Added tps + duration to AI history

src/open-ai.ts

@@ -21,13 +21,15 @@ export class OpenAi extends LLMProvider {
 			const h = history[i];
 			if(h.role === 'assistant' && h.tool_calls) {
 				const items: any[] = [];
-				if(h.content) items.push({role: 'assistant', content: h.content, timestamp: h.timestamp});
+				if(h.content) items.push({role: 'assistant', content: h.content, timestamp: h.timestamp, duration: h.duration, tps: h.tps});
 				items.push(...h.tool_calls.map((tc: any) => ({
 					role: 'tool',
 					id: tc.id,
 					name: tc.function.name,
 					args: JSONAttemptParse(tc.function.arguments, {}),
-					timestamp: h.timestamp
+					timestamp: h.timestamp,
+					duration: h.duration,
+					tps: h.tps
 				})));
 				history.splice(i, 1, ...items);
 				i += items.length - 1;
@@ -110,23 +112,27 @@ export class OpenAi extends LLMProvider {
 				};
 			}
 
-			let resp: any, terminal = false;
+			if(options.stream) requestParams.stream_options = {include_usage: true};
+			let resp: any, terminal = false, duration = 0, tps = 0;
 			do {
 				requestParams.messages = history.map(({timestamp, ...m}) => m);
+				const callStart = Date.now();
 				resp = await this.client.chat.completions.create(requestParams).catch(err => {
 					err.message += `\n\nMessages:\n${JSON.stringify(history, null, 2)}`;
 					throw err;
 				});
 
+				let usage: any;
 				if(options.stream) {
 					resp.choices = [{message: {role: 'assistant', content: '', tool_calls: [], timestamp: Date.now()}}];
 					for await (const chunk of resp) {
 						if(controller.signal.aborted) break;
-						if(chunk.choices[0].delta.content) {
+						if(chunk.usage) usage = chunk.usage;
+						if(chunk.choices[0]?.delta?.content) {
 							resp.choices[0].message.content += chunk.choices[0].delta.content;
 							options.stream({text: chunk.choices[0].delta.content});
 						}
-						if(chunk.choices[0].delta.tool_calls) {
+						if(chunk.choices[0]?.delta?.tool_calls) {
 							for(const deltaTC of chunk.choices[0].delta.tool_calls) {
 								const existing = resp.choices[0].message.tool_calls.find(tc => tc.index === deltaTC.index);
 								if(existing) {
@@ -151,19 +157,22 @@ export class OpenAi extends LLMProvider {
 							}
 						}
 					}
+				} else {
+					usage = resp.usage;
 				}
+				duration = Date.now() - callStart;
+				tps = usage?.completion_tokens && duration > 0 ? usage.completion_tokens / (duration / 1000) : 0;
 
 				if(resp.error) throw new Error(resp.error);
 				const toolCalls = resp.choices[0].message.tool_calls || [];
 				if(toolCalls.length && !controller.signal.aborted) {
-					history.push(resp.choices[0].message);
+					history.push({...resp.choices[0].message, duration, tps});
 					const results = await Promise.all(toolCalls.map(async (toolCall: any) => {
 						const tool = tools?.find(findByProp('name', toolCall.function.name));
 						if(options.stream) options.stream({tool: toolCall.function.name});
 						if(!tool) return {role: 'tool', tool_call_id: toolCall.id, content: '{"error": "Tool not found"}', timestamp: Date.now()};
 						try {
 							const args = JSONAttemptParse(toolCall.function.arguments, {});
-							// Wrap stream so a tool's `done` ends turn gracefully
 							const toolStream = options.stream && ((chunk: any) => {
 								if(chunk.done) { terminal = true; return; }
 								options.stream!(chunk);
@@ -181,8 +190,9 @@ export class OpenAi extends LLMProvider {
 
 			if(!terminal) {
 				const textContent = resp.choices[0].message.content || '';
-				history.push({role: 'assistant', content: textContent.trim(), timestamp: Date.now()});
+				history.push({role: 'assistant', content: textContent.trim(), timestamp: Date.now(), duration, tps});
 			}
+
 			history = this.toStandard(history);
 			if(options.history) options.history.splice(0, options.history.length, ...history.filter(h => h.role !== 'system'));
 			if(options.stream) options.stream({done: true});