openai 模型改为原生代码，移除 voyage

Author: linyuan.yang

packages/apps/admin/src/i18n/en.ts

@@ -334,7 +334,7 @@ export default {
     empty: 'No embedding configurations',
     add_title: 'Add Embedding',
     edit_title: 'Edit Embedding',
-    name_placeholder: 'e.g. openai-ada, gemini-004, voyage-3',
+    name_placeholder: 'e.g. openai-ada, gemini-004',
     confirm_delete: 'Delete embedding "{name}"?',
   },
   savers: {

packages/apps/admin/src/i18n/zh.ts

@@ -334,7 +334,7 @@ export default {
     empty: '暂无向量模型配置',
     add_title: '添加向量模型',
     edit_title: '编辑向量模型',
-    name_placeholder: '如 openai-ada、gemini-004、voyage-3',
+    name_placeholder: '如 openai-ada、gemini-004',
     confirm_delete: '确定要删除向量模型 "{name}" 吗?',
   },
   savers: {

packages/apps/docs/guide/features.md

@@ -7,7 +7,7 @@
 - **Multi-agent orchestration** — Single, ReAct (recursive task decomposition), and Generative (multimodal) modes; agents can be nested and composed
 - **ACP agent support** — Agent Client Protocol integration with persistent and transient agent modes
 - **Knowledge base** — Pluggable wiki data sources (local files, Google Drive, and third-party providers) with keyword + optional semantic search
-- **Long-term memory** — Vector-embedding semantic search for persistent context recall (OpenAI, Google, Ollama, Cohere, VoyageAI)
+- **Long-term memory** — Vector-embedding semantic search for persistent context recall (OpenAI, Google, Ollama, Cohere)
 - **Conversation compaction** — Automatic conversation summarization when token usage exceeds threshold, preserving continuity while reducing consumption
 - **Memory** — Per-agent automatic long-term memory: a background MemoryLLM extracts durable knowledge after each conversation idles; the agent reads it back via `search_memory` / `read_memory`, with consolidate/reconcile maintenance
 - **Agenda** — Conversation-driven reminders, schedules, and routines with absolute / interval / cron triggers; optionally synced from the conversation after every turn and delivered to any session or channel

packages/apps/docs/guide/models.md

@@ -36,4 +36,4 @@ Any endpoint that implements the OpenAI chat completions API can be used as a pr
 
 Sidebar → **Embedding Models** → New
 
-Embeddings are required for vector-based features ([Notes](./note), [Wiki](./wiki) semantic search, [Memory](./memory) hybrid search). Supported: OpenAI, Google, Ollama, Cohere, VoyageAI.
+Embeddings are required for vector-based features ([Notes](./note), [Wiki](./wiki) semantic search, [Memory](./memory) hybrid search). Supported: OpenAI, Google, Ollama, Cohere.

packages/apps/docs/guide/note.md

@@ -13,7 +13,7 @@ An embedding model is required first: sidebar → **Embedding Models** → New.
 | Field | Description |
 |-------|-------------|
 | Name  | Display name for this notebook |
-| Embedding | Embedding model used for semantic search (OpenAI, Google, Ollama, Cohere, VoyageAI) |
+| Embedding | Embedding model used for semantic search (OpenAI, Google, Ollama, Cohere) |
 
 ## How It Works
 

packages/apps/docs/zh/guide/features.md

@@ -7,7 +7,7 @@
 - **多 Agent 编排** —— Single、ReAct（递归任务分解）、Generative（多模态）三种模式，Agent 可嵌套组合
 - **ACP Agent 支持** —— Agent Client Protocol 集成，支持持久化与临时两种 Agent 模式
 - **知识库** —— 可插拔 Wiki 数据源（本地文件、Google Drive 与第三方来源），支持关键词与可选语义检索
-- **长期记忆** —— 基于向量 Embedding 的语义检索，持久化上下文召回（OpenAI、Gemini、Ollama、Cohere、VoyageAI）
+- **长期记忆** —— 基于向量 Embedding 的语义检索，持久化上下文召回（OpenAI、Gemini、Ollama、Cohere）
 - **对话压缩** —— Token 用量超阈值时自动摘要早期消息，保持上下文连续性同时降低消耗
 - **记忆（Memory）** —— Agent 级自动长期记忆：会话空闲后由后台 MemoryLLM 提炼持久知识，主 Agent 通过 `search_memory` / `read_memory` 召回，并有 consolidate / reconcile 维护任务
 - **日程（Agenda）** —— 由对话驱动的提醒、日程与周期任务，支持 absolute / interval / cron 触发器；可每轮对话后从对话自动同步，并投递到任意会话或渠道

packages/apps/docs/zh/guide/models.md

@@ -36,4 +36,4 @@
 
 侧栏 → **向量模型** → 新建
 
-向量模型是基于向量的功能（[Notes](./note)、[Wiki](./wiki) 语义检索、[Memory](./memory) 混合检索）的前置依赖。支持：OpenAI、Gemini、Ollama、Cohere、VoyageAI。
+向量模型是基于向量的功能（[Notes](./note)、[Wiki](./wiki) 语义检索、[Memory](./memory) 混合检索）的前置依赖。支持：OpenAI、Gemini、Ollama、Cohere。

packages/apps/docs/zh/guide/note.md

@@ -13,7 +13,7 @@ Note 仓库是供 Agent 召回的人工维护知识库。你可以在 Web UI 中
 | 字段 | 说明 |
 |-------|-------------|
 | 名称 | 该笔记本的显示名称 |
-| 向量模型 | 用于语义检索的模型（OpenAI、Google、Ollama、Cohere、VoyageAI） |
+| 向量模型 | 用于语义检索的模型（OpenAI、Google、Ollama、Cohere） |
 
 ## 工作原理
 

packages/backend/llm/scorpio.llm.anthropic/src/AnthropicModelService.ts

@@ -1,6 +1,7 @@
 import { ChatAnthropic } from "@langchain/anthropic";
 import { createFunctionCallingParser } from "@langchain/core/language_models/structured_output";
 import { AIMessageChunk, BaseMessage, SystemMessage } from "@langchain/core/messages";
+import type { AgentTool } from "scorpio.llm";
 import { toJsonSchema } from "@langchain/core/utils/json_schema";
 import {
   type ChatMessage,
@@ -53,14 +54,19 @@ export class AnthropicModelService extends ModelServiceBase<ChatAnthropic> {
     return typeof input === "string" ? input : this.applyCache(toBaseMessages(input));
   }
 
-  bindTools(tools: any[]): void {
-    if (this.cacheControl && tools.length > 0) {
-      const formatted = (this.model! as any).formatStructuredToolToAnthropic(tools);
+  bindTools(tools: AgentTool[]): void {
+    // AgentTool 的 schema 已是 JSON Schema，直接构造 Anthropic 原生工具格式
+    // （等价于 ChatAnthropic.formatStructuredToolToAnthropic 的输出形态）
+    const formatted: Array<{ name: string; description?: string; input_schema: Record<string, any>; cache_control?: any }> =
+      tools.map(tool => ({
+        name: tool.name,
+        description: tool.description,
+        input_schema: tool.schema,
+      }));
+    if (this.cacheControl && formatted.length > 0) {
       formatted[formatted.length - 1].cache_control = this.cacheControl;
-      this.boundModel = (this.model! as any).withConfig({ tools: formatted });
-    } else {
-      this.boundModel = this.model!.bindTools(tools);
     }
+    this.boundModel = (this.model! as any).withConfig({ tools: formatted });
   }
 
   async invokeStructured<T = any>(schema: any, prompt: string | ChatMessage[], options?: StructuredInvokeOptions): Promise<T> {

packages/backend/llm/scorpio.llm.gemini/package.json

@@ -6,6 +6,7 @@
   "types": "dist/index.d.ts",
   "scripts": { "build": "rimraf dist && tsc -b --force" },
   "dependencies": {
+    "@langchain/core": "catalog:",
     "@langchain/google-genai": "^2.1.31",
     "scorpio.llm": "workspace:*"
   },