@easynet/agent-model 1.0.77 → 1.0.79

package/README.md

@@ -1,76 +1,79 @@
 # @easynet/agent-model
 
-统一的 Model Hub：从 `model.yaml` 创建 LangChain ChatModel、Embedding 等模型。
+## Introduction
 
-## 最小接口
+`@easynet/agent-model` initializes the framework model layer from `model.yaml` and registers the configured chat model, embedding model, and optional VLM model into the default `AgentContext`.
 
-```ts
-import { AgentContextTokens, getDefaultAgentContext } from "@easynet/agent-common";
-import { createAgentModel } from "@easynet/agent-model";
+## API Reference
 
-await createAgentModel();
-const llm = getDefaultAgentContext().get(AgentContextTokens.ChatModel);
-const result = await llm.invoke("hello");
-console.log(result.content);
-```
+| API | What it does | Minimal usage |
+| --- | --- | --- |
+| `createAgentModel` | Load model config and register models into `AgentContext`. | `await createAgentModel({ configPath: "./model.yaml" })` |
 
-## 最小 YAML（推荐先用这个）
+## Usage
 
-`model.yaml`
+Create `model.yaml`:
 
 ```yaml
 llm:
-  # 默认使用 small 这个实例
   default: small
-
-  # 可按需保留 strong；不需要就删掉
-  strong:
-    provider: openai
-    base_url: ${LLM_BASE_URL}
-    model: ${LLM_MODEL}
-
-  # 本地 Ollama 最常见配置
   small:
     provider: openai
     base_url: http://localhost:11434/v1
     model: qwen3:0.6b
-
-embed:
-  default: gemma
-  gemma:
-    provider: openai
-    base_url: https://ollama-nvidia-8g-2.easynet.world/v1
-    model: embeddinggemma:latest
-    apiKey: ollama
-
-runtime:
-  check_connectivity: false
 ```
 
-## 最简单示例（带注释）
+Initialize the model layer:
 
 ```ts
-import { AgentContextTokens, getDefaultAgentContext } from "@easynet/agent-common";
 import { createAgentModel } from "@easynet/agent-model";
 
-async function main() {
-  // 1) 从当前目录 model.yaml 读取配置并创建模型
-  await createAgentModel();
-  const llm = getDefaultAgentContext().get(AgentContextTokens.ChatModel);
+await createAgentModel({
+  configPath: "./model.yaml",
+});
+```
 
-  // 2) 直接调用模型
-  const response = await llm.invoke("请用一句话介绍你自己。");
+### YAML config
 
-  // 3) 输出文本
-  console.log(response.content);
-}
+Example `model.yaml`:
 
-main().catch(console.error);
+```yaml
+apiVersion: easynet.world/v1
+kind: ModelConfig
+metadata:
+  name: agent-model-default
+spec:
+  llm:
+    default: small
+    small:
+      provider: openai
+      base_url: http://localhost:11434/v1
+      model: qwen3:0.6b
+
+  embed:
+    default: local
+    local:
+      provider: openai
+      base_url: http://localhost:11434/v1
+      model: qwen3-embedding:0.6b
+      apiKey: not-needed
+
+  vlm:
+    default: glm_ocr
+    glm_ocr:
+      type: image
+      provider: openai
+      base_url: http://localhost:11434/v1
+      model: glm-ocr:q8_0
+      apiKey: not-needed
 ```
 
-## 可选扩展
+Explanation:
 
-- 自定义 provider：`registerChatModelProvider(...)`
-- 只取默认模型配置：`getDefaultLlmConfig(...)`
-- 直接从对象构建模型：`createChatModelFromLlmConfig(...)`
-- 从 model.yaml 创建 EmbedFn：`createEmbedFnFromModelsConfig(...)`
+- `spec.llm`: chat model definitions.
+- `spec.llm.default`: the default LLM key to use.
+- `spec.llm.<name>.provider`: provider type, usually `openai` for OpenAI-compatible APIs.
+- `spec.llm.<name>.base_url`: base URL of the model service.
+- `spec.llm.<name>.model`: model name sent to the provider.
+- `spec.embed`: embedding model definitions used for retrieval or semantic search.
+- `spec.vlm`: optional vision model definitions for image-capable workflows.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@easynet/agent-model",
-  "version": "1.0.77",
+  "version": "1.0.79",
   "description": "Agent LLM: multi-provider, multi-model, simple chat/image API. Consumes agent.yaml llm section.",
   "type": "module",
   "bin": {