Create your first agent
Name it, pick a model, write its instructions and set its temperature.
On this page
An agent is the AI assistant your customers talk to. You can have several per workspace; how many depends on your plan.
Create the agent
New agent starts an interview: describe your business, answer a few questions, and the agent is built from your answers. See Build an agent with AI.
To set one up by hand instead:
- Open Agents in the sidebar, select New agent, then Set up manually.
- Give it a name, which customers will see, and choose a model.
- Create it. You land on its Settings tab.
Every agent has the same tabs, in the order you will usually use them:
| Tab | What you do there |
|---|---|
| Settings | Name, description, instructions, model and temperature |
| Knowledge | What it knows: files, websites, text, Q&A |
| Actions | What it can do: call APIs, collect leads, hand over |
| Procedures | Step-by-step playbooks it follows |
| Playground | Talk to it and test changes |
| Deploy | Put it on your website, and style the widget |
| Analytics | How its conversations are going |
Only owners and admins can create and change agents. Other roles can see them.
Write good instructions
Instructions tell the agent who it is and how to behave, and they are sent with every message. Short and specific works best:
You are the support assistant for Kigali Shoes, an online shoe shop.
Answer questions about orders, sizes, delivery and returns.
Be friendly and brief: two or three sentences unless the customer asks for detail.
Answer only from the knowledge you are given. If you do not know, say so
and offer to hand the customer to our team.
Never promise a refund; explain the returns policy instead.
Tips:
- Say what it should not do as clearly as what it should.
- Do not paste facts here. Prices, policies and product details belong in Knowledge, where the agent can look them up and cite them.
- Change one thing at a time and test it in the Playground.
Choose a model
The model is the AI that writes the replies. Models differ in quality, speed and cost:
- Each model shows the credits it costs per reply. Capable models cost more.
- Start with a mid-priced model, test it on real questions, and only move up if the answers fall short.
- If a model is ever withdrawn by its provider, the agent shows model unavailable. Pick another to keep it answering.
Temperature
Temperature controls how varied the wording is.
- Low (0–0.3): consistent and factual. Best for support.
- Higher (0.7 and up): more varied and creative. Better for marketing copy, rarely for support.
Duplicate and delete
From the agent's menu you can Duplicate an agent, which copies its settings so you can try a variation safely, or delete it. Deleting removes the agent and all its knowledge for good.
Related
- Build an agent with AIDescribe your business and answer a few questions; the agent is built from your answers.
- Knowledge sourcesTeach an agent from files, websites, text and questions with their answers.
- Test in the playgroundTry changes without saving, compare two models and inspect exactly what a reply was given.
Something not working as described here? See Troubleshooting.