Quick start
From a blank workspace to a live, embeddable agent.
1. Create an agent
From Your agents, click + New agent and give it a name. Each agent has its own isolated knowledge base, conversations, and settings — nothing is shared between agents in the same workspace.
2. Add a data source
Open the agent, go to Data sources, and add at least one: a website (crawled and chunked automatically), an uploaded file, a pasted block of text, or manually written Q&A pairs. The agent can only answer from what you give it here — it won't fall back to general knowledge unless you tell it to in Instructions.
3. Write instructions
In Instructions, describe who the agent is speaking as, its tone, and any hard rules (e.g. always escalate billing questions to a human). Pick a model and temperature under the same tab — temperature controls how strictly it sticks to your sources versus how much it improvises.
4. Test it
Open Playground and ask it a real question. A grounded answer shows a citation back to the exact source chunk it used, with a similarity score — open the retrieval trace on any reply to see every candidate chunk that was considered, not just the ones it cited.
5. Give it something to do
Under Actions, add a template — Escalations (create a real support ticket), Collect leads, or a custom HTTP call to your own API — so the agent can do things, not just answer questions.
6. Deploy it
Open Channels and pick where it goes live: paste the Chat bubble script tag on your site, share the Help page link, or set up WhatsApp / Phone with your own Twilio credentials. Every channel answers from the exact same sources and instructions.
That's the whole loop — Zetra doesn't require any separate deploy step or build pipeline. Every change you save takes effect on the next message.