@xuda.io/ai_module 1.1.5126 → 1.1.5127

package/index.mjs

@@ -4785,7 +4785,7 @@ setTimeout(async () => {
 
   // await init_studio_units();
   let ret;
-  // ret = await get_chat_suggestions(uid, [], 'coi_7b65b783af38b91c8ec4c499d3694bab');
+  ret = await get_chat_suggestions(uid, [], 'coi_7b65b783af38b91c8ec4c499d3694bab');
   // ret = await get_active_account_profile_info(uid);
   // ret = await is_spam_email(uid, 'dsgr8-2994400420@gigs.craigslist.org', 'To: You, From: Us - 70% Off Stickers & Labels Today Only');
   // ret = await is_spam_email(uid, 'noreply@tm.openai.com', 'Please update your Plus payment method', account_profile_info);
@@ -10888,39 +10888,43 @@ export const get_chat_suggestions = async function (uid, ai_agents, conversation
 
     const agents_docs = await get_agents();
 
+    const AgentItemSchema = z.object({
+      agent_id: z.string().describe('The unique identifier for the agent'),
+      short_action: z.string().describe('A brief, 2-4 word call to action'),
+      agent_description: z.string().describe("A professional description of the agent's capabilities"),
+    });
+
+    // Define the response format as a fixed-length array of 5
+    const AgentResponseSchema = z.object({
+      agents: z.array(AgentItemSchema).length(5),
+    });
+
     const chat_ret = await submit_chat_gpt_prompt({
       uid,
       prompt: `
      
-      ## goal:
-      return 5 agent names best fit from the ## agent list
-      considure the following parametes:
-      
-      1. the ## user info::
-      2. the ## user prompt
-      3. the ## prompt category
-      4. the ## prompt response
+     Role: You are a routing assistant.
+
+    Task: Analyze the provided user context and select the 5 most relevant agents from the ## Agent List.
+
+    Selection Logic:
+    Base your selection on the alignment between the agent's purpose and the following context:
 
-      ## user info: 
-      ${JSON.stringify(account_doc.account_info)}
+    User Profile: ${JSON.stringify(account_doc.account_info)}
 
-      ## user prompt:
-      ${conversation_item_doc.prompt || conversation_doc.prompt}
+    Recent Interaction: The user asked ${conversation_item_doc.prompt || conversation_doc.prompt}, categorized as ${conversation_doc?.category_info?.category}, which resulted in the response: ${conversation_item_doc.text}.
 
-      ## prompt category:
-      ${conversation_doc?.category_info?.category}
+    Data Source: > ${JSON.stringify(agents_docs)}
 
-      ## prompt response:
-      ${conversation_item_doc.text} 
+    Output: > Return a JSON array of exactly 5 objects following this schema:
+    [{ "agent_id": "string", "short_action": "string", "agent_description": "string" }]
 
-      ## agent list:
-      ${JSON.stringify(agents_docs)}
+      ## output example:
+      [{agent_id:"392_agn_55cb328d2743",short_action:"Create article","agent_description":"create proffessional article..."}]
      
       `,
       model: ai_model,
-      // response_format: z.object({
-      //   mood_level: z.number().describe('mood level of interaction: -2: very negative,-1: negative 0:natural 1:positive 2: very positive'),
-      // }),
+      response_format: AgentResponseSchema,
       metadata: { conversation_item_id, func: 'get_chat_suggestions' },
       account_profile_info,
     });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@xuda.io/ai_module",
-  "version": "1.1.5126",
+  "version": "1.1.5127",
   "description": "Xuda AI Module",
   "main": "index.mjs",
   "type": "module",