The agent editor, card by card
Every setting on the agent page in NexaVoxa, what it actually controls, and which ones you can ignore until you have a reason not to.
7 min readUpdated 26 August 2026
Select an agent from the list on the left and every setting it has fills the page. It looks like a lot the first time. It is not. Two cards decide almost everything about how a call goes, and the rest ship with defaults that hold until you have a specific reason to change them.
This page walks the editor top to bottom so you know what each thing does before you touch it.
What sits above the cards
The strip at the top carries the agent's name, its ID, and whether it is Active. On the right is Talk to your agent.
That button is the most useful control on the page. Every change you make below is a guess until you have heard it, and hearing it takes about fifteen seconds. Use it after every meaningful edit rather than saving up ten changes and then wondering which one caused the problem.
Tip
The agent ID is worth copying when you contact support. Ten agents called some variant of Front Desk is normal; the ID is unambiguous.
How your agent runs
The first choice on the page, sitting above everything else because it frames all of it. The page states the difference plainly: a single prompt means one set of instructions drives the whole call, and you switch to a guided flow for complex, multi-step calls.
Start with Single prompt. The overwhelming majority of agents work this way, and when something goes wrong you edit a paragraph instead of rewiring a diagram. There is a separate article on when a flow genuinely earns its complexity.
The five cards at a glance
| Card | What it controls | Touch it on day one |
|---|---|---|
| Create Agent | Name, who speaks first, welcome message, whether the greeting can be interrupted | Yes |
| Instructions | The prompt the agent follows on every call | Yes |
| Voice & model | Model, voice, language | Only if the default sounds wrong |
| Knowledge base | Which of your documents it can search mid-call | Only if it needs your content |
| More to configure | Seven panels: AI & voice models, Call settings, Conversation intelligence, Tools, Data capture, Workflow, Embed | No |
Create Agent
The card is named for creation but it is also where you edit an agent that already exists. The page describes it as naming your agent and deciding how the call opens. Four settings:
- Agent name. This is what you pick out of a list once there are twenty of them, so name it after the job and the place, not after the model.
- Who speaks first, a choice between Agent and Caller.
- Welcome message, with the placeholder "Greet the user."
- Greeting can be interrupted, off by default.
Should the agent or the caller speak first?
This is a small toggle with a large effect on how the call feels, and the right answer depends entirely on why the phone is ringing.
| Setting | The call opens with | Suits |
|---|---|---|
| Agent | Your welcome message, immediately | Inbound lines where the caller may not know who they have reached |
| Caller | Silence, until the caller speaks | Calls where the person already knows exactly why they rang |
An agent that speaks first takes control of the call and sets expectations, which is usually what an inbound support or reception line wants. An agent that waits feels less like a machine answering, which suits a line people ring with a specific request already in mind.
Should callers be able to interrupt the greeting?
The page explains both positions directly: on, callers can talk over the welcome message and the agent stops; off, which is the default, the greeting always plays to the end.
The default suits a short greeting. It stops the agent being cut off by background noise or by the click of the line connecting, which is a real problem on phone calls and makes an agent look broken in the first two seconds.
The moment your welcome message runs longer than a sentence, turn interruption on. A caller who already knows what they want should never be held hostage by a greeting, and forcing them to wait is the fastest way to make an automated line feel worse than a queue.
Careful
If callers routinely start talking before the agent finishes, that is not an interruption problem. It is a greeting that is too long. Fix the message before reaching for the toggle.
The remaining cards
Instructions is where call quality is actually won or lost, and it has its own article. So does Voice & model.
Knowledge base attaches a collection of your own content so the agent can answer from your material rather than from what the model happens to know.
More to configure is where the rest lives, and it holds seven panels rather than the handful it looks like at a glance. The page describes them as voice and models, call handling, tools, data capture, workflow and embed, and says you can open any section to fine-tune the agent. Each one has its own article.
Why is my change not affecting calls?
Because editing is not publishing. Changes to an existing agent have to be saved and published before a caller hears them, and before you can test.
This is deliberate rather than an annoyance. It means you can rewrite a prompt in the middle of a working day without touching what live callers are getting until you are ready.
An agent created from a template or from a description is published as it is created, which is why a brand new agent can be tested immediately and an edited one cannot.
Common questions
- Why can I not test my agent?
- Testing requires a published agent. If you have just edited one, save and publish it first. A newly created agent is published automatically, so this only bites after an edit.
- Do I have to fill in every card before the agent works?
- No. A name, instructions and the default voice and model are enough for a working call. Knowledge base, tools and the rest are additions, not requirements.
- What is the agent ID for?
- It identifies the agent to the API and to support. Quoting it in a support request saves a round of questions about which agent you mean.
- Can I copy an agent instead of building a new one?
- Yes, duplicate it from the agent list. Once you have one agent you are happy with, duplicating it is usually faster than starting from a template.
- What does the Active badge mean?
- That the agent is published and able to take calls. An agent that has been edited but not republished still shows what is currently live, not your unsaved changes.
Was this helpful?