constructor-super": 2,
+    "no-class-assign": 2,
+    "no-confusing-arrow": 2,
+    "no-const-assign": 2,
+    "no-dupe-class-members": 2,
+    "no-new-symbol": 2,
+    "no-this-before-super": 2,
+    "prefer-const": 2
+  },
+  "globals": {
+    "DTRACE_HTTP_CLIENT_REQUEST"           : false,
+    "LTTNG_HTTP_CLIENT_REQUEST"            : false,
+    "COUNTER_HTTP_CLIENT_REQUEST"          : false,
+    "DTRACE_HTTP_CLIENT_RESPONSE"          : false,
+    "LTTNG_HTTP_CLIENT_RESPONSE"           : false,
+    "COUNTER_HTTP_CLIENT_RESPONSE"         : false,
+    "DTRACE_HTTP_SERVER_REQUEST"           : false,
+    "LTTNG_HTTP_SERVER_REQUEST"            : false,
+    "COUNTER_HTTP_SERVER_REQUEST"          : false,
+    "DTRACE_HTTP_SERVER_RESPONSE"          : false,
+    "LTTNG_HTTP_SERVER_RESPONSE"           : false,
+    "COUNTER_HTTP_SERVER_RESPONSE"         : false,
+    "DTRACE_NET_STREAM_END"                : false,
+    "LTTNG_NET_STREAM_END"                 : false,
+    "COUNTER_NET_SERVER_CONNECTION_CLOSE"  : false,
+    "DTRACE_NET_SERVER_CONNECTION"         : false,
+    "LTTNG_NET_SERVER_CONNECTION"          : false,
+    "COUNTER_NET_SERVER_CONNECTION"        : false  
+  }
+}

.gitignore

@@ -0,0 +1,45 @@
+# Logs
+logs
+*.log
+
+# Runtime data
+pids
+*.pid
+*.seed
+
+# github pages site
+_site
+
+#transient test cases
+examples/nosave.*.js
+
+# Directory for instrumented libs generated by jscoverage/JSCover
+lib-cov
+
+# Coverage directory used by tools like istanbul
+coverage
+.nyc_output/
+
+# Grunt intermediate storage (http://gruntjs.com/creating-plugins#storing-task-files)
+.grunt
+
+# node-waf configuration
+.lock-wscript
+
+# Compiled binary addons (http://nodejs.org/api/addons.html)
+build/Release
+
+# Dependency directory
+# https://www.npmjs.org/doc/misc/npm-faq.html#should-i-check-my-node_modules-folder-into-git
+node_modules
+
+.DS_Store
+
+examples/*
+
+ecosystem.config.js
+.vscode
+test/credentials/*.json
+run-tests.sh
+run-coverage.sh
+.vscode

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2024 Drachtio Communications Services, LLC
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -0,0 +1,25 @@
+# talk-to-your-llm
+
+This is a basic voicebot connected to using their Assistant framework.
+
+## Installing
+
+The basics:
+```bash
+npm ci
+WS_PORT=3000 OPENAI_API_KEY=xxxx node app.js
+```
+
+In the jambonz webapp create an application with url `wss://<your-domain>/llm-voicebot` and route calls to it.
+
+## Environment variables
+
+|variable|meaning|required|
+|---------|------|--------|
+|OPENAI_API_KEY|Your api key|yes|
+|OPENAI_MODEL|model to use|no (default: gpt-4-turbo)|
+|BOT_NAME|Name of Assistant to create|no (default: jambonz-llm-voicebot)|
+
+## Limitations
+
+Currently, there is no support for adding [Tools](https://platform.openai.com/docs/assistants/tools) to the Assistant.  User input is transcribed as presented as simple text and the OpenAI assistant is encouraged (via system instructions) to respond with text that is brief and un-annotated.

app.js

@@ -0,0 +1,12 @@
+const {createServer} = require('http');
+const {createEndpoint} = require('@jambonz/node-client-ws');
+const server = createServer();
+const makeService = createEndpoint({server});
+const logger = require('pino')({level: process.env.LOGLEVEL || 'info'});
+const port = process.env.WS_PORT || 3000;
+
+require('./lib/routes')({logger, makeService});
+
+server.listen(port, () => {
+  logger.info(`jambonz websocket server listening at http://localhost:${port}`);
+});

data/settings.json

@@ -0,0 +1,3 @@
+{
+  "system_instructions": "Answer questions directly without greeting the user.  Respond with plain text only: no annotations, markup or citations.  Separate paragraphs with 2 newline characters. Keep your responses to between 1-4 paragraphs if possible while still providing a complete answer."
+}

lib/routes/index.js

@@ -0,0 +1,4 @@
+module.exports = ({logger, makeService}) => {
+  require('./llm-voicebot')({logger, makeService});
+};
+

lib/routes/llm-voicebot.js

@@ -0,0 +1,110 @@
+
+const OpenAIAssistant = require('../utils/ai-assistants/llm-openai');
+const {system_instructions} = require('../../data/settings.json');
+const {processStreamingResponse, streamingResponseComplete} = require('../utils/process-streaming-response');
+
+const service = ({logger, makeService}) => {
+  const assistant = new OpenAIAssistant({
+    logger,
+    model: process.env.OPENAI_MODEL || 'gpt-4-turbo',
+    name: process.env.BOT_NAME || 'jambonz-llm-voicebot',
+    instructions: system_instructions
+  });
+  assistant.init();
+
+  const svc = makeService({path: '/llm-voicebot'});
+
+  svc.on('session:new', (session, path) => {
+    session.locals = { ...session.locals,
+      logger: logger.child({call_sid: session.call_sid}),
+      deepgramOptions: {
+        endpointing: 350,
+        utteranceEndMs: 1000,
+      },
+      turns: 0,
+      says: 0,
+      textOffset: 0,
+      assistant
+    };
+    session.locals.logger.info({session, path}, `new incoming call: ${session.call_sid}`);
+
+    session
+      .on('/user-input-event', onUserInputEvent.bind(null, session))
+      .on('close', onClose.bind(null, session))
+      .on('error', onError.bind(null, session));
+
+    session
+      .answer()
+      .pause({length: 0.5})
+      .config({
+        recognizer: {
+          vendor: 'default',
+          language: 'default',
+          deepgramOptions: session.locals.deepgramOptions
+        },
+        bargeIn: {
+          enable: true,
+          input: ['speech'],
+          actionHook: '/user-input-event',
+          sticky: true,
+        }
+      })
+      .say({text: 'Hi there!  You are speaking to chat GPT.  What would you like to know?'})
+      .reply();
+  });
+};
+
+const onUserInputEvent = async(session, evt) => {
+  const {logger} = session.locals;
+  logger.info({evt}, 'got speech evt');
+
+  switch (evt.reason) {
+    case 'speechDetected':
+      handleUserUtterance(session, evt);
+      break;
+    case 'timeout':
+      break;
+    default:
+      session.reply();
+      break;
+  }
+
+};
+
+const handleUserUtterance = async(session, evt) => {
+  const {logger, assistant, thread} = session.locals;
+  const {speech} = evt;
+  const userMessage = speech.alternatives[0].transcript;
+  logger.info({utterance: speech.alternatives[0]}, 'handling user utterance');
+
+  if (!thread) {
+    const thread = await assistant.createThread(userMessage);
+    thread
+      .on('botStreamingResponse', (response) => {
+        processStreamingResponse(session, response);
+      })
+      .on('botStreamingCompleted', () => {
+        streamingResponseComplete(session);
+      });
+    session.locals.thread = thread;
+  }
+  else {
+    thread.addUserMessage(speech.alternatives[0].transcript);
+  }
+  session.reply();
+};
+
+const onClose = (session) => {
+  const {logger, thread} = session.locals;
+  logger.info('call ended');
+  if (thread) {
+    thread.close();
+  }
+};
+
+const onError = (session, err) => {
+  const {logger} = session.locals;
+  logger.error(err, 'Error in call');
+};
+
+module.exports = service;

lib/utils/ai-assistants/llm-openai.js

@@ -0,0 +1,98 @@
+const OpenAI = require('openai');
+const openai = new OpenAI();
+const Emitter = require('events');
+
+/* TODO: in future, add support for other AI Assistants beyond OpenAI */
+class OpenAIThread extends Emitter {
+  constructor(logger, assistant_id, thread, run) {
+    super();
+
+    this.logger = logger;
+    this.assistant_id = assistant_id;
+    this.thread = thread;
+    this.run = run;
+
+    this._attachListeners();
+  }
+
+  _attachListeners() {
+    this.run
+      .on('event', (evt) => {
+        if (evt.event === 'thread.run.completed') {
+          this.emit('botStreamingCompleted');
+        }
+      })
+      .on('textDelta', (delta, snapshot) => {
+        this.emit('botStreamingResponse', snapshot.value);
+      })
+      .on('connect', () => this.logger.info('connected'))
+      .on('end', () => this.logger.info('ended'));
+  }
+
+  async addUserMessage(userMessage) {
+    this.logger.info('adding user message');
+    await openai.beta.threads.messages.create(this.thread.id, {
+      role: 'user',
+      content: userMessage,
+    });
+    this.run = await openai.beta.threads.runs.stream(this.thread.id, {
+      assistant_id: this.assistant_id
+    });
+    this._attachListeners();
+  }
+
+  async close() {
+    await openai.beta.threads.del(this.thread.id);
+  }
+}
+
+class OpenAIAssistant {
+  constructor({logger, model, name, instructions}) {
+    this.logger = logger;
+    this.model = model;
+    this.name = name;
+    this.instructions = instructions;
+  }
+
+  async init() {
+    try {
+      this.assistant = await openai.beta.assistants.create({
+        model: this.model,
+        name: this.name,
+        instructions: this.instructions,
+      });
+    } catch (err) {
+      this.logger.error(err);
+      throw err;
+    }
+  }
+
+  async createThread(initialUserMessage) {
+    /* create a thread */
+    try {
+      this.logger.info({initialUserMessage}, `creating thread with message: ${initialUserMessage}`);
+      const thread = await openai.beta.threads.create({
+        messages: [
+          {
+            role: 'user',
+            content: initialUserMessage,
+          },
+        ],
+      });
+
+      /* run the thread */
+      this.logger.info('running thread');
+      const run = await openai.beta.threads.runs.stream(thread.id, {
+        assistant_id: this.assistant.id,
+      });
+
+      const t = new OpenAIThread(this.logger, this.assistant.id, thread, run);
+      return t;
+    } catch (err) {
+      this.logger.error(err);
+      throw err;
+    }
+  }
+}
+
+module.exports = OpenAIAssistant;

lib/utils/process-streaming-response.js

@@ -0,0 +1,66 @@
+
+const processStreamingResponse = async(session, text) => {
+  const {logger, says, textOffset} = session.locals;
+  let spoken = false;
+
+  /* trim the leading part of the text we have already said */
+  const trimmed = text.substring(textOffset);
+
+  /**
+   * when we have a new response, say the first sentence as soon as available to get it out,
+   * after that fall back to speaking the rest of the response in chunks of paragraphs
+   */
+
+  if (says === 0) {
+    const pos = trimmed.indexOf('.');
+    if (-1 !== pos) {
+      const firstSentence = trimmed.substring(0, pos + 1);
+      logger.info(`speaking first sentence: ${firstSentence}`);
+      session
+        .say({text: firstSentence})
+        .send({execImmediate: true});
+      session.locals.says++;
+      session.locals.textOffset = pos + 1;
+      spoken = true;
+      session.locals.unsent = trimmed.substring(pos + 1);
+    }
+  }
+  if (says > 0) {
+    const pos = trimmed.indexOf('\n\n');
+    if (-1 !== pos) {
+      const paragraph = trimmed.substring(0, pos);
+      logger.info(`speaking paragraph: ${paragraph}`);
+      session
+        .say({text: paragraph})
+        .send({execImmediate: false});
+      session.locals.says++;
+      session.locals.textOffset += (pos + 2);
+      spoken = true;
+      session.locals.unsent = trimmed.substring(pos + 2);
+    }
+  }
+
+  if (!spoken) {
+    session.locals.unsent = trimmed;
+  }
+};
+
+const streamingResponseComplete = async(session) => {
+  const {logger, unsent} = session.locals;
+
+  if (unsent) {
+    logger.info('sending final unsent text');
+    session
+      .say({text: unsent})
+      .send({execImmediate: false});
+  }
+  session.locals.turns++;
+  session.locals.says = 0;
+  session.locals.textOffset = 0;
+  session.locals.unsent = '';
+};
+
+module.exports = {
+  processStreamingResponse,
+  streamingResponseComplete
+};

package.json

@@ -0,0 +1,22 @@
+{
+  "name": "talk-to-your-llm",
+  "version": "0.0.1",
+  "description": "jambonz websocket application",
+  "main": "app.js",
+  "scripts": {
+    "start": "node app",
+    "jslint": "eslint app.js lib"
+  },
+  "author": "",
+  "license": "MIT",
+  "dependencies": {
+    "@jambonz/node-client-ws": "^0.1.41",
+    "bent": "^7.3.12",
+    "openai": "^4.33.0",
+    "pino": "^8.20.0"
+  },
+  "devDependencies": {
+    "eslint": "^8.57.0",
+    "eslint-plugin-promise": "^6.1.1"
+  }
+}