How-To Guides / User

Create an Agent

An agent is work you hand over: an event starts it, a policy bounds it, and you keep the record of what it did. This walks the builder end to end.

User

Agents

Before you start

Where this happens, and what you need first

Starting screen: in the Telara dashboard, open Agents → Agent Factory in the left sidebar. That page is the execution hub — its tabs are Agents, Tasks, Runs & Executions, and Approvals. Click Create agent to open the builder.

An agent cannot be saved with nothing it can do. Before you start, have at least one of these ready: a connected integration, a catalog agent, an attached permission policy, or a workflow you intend to draw. If you have none of them, connect an integration first — the builder will otherwise block Save at the last step with “This agent has nothing it can do yet.”

Overview

Two execution modes, chosen first

Mode is step one and it decides which steps follow. On an existing agent it is derived from the definition and shown as locked.

Autonomous

A trigger starts the run, the agent gathers context step by step, and approved response tools act when needed. Best when the shape of the work is not known up front.

Mode → Identity → Triggers → Investigation → Approvals → Outputs

Workflow

You lay the steps out before the run on a DAG canvas, branch on known conditions, and every run follows the saved path. Best when the work has a known shape.

Mode → Identity → Triggers → Workflow → Outputs

Steps

How to create an agent

Open Agent Factory → Create agent

From Agents → Agent Factory, click Create agent. The builder opens with a stepper down the left; Next advances it and you can also click a step directly once a mode is chosen.

Pick an execution mode

Choose Autonomous or Workflow. You can also describe what you want in plain language here and let Agent Architect propose the agent — you review the proposal in the same builder before anything is saved. Nothing else unlocks until a mode is set.

Name it on the Identity step

A name is required; everything else on this step is optional. Scope and permission policies live here too, and the MCP configuration the agent runs against is provisioned on save if you do not choose one.

Add a trigger

Triggers are the event sources that start a run. An autonomous agent needs at least one configured trigger before it can be saved. A half-filled trigger blocks Save in both modes — finish it or delete it, rather than leaving it as a draft, or the agent would be saved without the event that starts it.

Define what it may do

In Autonomous mode this is the Investigationstep — the read actions the agent may take to gather context — followed by Approvals, where you add human-in-the-loop gates. In Workflow mode it is the Workflow canvas, where you lay out the nodes.

Set outputs

The Outputs step covers response formats and notifications. Every output you enable needs both a destination action and a credential; one without the other blocks Save.

Save

On the last step the button reads Save Agent. If it reads Complete Required Fieldsinstead, the reason is printed directly beneath it — missing name, unfinished trigger, nothing the agent can do, or an incomplete output.

Result

What you should see when it worked

Saving redirects you to the agent's own page. Its tabs are Overview, Actions, Learnings, Executions and Webhooks. There is no separate activate step — a saved agent runs when its trigger fires.

Runs appear on the agent's Executions tab, and across all agents under Agents → Observability.

Known issue: the run count shown against each row of the agents list currently reads 0 regardless of how many runs the agent has had. The totals at the top of the page, and the Executions tab, are the accurate ones.