How to Build a Breakfast Shop AI Agent with Baidu Wenxin and DeepSeek R1

Wenxin Intelligent Agent Platform – Core Features

Zero‑code natural‑language creation : Configure agents by describing requirements in plain language; the system generates the underlying configuration automatically.

One‑click audiovisual scene setup : Choose from preset backgrounds and voices or customize them to give the agent a distinct personality.

High‑compatibility knowledge‑base : Supports large‑scale ingestion of multi‑format, multi‑source data, enabling rapid construction of both generic and domain‑specific agents.

Extensible plugins and workflows : Official and partner plugins are available, and developers can create custom plugins to add complex functionality.

Cross‑ecosystem deployment : Agents can be distributed to Baidu Search, Xiaodu, YiYan, Maps, in‑car infotainment and other devices.

Monetization hooks : Supports link mounting, lead collection, and product sales.

Data‑flywheel optimization : Provides QA tuning, agent diagnostics, and model‑level feedback loops for continuous improvement.

Building a Breakfast‑Shop Reception Agent with DeepSeek R1

Open the Wenxin Agent platform at https://agents.baidu.com/center.

Click “Create Agent”.

Enter the persona and response logic. An example configuration is shown below:

#角色规范
你是“早餐店智能助手”，是一家中式早餐点的全能帮手，帮顾客解答菜品问题，点餐问题，结账等，拒绝回答用户除了餐饮吃喝品菜之外的问题，不要在思考过程中暴露跟人社相关的内容。

#思考规范
以下点菜注意事项，要严格遵循！！
1. 一般一个人的食量为一笼包子或者两根油条，或者两个馅饼，一个油条。
2. 早餐注意搭配一份粥或豆浆等带汤食物。
3. 荤素搭配均衡，避免过多油腻食品。
4. 考虑顾客的饮食偏好和健康需求，提供适当的建议。
5. 当顾客决定下定后请给出选定的订单和价格。

#菜单数据
---markdown形式的菜单，过长不展示

#回复规范
1. 使用轻松有趣的语气与用户交流，提升用户的点餐体验。
2. 在回复中可以适当加入一些对菜品的描述或推荐理由，增加互动性。
3. 保持回复的简洁明了，避免使用过于复杂的词汇或句子结构。
4. 在结束时，可以询问用户是否有其他需求或是否满意当前的点餐建议。

Select the model “DeepSeek R1”.

Test the agent directly in the right‑hand panel of the interface.

The platform now offers the full‑capacity DeepSeek R1 model. Switching to this model improves reasoning, style, and overall interaction quality without additional fine‑tuning.

Key technical observations:

Knowledge‑base construction accepts various file formats; the platform parses and integrates them automatically, dramatically reducing development time.

Plugins act as “super‑power accessories”, allowing agents to handle complex scenarios, automatic follow‑up questions, and long‑term memory across sessions.

With DeepSeek R1’s strong inference, developers can focus on prompt engineering; fine‑tuning is generally unnecessary for most applications.