Turn spoken orders into POS actions
A custom LLM agent converts staff speech into API calls that add or remove items in a store POS and start card or QR payment. ·
Done with a custom-built or unnamed agent
- What they did
- The author built a LangChain agent on gpt-3.5-turbo with three tools: add items, remove items and pay. Each tool is a function that sends a request to a FastAPI endpoint standing in for the POS. A headset listens to the whole conversation. The prompt tells the agent to act only on speech between the cue words 'agent' and 'please' and to ignore everything else.
- How it went
- In the demo the agent picked out the relevant orders from the chatter, adjusted the order after a change of mind, and then started card payment. The source gives no accuracy or speed numbers, and it does not report a real store deployment.
- Worth knowing
- The agent decides which tool to call from the function names and docstrings, so write the docstrings as instructions to the LLM, including the exact data format you expect.
Try it yourself with n8n
Build a workflow that takes spoken or transcribed staff orders like [example phrase] and uses an LLM to turn them into API calls against [my POS system] to add or remove items on the current order. Test it with sample phrases in a sandbox first. Require a staff confirmation step that shows the parsed order before any item changes or payment starts.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: towardsai.com · Nov 17, 2026 · Submitted by from search gap: 收银系统
Related use cases
Run a solo company with eight agent departments
A founder set up CEO, CFO, COO, lawyer, accountant, marketing, CTO, and improver agents that share memory and consult each other.
Catch and prepare an unsent invoice for approval
An early tester's dot noticed an invoice to a publication was unsent and prepared it for approval.
Reschedule meetings and send daily briefings
OpenClaw checks availability, updates events and drafts confirmations for reschedule requests, and sends daily briefings via messaging apps.
Email daily ClickUp digests to stakeholders
A Make.com scenario sends a daily email digest from ClickUp to stakeholders without ClickUp accounts, and adds contacts to a selected action item.
Let an AI CEO run a small security scanner business
The poster gave an AI agent a security scanner business idea with three AI employees; at day 2 there was no revenue and a homepage outage.
Run an online store by texting commands to an agent
A reference build where an OpenClaw agent runs store API calls from the owner's Telegram texts, while a buyer-facing agent handles WhatsApp orders and status questions.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.