symposium 0.15.8 → 1.0.1

package/Agent.js

@@ -111,7 +111,10 @@ export default class Agent {
 		await this.log('user_message', content);
 		thread.addMessage('user', content);
 
-		return this.execute(thread);
+		const emitter = new BufferedEventEmitter();
+		emitter.emit('start', thread);
+
+		return this.execute(thread, emitter);
 	}
 
 	async beforeExecute(thread, emitter) {
@@ -120,9 +123,7 @@ export default class Agent {
 		return thread;
 	}
 
-	async execute(thread, counter = 0, existing_emitter = null) {
-		const emitter = existing_emitter || new BufferedEventEmitter();
-
+	async execute(thread, emitter, counter = 0) {
 		const execution = new Promise(async (resolve, reject) => {
 			try {
 				if (counter === 0)
@@ -189,7 +190,7 @@ export default class Agent {
 
 						case 'chat':
 							if (response?.type === 'continue')
-								return this.execute(thread, 0, emitter);
+								return this.execute(thread, emitter);
 
 							return resolve(null);
 
@@ -200,7 +201,7 @@ export default class Agent {
 					console.error(e);
 
 					if (counter < this.max_retries)
-						await this.execute(thread, counter + 1, emitter);
+						await this.execute(thread, emitter, counter + 1);
 				}
 			} catch (e) {
 				reject(e);

package/README.md

@@ -13,10 +13,22 @@ Symposium is a powerful and flexible Node.js framework for building Large Langua
 
 ## Installation
 
+Requires Node.js v18 or higher.
+
 ```bash
 npm install symposium
 ```
 
+## Configuration
+
+Symposium uses environment variables to configure access to various services. You can set these in a `.env` file at the root of your project.
+
+-   `OPENAI_API_KEY`: Required for using OpenAI models and for Real-time Voice Sessions.
+-   `ANTHROPIC_API_KEY`: Required for using Anthropic models.
+-   `GROQ_API_KEY`: Required for using Groq models.
+-   `DEEPSEEK_API_KEY`: Required for using DeepSeek models.
+-   `TRANSCRIPTION_MODEL`: (Optional) The name of the model to use for audio transcription (currently, only `gpt4o-transcribe` is supported).
+
 ## Core Concepts
 
 The framework is built around a few core components:
@@ -26,6 +38,9 @@ The framework is built around a few core components:
 -   **`Thread`**: Represents a single conversation with an agent. It maintains the message history and the agent's state for that conversation. Each thread has a unique ID.
 -   **`Tool`**: A base class for creating tools that an `Agent` can use. Tools expose functions that the LLM can call to interact with external APIs or data.
 -   **`Message`**: A wrapper for messages within a `Thread`, containing the role (`user`, `assistant`, `system`, `tool`), content, and other metadata.
+-   **`MemoryHandler`**: A class for managing an agent's long-term memory. It can be extended to create custom memory strategies.
+-   **`Summarizer`**: A utility agent for summarizing text or conversations.
+-   **`Logger`**: A simple logging utility that can be passed to an agent to log its activity.
 
 ## Getting Started
 
@@ -85,12 +100,25 @@ async function main() {
 	emitter.on('output', (content) => {
 		process.stdout.write(content);
 	});
+
+	emitter.on('error', (error) => {
+		console.error(`\nAn error occurred: ${error.message}`);
+	});
+
+	emitter.on('partial', (text) => {
+		console.log(`\n> ${text}\n`);
+	});
 }
 
 main();
 ```
 
-When you run this, the agent will respond to your message, and the response will be streamed to the console. The `message` method returns an `EventEmitter` that emits `data` events for text chunks, partial tool usage, and the final response object.
+When you run this, the agent will respond to your message, and the response will be streamed to the console. The `message` method returns an `EventEmitter` that emits several events:
+
+-   `start`: Emitted when the agent begins processing the message. The `thread` object is passed as an argument.
+-   `output`: Emitted for each chunk of text in the response stream.
+-   `partial`: Emitted to provide insight into the agent's internal state, like when it decides to use a tool.
+-   `error`: Emitted if an error occurs during processing.
 
 ## Advanced Usage
 
@@ -158,6 +186,43 @@ const emitter = await agent.message("What's the weather like in Paris?");
 
 The agent's underlying LLM will now be able to see the `get_weather` function and will call it when appropriate, passing the result back into the conversation.
 
+### Real-time Voice and Transcription
+
+Symposium has built-in support for audio transcription and real-time voice sessions, currently powered by OpenAI.
+
+#### Audio Transcription
+
+You can send audio content directly in a message. If the model doesn't support audio input, Symposium will automatically transcribe it to text.
+
+```javascript
+// Transcribing an audio file from a URL
+const emitter = await agent.message([
+    {
+        type: 'audio',
+        content: {
+            type: 'url',
+            data: 'http://example.com/audio.mp3'
+        }
+    }
+]);
+```
+
+You can also use the static `Symposium.transcribe()` method for standalone transcription.
+
+#### Real-time Voice Sessions
+
+For interactive voice conversations, you can create a real-time session. This is useful for building voice bots.
+
+```javascript
+// (inside an async function)
+const { response, thread } = await agent.createRealtimeSession();
+const sessionId = response.id;
+const clientSecret = response.client_secret.value;
+
+// You would then use this session ID and client secret on the client-side
+// to connect to the real-time session endpoint.
+```
+
 ### Switching Models
 
 You can set a default model for an agent or change it on a per-thread basis.
@@ -271,6 +336,12 @@ This is a high-level overview. For details, please refer to the source code.
 -   `getFunctions()`: **Abstract**. Must return an array of function definitions that the LLM can call.
 -   `callFunction(thread, name, payload)`: **Abstract**. Called when the LLM decides to use one of the tool's functions.
 
+### Other Classes
+
+-   **`MemoryHandler`**: Provides a base for implementing long-term memory for an agent.
+-   **`Summarizer`**: A utility agent for text summarization.
+-   **`Logger`**: A simple logger for agent activity.
+
 ## License
 
 ISC

package/models/{Whisper.js → Gpt4oTranscribe.js}

@@ -1,8 +1,8 @@
 import OpenAIModel from "./OpenAIModel.js";
 
-export default class Whisper extends OpenAIModel {
+export default class Gpt4oTranscribe extends OpenAIModel {
 	type = 'stt';
-	name = 'whisper';
+	name = 'gpt4o-transcribe';
 
 	async transcribe(file, prompt = null) {
 		const response = await this.getOpenAi().audio.transcriptions.create({

package/package.json

@@ -1,7 +1,7 @@
 {
   "type": "module",
   "name": "symposium",
-  "version": "0.15.8",
+  "version": "1.0.1",
   "description": "Agents",
   "main": "index.js",
   "author": "Domenico Giambra",