Creating an agent

Status: Current | Last reviewed: 2026-09-23

An agent is the thing that talks to your customers. You create one in the Agents section of TrueServe, give it an identity, then build out its behavior in the pipeline editor.

Prefer to describe what you want and let an AI assistant build the first draft? See Generate with AI (if enabled for your tenant).

Browse the agent repository

The Agents repository brings conversational agents and background agents together, limited to the types your role can access. It opens in Card view; use List view for compact rows. Type filters show Voice, Chat, Copilot, Translation, and Background.

Search names, descriptions, IDs, types, or tags. My favorites, My agents, and My recents narrow the list. Open Filters for publication/activation status and created/modified dates; use Clear all to reset. Sort by name, type, status, creation time, last modification, or recently opened. Large lists have page controls at the bottom.

Cards show publication state separately from Draft available and a background agent’s Active/Inactive state. Click the card body or list row to open the editor; use the heart/star for favorites and ⋯ Manage for administrative actions. Older workflow links still work and open the repository with Background selected.

Create the agent

  1. From the left nav, open Agents.
  2. Click New agent in the top right and choose Voice agent, Chat agent, Copilot or Translation agent. Background agent opens the workflow creation dialog.
  3. Fill in:
  4. Name — what people in your tenant will see. Be specific ("Auto Claims Triage" beats "Bot 1").
  5. Description — one or two sentences explaining what the agent is for. This shows up in pickers when wiring the agent into a workflow or phone number.
  6. Agent type — voice, chat, or multimodal. You can change this later but most properties of the pipeline are scoped to the type.
  7. Surface — which embedded UI shell the agent renders in (only relevant for chat and multimodal). Pick "default" if you're not sure.
  8. Click Create.

You land on the new agent's detail page with one draft version pre-created.

Pick a stack (the default)

The recommended starting point is Pick a stack: choose a technology family (Amazon, OpenAI, Google, or self-hosted) and the channels the agent must serve (voice, chat, or both). The wizard starts your draft on that family's best-practice voice pipeline — or its chat pipeline for chat-only agents — and when you publish, the platform derives the remaining variants (cascade and backup pipelines, a chat variant) automatically. Everything stays fully editable in the pipeline editor; families without a speech-to-speech service (self-hosted) run voice as a cascaded pipeline.

After you pick a family, the step lists an integration select for each provider the family uses (for example AWS and Deepgram for the Amazon stack). If your tenant has exactly one matching integration it is picked for you; if there is none, the wizard tells you the agent will need credentials before it can run — you can still continue, then add one later in Settings → Integrations. The picks apply to every derived variant.

The full set of stack variants is materialized when you publish. The former Variants section is hidden from Agent Settings → Routing Rules; stored runtime profiles and publication behavior are unchanged. Agents that serve both voice and chat retain the Voice | Chat toggle in Preview so you can test both channels without leaving the editor.

Find and copy an agent ID

In the agent repository, select Manage on a card or row, then Copy ID. The displayed ID is the identity's base ID; use versioned references where an integration requires a particular version.

What's in an agent

An agent is the identity: its name, description, surface configuration, and a list of versions. Everything that controls how it talks — speech-to-text provider, language model, voice, system prompt, tools, workflows — lives on its versions. See Agent versions for the version model.

When you choose Custom, the first version is empty. Open the pipeline editor to add stages and choose providers deliberately. Template and Generate with AI starts may already contain configured stages.

Naming convention

Tenants commonly use a Domain - Purpose shape:

That makes agents easy to find when your tenant has dozens of them.

Deleting an agent

Open the agent's detail page, click ⋯ in the top right, and choose Delete. Deleting an agent removes all its versions, all its conversation history, and any phone-number bindings. This cannot be undone.

Warning: Deleting an agent that's currently bound to a live phone number drops calls in flight. Move the binding to a different agent first.

See also