Agent runs and runtimes

Every time an agent does work in Helpin, it creates a run. A run records what the agent was asked to do, which record it worked on, every step it took, and how it ended. This page explains how runs start, what you can choose when you start one, and how to follow a run to the end.

How a run starts

Start	Example
Manually	You select Run now on an agent, or run an agent from a task, epic or document.
From a flow	A flow starts an agent when its trigger fires, such as a task entering a state or a schedule.
From a module	Support can run your support agent on new customer messages. Mentioning an agent in a document can start a run on that document.

Every run, however it starts, appears in Automation → Activity. See Automation areas: Flows, Agents, Activity and Catalog.

Start a run manually

Open Automation → Agents and select an agent.
Select Run now.
Under Target, choose what the agent should work on, such as a task, epic, repository, CRM record or the workspace. Only targets the agent is allowed to use are listed.
Pick the specific record. For a repository, you can also set a Base branch.
Optionally choose an AI profile. Leave Agent default to use the agent's own setting.
Optionally add Run instructions to focus this run.
Select Run now.

What you can choose

Setting	What it controls
AI profile	Which AI model connection the run uses. Profiles are set up in your AI settings.
Run mode	Autonomous runs work in the background from start to finish. Interactive runs can stop to ask you questions or request approval. This is set on the agent.
Approval policy	When a person must approve the agent's work. This is set on the agent.
Repository delivery	For task, epic and repository runs: Publish changes or Preview changes. See Coding runs and delivery.

Runtimes

All runs use Helpin's agent runtime, which carries out the agent's model and tool steps. You don't choose an execution engine for an agent. The choices that shape a run are the agent's settings and the AI profile.

Follow a run

Open a run from Activity, or from the task, epic or conversation it worked on. The run view shows the transcript, tool calls, plans and any previews the agent produced. The status tells you where it is:

Status	Meaning
Queued	Waiting to start.
Preparing or Starting agent	Loading context and setting up.
Agent working	The agent is active.
Approval, Input, Review or Sign-in	The run is paused and needs you. See Respond to approvals and questions during a run.
Completed	The run finished.
Failed	The run stopped with an error.
Cancelled	Someone stopped the run.

Stop or continue a run

To stop a run, select Cancel run in the run view.
After a run fails or is cancelled, you can type in Add instructions to continue this run... to pick up where it left off.
A completed run can't be continued. Start a new run instead.

Was this article helpful?