wingbot 3.76.12 → 3.76.13

package/.claude/settings.local.json

@@ -0,0 +1,11 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(npx mocha:*)",
+      "Bash(npx tsc *)",
+      "Bash(npm test *)",
+      "Bash(node -e \"const p = require\\('./bot/plugins/BTMLLM'\\); console.log\\('factory type:', typeof p\\); console.log\\('factory name:', p.name\\);\")",
+      "Bash(grep -n \"prompt\\\\`\\\\|tagged\\\\|render\\\\|compile\\\\|hbs\" /Users/ondrejveres/Wingbot/wingbot-llm/src/prompt.js)"
+    ]
+  }
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "wingbot",
-  "version": "3.76.12",
+  "version": "3.76.13",
   "description": "Enterprise Messaging Bot Conversation Engine",
   "main": "index.js",
   "type": "commonjs",

package/src/ChatGpt.js

@@ -14,6 +14,7 @@ const LLM = require('./LLM');
 /** @typedef {import('./Responder').QuickReply} QuickReply */
 /** @typedef {import('./LLM').LLMMessage} LLMMessage */
 /** @typedef {import('./LLM').LLMProviderOptions} LLMProviderOptions */
+/** @typedef {import('./LLM').LLMUsage} LLMUsage */
 /** @typedef {import('./LLMSession').ToolFunction} ToolFunction */
 /** @typedef {import('./LLMSession').JsonSchemaProp} SimpleJsonSchema */
 
@@ -285,10 +286,13 @@ class ChatGpt {
      * @param {LLMMessage[]} prompt
      * @param {LLMProviderOptions} [options]
      * @param {ToolFunction[]} [tools=[]]
+     * @param {LLMUsage} [usage] - gets filled with the reported token usage
      * @returns {Promise<LLMMessage>}
      */
-    async requestChat (prompt, options, tools = []) {
-        const choice = await this._request(prompt, options, tools);
+    async requestChat (prompt, options, tools = [], usage = {}) {
+        const { choice, usage: reportedUsage } = await this._request(prompt, options, tools);
+
+        Object.assign(usage, reportedUsage);
 
         const { finish_reason: finishReason, message } = choice;
 
@@ -310,7 +314,7 @@ class ChatGpt {
      * @param {LLMMessage[]} chat
      * @param {LLMProviderOptions} requestOptions
      * @param {ToolFunction[]} [tools=[]]
-     * @returns {Promise<ChatGPTChoice>}
+     * @returns {Promise<{choice:ChatGPTChoice, usage:LLMUsage}>}
      */
     async _request (chat, requestOptions, tools = []) {
         const {
@@ -485,11 +489,28 @@ class ChatGpt {
 
             const [choice] = responseData.choices;
 
+            const { usage = {}, model: usedModel } = responseData;
+            const {
+                prompt_tokens: promptTokens,
+                completion_tokens: completionTokens,
+                prompt_tokens_details: promptDetails = {},
+                completion_tokens_details: completionDetails = {}
+            } = usage;
+
+            /** @type {LLMUsage} */
+            const llmUsage = {
+                model: usedModel,
+                promptTokens,
+                completionTokens,
+                cachedTokens: promptDetails.cached_tokens,
+                reasoningTokens: completionDetails.reasoning_tokens
+            };
+
             const durationMs = Date.now() - startTime;
 
             this._log('#GPT response', { choice, responseData, durationMs });
 
-            return choice;
+            return { choice, usage: llmUsage };
         } catch (e) {
             this._logger.log('#GPT failed', {
                 message: e.message, body, responseText, responseData, status, statusText
@@ -516,7 +537,9 @@ class ChatGpt {
             { role: 'user', content }
         ];
 
-        return this._request(messages, requestOptions);
+        const { choice } = await this._request(messages, requestOptions);
+
+        return choice;
     }
 
     /**

package/src/LLM.js

@@ -88,6 +88,7 @@ const LLMSession = require('./LLMSession');
  * @param {LLMMessage[]} prompt
  * @param {LLMProviderOptions} [options]
  * @param {ToolFunction[]} [tools]
+ * @param {LLMUsage} [usage] - gets filled with the reported token usage
  * @returns {Promise<LLMMessage>}
  */
 
@@ -158,6 +159,15 @@ const LLMSession = require('./LLMSession');
  * @prop {RegExp} regex
  */
 
+/**
+ * @typedef {object} LLMUsage
+ * @prop {string} [model]
+ * @prop {number} [promptTokens]
+ * @prop {number} [cachedTokens]
+ * @prop {number} [completionTokens]
+ * @prop {number} [reasoningTokens]
+ */
+
 /**
  * @typedef {object} PromptInfo
  * @prop {LLMMessage[]} prompt
@@ -167,6 +177,7 @@ const LLMSession = require('./LLMSession');
  * @prop {number} [durationMs]
  * @prop {LLMCallOptions} [options]
  * @prop {string} [error]
+ * @prop {LLMUsage} [usage]
  */
 
 /**
@@ -443,8 +454,10 @@ class LLM {
         const startTime = Date.now();
         let result = null;
         let error = null;
+        /** @type {LLMUsage} */
+        const usage = {};
         try {
-            result = await this._provider.requestChat(prompt, opts, session.tools);
+            result = await this._provider.requestChat(prompt, opts, session.tools, usage);
         } catch (e) {
             error = e;
         }
@@ -456,7 +469,8 @@ class LLM {
             session.tools,
             durationMs,
             opts,
-            error
+            error,
+            Object.keys(usage).length ? usage : null
         );
         if (error) throw error;
         return result;
@@ -471,8 +485,9 @@ class LLM {
      * @param {number} [durationMs]
      * @param {LLMCallOptions} [options]
      * @param {Error} [error]
+     * @param {LLMUsage} [usage]
      */
-    logPrompt (prompt, result, vectorSearchResult, tools, durationMs, options, error) {
+    logPrompt (prompt, result, vectorSearchResult, tools, durationMs, options, error, usage) {
         this._lastResult = result;
         this._configuration.logger.logPrompt({
             prompt,
@@ -481,6 +496,7 @@ class LLM {
             tools,
             durationMs,
             ...(options != null && { options }),
+            ...(usage != null && { usage }),
             ...(error != null && { error: error?.message || String(error) })
         });
     }

package/src/graphApi/schema.gql

@@ -129,6 +129,14 @@ type ToolFunction {
     parameters: Any
 }
 
+type LLMUsage {
+    model: String
+    promptTokens: Float
+    cachedTokens: Float
+    completionTokens: Float
+    reasoningTokens: Float
+}
+
 type PromptInfo {
     prompt: [LLMMessage!]!
     result: LLMMessage
@@ -137,6 +145,7 @@ type PromptInfo {
     durationMs: Float
     options: Any
     error: String
+    usage: LLMUsage
 }
 
 type UserInteraction {