@yeaft/webchat-agent 0.1.416 → 0.1.418

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@yeaft/webchat-agent",
-  "version": "0.1.416",
+  "version": "0.1.418",
   "description": "Remote agent for Yeaft WebChat — connects worker machines to the central server",
   "main": "index.js",
   "type": "module",

package/unify/llm/chat-completions.js

@@ -12,6 +12,12 @@
  *   Response: delta.tool_calls[i].function.arguments (JSON string) → accumulate → JSON.parse → UnifiedToolCall
  *   Result:   UnifiedToolResult → { role: "tool", tool_call_id, content }
  *   Finish:   "tool_calls" → "tool_use", "stop" → "end_turn", "length" → "max_tokens"
+ *
+ * max_tokens strategy (based on model ID):
+ *   OpenAI models (gpt-*, o1*, o3*, o4*) use "max_completion_tokens" (new standard).
+ *   All other models (DeepSeek, Gemini, etc.) use "max_tokens" (legacy/compat).
+ *   CopilotProxy transparently forwards whatever the client sends.
+ *   Callers can override via extraBody to pass any parameter directly.
  */
 
 import {
@@ -23,6 +29,26 @@ import {
   LLMAbortError,
 } from './adapter.js';
 
+/**
+ * Check if a model ID is an OpenAI model that supports max_completion_tokens.
+ * OpenAI introduced max_completion_tokens with o1 and made it standard for
+ * GPT-4.1+, o-series, and GPT-5+. Other OpenAI-compatible APIs (DeepSeek,
+ * Gemini, Ollama) still only understand max_tokens.
+ *
+ * @param {string} model — The model ID (e.g. "gpt-5", "deepseek-chat", "o3")
+ * @returns {boolean} true = use max_completion_tokens
+ */
+export function useNewMaxTokensParam(model) {
+  if (!model) return false;
+  const m = model.toLowerCase();
+  // GPT-4.1+ and GPT-5+
+  if (m.startsWith('gpt-')) return true;
+  // o-series reasoning models (o1, o3, o4-mini, etc.)
+  if (/^o\d/.test(m)) return true;
+  // Everything else (deepseek-*, gemini-*, claude-*, custom models): legacy
+  return false;
+}
+
 /**
  * ChatCompletionsAdapter — Talks to OpenAI Chat Completions API and compatibles.
  */
@@ -39,6 +65,24 @@ export class ChatCompletionsAdapter extends LLMAdapter {
     this.#baseUrl = baseUrl.replace(/\/+$/, ''); // strip trailing slash
   }
 
+  /** Expose baseUrl for testing. */
+  get baseUrl() { return this.#baseUrl; }
+
+  /**
+   * Build the max-tokens portion of the request body.
+   * Uses max_completion_tokens for OpenAI models, max_tokens for others.
+   *
+   * @param {string} model
+   * @param {number} maxTokens
+   * @returns {object}
+   */
+  #maxTokensBody(model, maxTokens) {
+    if (useNewMaxTokensParam(model)) {
+      return { max_completion_tokens: maxTokens };
+    }
+    return { max_tokens: maxTokens };
+  }
+
   /**
    * Translate UnifiedToolDef[] → Chat Completions tool format.
    * @param {import('./adapter.js').UnifiedToolDef[]} tools
@@ -140,16 +184,16 @@ export class ChatCompletionsAdapter extends LLMAdapter {
   }
 
   /**
-   * @param {{ model: string, system: string, messages: import('./adapter.js').UnifiedMessage[], tools?: import('./adapter.js').UnifiedToolDef[], maxTokens?: number, signal?: AbortSignal }} params
+   * @param {{ model: string, system: string, messages: import('./adapter.js').UnifiedMessage[], tools?: import('./adapter.js').UnifiedToolDef[], maxTokens?: number, extraBody?: object, signal?: AbortSignal }} params
    * @returns {AsyncGenerator<import('./adapter.js').StreamEvent>}
    */
-  async *stream({ model, system, messages, tools, maxTokens = 16384, signal }) {
+  async *stream({ model, system, messages, tools, maxTokens = 16384, extraBody, signal }) {
     if (signal?.aborted) throw new LLMAbortError();
 
     const body = {
       model,
       messages: this.#translateMessages(system, messages),
-      max_tokens: maxTokens,
+      ...this.#maxTokensBody(model, maxTokens),
       stream: true,
       stream_options: { include_usage: true },
     };
@@ -157,6 +201,9 @@ export class ChatCompletionsAdapter extends LLMAdapter {
     const translatedTools = this.#translateTools(tools);
     if (translatedTools) body.tools = translatedTools;
 
+    // extraBody allows callers to pass through any additional/override parameters
+    if (extraBody) Object.assign(body, extraBody);
+
     const response = await fetch(`${this.#baseUrl}/chat/completions`, {
       method: 'POST',
       headers: {
@@ -277,15 +324,18 @@ export class ChatCompletionsAdapter extends LLMAdapter {
   /**
    * Non-streaming call for side queries.
    */
-  async call({ model, system, messages, maxTokens = 4096, signal }) {
+  async call({ model, system, messages, maxTokens = 4096, extraBody, signal }) {
     if (signal?.aborted) throw new LLMAbortError();
 
     const body = {
       model,
       messages: this.#translateMessages(system, messages),
-      max_tokens: maxTokens,
+      ...this.#maxTokensBody(model, maxTokens),
     };
 
+    // extraBody allows callers to pass through any additional/override parameters
+    if (extraBody) Object.assign(body, extraBody);
+
     const response = await fetch(`${this.#baseUrl}/chat/completions`, {
       method: 'POST',
       headers: {