@dograh/ sdk (英语)

Dograh语音-AI工作流程的打入构建器. 获取节点光谱目录 dograh 后端在会话开始时, 验证每个反对它的呼叫 调用网站,并生产通过 Python 绕行的线条格式 JSON ReactFlowDTO。 。 。 .

安装

npm install @dograh/sdk
# or
pnpm add @dograh/sdk

对于本地开发的取出单列,请添加 tsconfig 路径 条目 :

{
  "paths": {
    "@dograh/sdk": ["../sdk/typescript/src/index.ts"]
  }
}

使用量

import { DograhClient, Workflow } from "@dograh/sdk";

const client = new DograhClient({
  baseUrl: "http://localhost:8000",
  apiKey: process.env.DOGRAH_API_KEY,
});

const wf = new Workflow({ client, name: "loan_qualification" });

const start = await wf.add({
  type: "startCall",
  name: "greeting",
  prompt: "You are Sarah from Acme Loans. Greet the caller warmly.",
  greeting_type: "text",
  greeting: "Hi {{first_name}}, this is Sarah.",
});

const qualify = await wf.add({
  type: "agentNode",
  name: "qualify",
  prompt: "Ask about loan amount and timeline.",
});

const done = await wf.add({ type: "endCall", name: "done", prompt: "Thank them." });

wf.edge(start, qualify, { label: "interested", condition: "Caller expressed interest." });
wf.edge(qualify, done, { label: "done", condition: "Qualification complete." });

await client.saveWorkflow(123, wf);

客户端验证

每个 add() 调用以获取的光谱验证 kwargs . ValidationError 当:

• 传递未知字段(捕获物打字)
• 缺少或空白所需的字段
• scalar 类型错误( 如布尔字符串)
• a 一个 options 数值不在允许的列表中

当一个Spec带一个 llm_hint,提示附在错误之后,所以是一个 LLM 代理在重试时可以自我更正 :

tool_uuids: expected tool_refs, got string
  Hint: List of tool UUIDs from `list_tools`.

服务器侧 Pydantic 验证器运行到保存并显示客户端的任何内容 让我们通过.

环境

DOGRAH_API_URL=http://localhost:8000   # default
DOGRAH_API_KEY=sk-...                  # sent as X-API-Key

许可证

BSD 2- Clause — 见 LICENSE。 。 。 .