Case: maintain a restaurant assistant through WeChat

We will use fictional Westbank Restaurant. The owner prepares a menu, explains reception rules, and publishes the assistant for customers without writing code.

Try it · Start a DeepAgents admin conversation to prepare the restaurant workflow.

Upload the menu and reception notes into the restaurant folder. Keep prices in the menu and escalation rules in the notes. Share only information intended for customers.

Try it · Upload two synthetic files and inspect the menu price.

Tell your private assistant to use menu prices and refer six-person private-room requests to the owner. Review its understanding, then save the confirmed reception prompt version.

Try it · Explain the reception boundary and save the confirmed Prompt version.

Before publishing, test menu and private-room questions. Check the quoted price and whether the booking waits for confirmation. Revise the rules and test again if needed.

Try it · Try demonstration menu and special-booking questions.

Create the restaurant Service with a name, authorized subscription connection, and model. In this example, share its entry only with invited, trusted customers.

Try it · Open the creation form, enter the name, and select the engine.

Choose the saved reception prompt version and allow only the menu and reception notes. Select Publish immediately, then Save. These choices determine which documents customers can access.

Try it · Select the prompt and two documents, then show publish-on-save in the demonstration.

First set DeepAgents as the new-chat default in Models & connections. The owner then scans in WeChat connection. Customers scan the separate Service channel.

Try it · Customers scan the Service channel. The owner sets the DeepAgents default, then scans in WeChat connection.

The customer asks about today’s menu in Service WeChat. The assistant reads the shared menu and quotes CNY 168. It does not take orders or payments.

Try it · Show the customer’s menu question and illustrative reply.

For a private-room request, the assistant collects the time and contacts the owner. Numbered feedback enters Inbox and a WeChat notification is attempted. Submission does not confirm the booking.

Try it · Submit the demonstration booking request and show its feedback ID.

After receiving the notification in their own WeChat, the owner replies with the feedback ID. The response is queued for the original conversation; delivery can be checked in the web Inbox.

Try it · Send an ID-based demonstration reply from administrator WeChat and locate the web Inbox.

In the owner’s DeepAgents WeChat conversation, ask to change the shared menu price from 168 to 198 yuan. Specify the file, then check the saved shared document in the file panel.

Try it · Name the shared menu file, request the edit, then verify the saved result on the web.

Ask again through the customer entry and request a fresh menu read. Verify CNY 198. Updating the document does not rewrite earlier replies; this completes the reception and maintenance loop.

Try it · Send a fresh customer question and inspect the price after rereading.