Yesterday I set up my OpenClaw, running the Hugo’s Bistro demo. Today the next step is to make the information available to the customer by adding a chatbot.
Day 2 — Build your first chatbot
- Create a free Voiceflow account
- Build a chatbot for “Hugo’s Bistro”
- Program it to answer: hours, location, menu, booking, parking, dietary options, pricing, private events, wait times, contact info
- Test it thoroughly — try to break it with weird questions.
I didn’t know Voiceflow, but I created an account and then had Claude Cowork start to build the chatbot. After a few minutes on Claude’s Sonnet 4.6 model it had already reached my usage limits for the day without any visible results in Voiceflow. Super annoying! All after just setting up a connection to my OpenClaw instance and querying the opening and closing times of Hugo’s Bistro once.
Anyway, I just switched to my trusted Codex instead to see if it could run without annoying usage limits. Of course it ran without any hiccups and configured Voiceflow via the web interface. After that it automatically set up an API key endpoint so it could do future updates without depending on brittle browser automations.
Claude barely touched the task. Then Codex completed it while also improving the method how it did it.
Computer use is so cool when it works. Just seeing entries appear in a website out of nowhere is magical. When Codex was ready it had set up everything in Voiceflow correctly.
I only had $5 in free credits in Voiceflow which should be sufficient for this test case. But to make sure I was not flying through the credits I opted for the GPT-5 mini model in Voiceflow.
I asked my first question to the chatbot, and it answered well. Based on this I prompted about 20 different questions to the chatbot. Each one a bit weirder and less in line with expected customer questions.
I had a list of gradually weirder questions prepared. Starting from basic things like “Is your outdoor seating open when it rains?” to “Has anyone ever proposed marriage in your restaurant, and if yes, which table has the highest success rate?”
That’s a fun question! I honestly don’t have that kind of data on hand — and I wouldn’t want to guess which table has the magic touch.
If you’re planning something special, I’d love to help make it memorable. Feel free to reach out to the team directly and they can help arrange the perfect setting for a proposal.
Would you like help with a reservation, or is there anything else I can do for you?
So far I am quite pleased with the progress, it helps me get more into the flow of the project. It helps me to continue without quitting after the first hurdle.
Key Insight:
A chatbot is not proven by answering easy FAQ questions. It is proven by how it behaves when the user stops behaving normally.