Skip to main content

Agents

Agents are the heart of HubTalk AI v4. They are the voice interfaces you build once and run for customers, combining speech, prompts, and business actions into a single experience.

What agents do

An agent can:
  • greet callers and answer questions
  • collect customer data during the conversation
  • perform handoffs to a human operator
  • end the call cleanly when the task is completed
  • send call data to your systems through webhooks

Create your first agent

Open the Agents section and click Create Agent. A new agent opens immediately in the builder interface. The creation screen shows the agent setup panel with sections such as Functions, Variables, Post Call Analysis, VAD, Call Settings, Speech settings, Webhook Settings, and Knowledge Base — all of them described in Agent settings. Create agent button and the settings panel right after creation

Where to start

Start from a clear goal for the call: support, booking an appointment, confirming an order, qualifying a lead. Everything else follows from it. Then build a minimal flow and finish that one rather than every branch at once:
  1. Welcome message — the first thing the caller hears. Friendly and clear; this is also where you decide who speaks first.
  2. Intent and questions — what the agent has to learn from the caller, and what counts as success.
  3. Handoff and ending conditions — when the agent gives the call to a human, and when it says goodbye itself.
After that configure the language and voice, the prompt and the actions — transfers, ending the call, webhooks and post-call logic. Once the happy path works end to end, start adding branches.
Publish first, hand out afterwards. A member can only pick a published agent for a campaign, so the first publication makes the agent visible to the whole organization. An agent that is still being built is hidden before that — see The agent list and access.

Where everything is documented

The agent list and access

The agent table and organization folders, renaming and deletion, the Visible to members setting and what exactly it hides.

Flow nodes

Nine node types — Conversation, Function, Code, Logic Split, Extract Variables, Transfer, Component, MCP Tool, End — and the fields of each.

Working in the editor

Document search, agent variables, the issues tab, testing the flow and starting a test from a selected node.

Agent settings

Prompt and models, voice, recording and call control, conversation goals, analytics and webhooks.
After the agent is ready, connect it to telephony and configure webhooks so call results are visible to your team.