Attach tps to response promise

package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@ztimson/ai-utils",
-	"version": "1.3.5",
+	"version": "1.3.6",
 	"description": "AI Utility library",
 	"author": "Zak Timson",
 	"license": "MIT",

src/llm.ts

@@ -266,7 +266,10 @@ ${a.system}`,
 			nestedAborts.forEach(a => a());
 		};
 
-		const promise = (async () => {
+		let promise: any;
+		const requestStart = Date.now();
+
+		promise = (async () => {
 			let tools: AiTool[] = options.tools || this.ai.options.llm?.tools || [];
 			const prompts: string[] = [];
 			let history = options.history || [];
@@ -336,6 +339,8 @@ Also relevant but not preloaded (use \`memory_recall\`): ${listed.map(r => r.nam
 			const toolTimings = new Map<string, {duration: number, tps: number}>();
 			tools = this.wrapToolTiming(tools, toolTimings);
 
+			if(aborted) throw Object.assign(new Error('Aborted'), {name: 'AbortError'});
+
 			prompts.unshift(options.system || this.ai.options.llm?.system || '');
 			request = this.models[m].ask(message, {...options, tools, system: prompts.filter(Boolean).join('\n\n')});
 			let resp = await request;
@@ -374,6 +379,13 @@ Also relevant but not preloaded (use \`memory_recall\`): ${listed.map(r => r.nam
 				if(options.history) options.history.splice(0, options.history.length, ...compressed);
 			}
 
+			const requestDuration = Date.now() - requestStart;
+			const totalTokens = history
+				.filter((h: any) => h.role === 'assistant' && h.duration && h.tps)
+				.reduce((sum: number, h: any) => sum + h.tps * (h.duration / 1000), 0);
+			const requestTps = requestDuration > 0 ? totalTokens / (requestDuration / 1000) : 0;
+			Object.assign(promise, {duration: requestDuration, tps: requestTps});
+
 			return resp;
 		})();