symposium 0.13.9 → 0.14.1

package/Agent.js

@@ -88,6 +88,17 @@ export default class Agent {
 			this.callbacks[i + '-' + thread.id].push(callback);
 		}
 
+		const model = Symposium.getModelByName(thread.state.model);
+		if (!model.supports_audio) {
+			for (let c of content) {
+				if (c.type === 'audio' && !c.content?.transcription) {
+					const words = await this.getPromptWordsForTranscription(thread);
+					const prompt = words.length ? 'Possibili parole usate: ' + words.join(', ') : null;
+					c.content.transcription = await Symposium.transcribe(c.content, prompt);
+				}
+			}
+		}
+
 		await this.log('user_message', content);
 		thread.addMessage('user', content);
 
@@ -205,19 +216,6 @@ export default class Agent {
 	async generateCompletion(thread, options = {}, retry_counter = 1) {
 		try {
 			const model = Symposium.getModelByName(thread.state.model);
-
-			for (let message of thread.messages) {
-				for (let c of message.content) {
-					if (c.type === 'audio' && !model.supports_audio) {
-						const words = await this.getPromptWordsForTranscription(thread);
-						const prompt = words.length ? 'Possibili parole usate: ' + words.join(', ') : null;
-						const transcribed = await Symposium.transcribe(c.content, prompt);
-						c.type = 'text';
-						c.content = '[voice message] ' + transcribed;
-					}
-				}
-			}
-
 			const messages = await model.generate(thread, await this.getFunctions(), options);
 			return model.supports_functions ? messages : messages.map(m => this.parseFunctions(m));
 		} catch (error) {
@@ -261,6 +259,15 @@ export default class Agent {
 			return i.output(thread, msg);
 	}
 
+	async partial(thread, msg) {
+		if ((typeof msg) === 'text')
+			msg = {summary: msg};
+
+		const i = this.options.interfaces.find(i => i.name === thread.interface);
+		if (i)
+			return i.partial(thread, msg);
+	}
+
 	parseFunctions(message) {
 		const newContent = [];
 		for (let m of message.content) {
@@ -373,11 +380,18 @@ export default class Agent {
 		if (!functions.has(function_call.name))
 			throw new Error('Unrecognized function ' + function_call.name);
 
+		const func = functions.get(function_call.name);
+		const partialOutput = func.partialOutput ? ((typeof func.partialOutput) === 'text' ? func.partialOutput : func.partialOutput.call(this, function_call.arguments)) : 'Uso lo strumento ' + function_call.name + '...';
+		this.partial(thread, partialOutput);
+
 		await this.log('function_call', function_call);
 
 		try {
-			return await functions.get(function_call.name).tool.callFunction(thread, function_call.name, function_call.arguments);
+			const response = await func.tool.callFunction(thread, function_call.name, function_call.arguments);
+			this.partial(thread, 'Risposta ricevuta da ' + func.tool.name);
+			return response;
 		} catch (error) {
+			this.partial(thread, 'Ricevuto errore da ' + func.tool.name);
 			return {error};
 		}
 	}

package/Interface.js

@@ -7,6 +7,9 @@ export default class Interface {
 	static async output(thread, msg) {
 	}
 
+	static async partial(thread, msg) {
+	}
+
 	static async error(thread, error) {
 	}
 }

package/models/AnthropicModel.js

@@ -148,6 +148,17 @@ export default class AnthropicModel extends Model {
 							}
 							break;
 
+						case 'audio':
+							if (c.content.transcription) {
+								content.push({
+									type: 'text',
+									text: '[transcribed] ' + c.content.transcription,
+								});
+							} else {
+								throw new Error('Audio content is not supported by this model');
+							}
+							break;
+
 						default:
 							throw new Error('Message type "' + c.type + '" unsupported by this model');
 					}

package/models/GroqModel.js

@@ -158,6 +158,19 @@ export default class GroqModel extends Model {
 					}
 					break;
 
+
+				case 'audio':
+					if (c.content.transcription) {
+						messages.push({
+							role: message.role,
+							content: '[transcribed] ' + c.content.transcription,
+							name: message.name,
+						});
+					} else {
+						throw new Error('Audio content is not supported by this model');
+					}
+					break;
+
 				default:
 					throw new Error('Message type unsupported by this model');
 			}

package/models/OpenAIModel.js

@@ -140,24 +140,34 @@ export default class OpenAIModel extends Model {
 					break;
 
 				case 'audio':
-					if (c.content.type !== 'base64')
-						throw new Error('Audio content must be base64 encoded for this model');
-					if (!['audio/mpeg', 'audio/wav'].includes(c.content.mime))
-						throw new Error('Audio content must have a valid MIME type');
+					if (this.supports_audio) {
+						if (c.content.type !== 'base64')
+							throw new Error('Audio content must be base64 encoded for this model');
+						if (!['audio/mpeg', 'audio/wav'].includes(c.content.mime))
+							throw new Error('Audio content must have a valid MIME type');
 
-					messages.push({
-						role,
-						content: [
-							{
-								type: 'input_audio',
-								input_audio: {
-									data: c.content.data,
-									format: c.content.mime === 'audio/mpeg' ? 'mp3' : 'wav',
+						messages.push({
+							role,
+							content: [
+								{
+									type: 'input_audio',
+									input_audio: {
+										data: c.content.data,
+										format: c.content.mime === 'audio/mpeg' ? 'mp3' : 'wav',
+									},
 								},
-							},
-						],
-						name: message.name,
-					});
+							],
+							name: message.name,
+						});
+					} else if (c.content.transcription) {
+						messages.push({
+							role,
+							content: '[transcribed] ' + c.content.transcription,
+							name: message.name,
+						});
+					} else{
+						throw new Error('Audio content is not supported by this model');
+					}
 					break;
 
 				case 'function':

package/package.json

@@ -1,14 +1,14 @@
 {
   "type": "module",
   "name": "symposium",
-  "version": "0.13.9",
+  "version": "0.14.1",
   "description": "Agents",
   "main": "index.js",
   "author": "Domenico Giambra",
   "license": "ISC",
   "dependencies": {
-    "@anthropic-ai/sdk": "^0.53.0",
-    "groq-sdk": "^0.23.0",
+    "@anthropic-ai/sdk": "^0.54.0",
+    "groq-sdk": "^0.25.0",
     "openai": "^5.0.0",
     "tiktoken": "^1.0.10"
   }