Agent Management

Agents are the core workers in Horizon. Each agent is an autonomous AI entity configured with specific skills, connected to your business tools, and assigned to a department. The Agents page in your workspace is where you manage the full lifecycle of every agent.

Agent List View

The agent list displays all agents in your organization in a sortable, filterable table:

Column	Description
Name	The agent’s display name.
Status	Active, Idle, Paused, or Error.
Department	The department the agent belongs to.
Skills	Number of skills assigned to the agent.
Conversations	Total conversation threads the agent has participated in.
Last Active	Timestamp of the agent’s most recent action.
Token Usage	Tokens consumed by this agent in the current billing period.

Use the search bar to find agents by name and the filter dropdowns to narrow by status or department.

Creating a New Agent

1. Click the + New Agent button in the top-right corner.
2. Enter a name and optional description for the agent.
3. Select a department to assign the agent to.
4. Choose an agent template from the Store or start from scratch.
5. Assign skills from your installed skill library. You can add or remove skills later.
6. Configure connections — select which integrations the agent can access (for example, Salesforce, QuickBooks, Slack).
7. Set the agent’s behavioral instructions — a natural-language prompt that defines the agent’s personality, goals, and constraints.
8. Review the summary and click Create Agent.

Note

Agents can only use connections that have already been set up in your workspace. If you need a new integration, visit Connections first.

Viewing Conversations

Each agent maintains a history of all its conversation threads. To view them:

1. Click on an agent’s name to open the agent detail page.
2. Select the Conversations tab.
3. Browse conversations sorted by most recent activity.

Each conversation shows the full message history between the agent and any users or other agents it interacted with, along with skill execution results inline. You can search within conversations and filter by date range.

Conversations are the primary audit trail for agent actions. Every skill execution, decision, and external API call is logged as part of the conversation thread.

Agent Configuration

The Configuration tab on an agent’s detail page lets you adjust:

• Skills — add or remove skills. Changes take effect on the agent’s next conversation turn.
• Connections — grant or revoke access to specific integrations.
• Behavioral instructions — update the agent’s system prompt to refine its behavior.
• Memory — view and manage the agent’s long-term memory store. You can clear specific memories or reset the entire memory.
• Model settings — choose the underlying LLM model and adjust parameters like temperature and max tokens.
• Rate limits — set maximum tasks per hour or per day to control costs.

Tip

Use behavioral instructions to set boundaries. For example: “Only create invoices after confirming the line items with the user” prevents the agent from taking unreviewed actions.

Agent Status

Agents can be in one of four states:

• Active — the agent is currently processing a task or conversation.
• Idle — the agent is running but has no active tasks.
• Paused — an administrator has manually paused the agent. It will not accept new tasks until resumed.
• Error — the agent encountered a failure and requires attention. Check the agent’s logs for details.

To pause or resume an agent, use the toggle on the agent’s detail page or the quick-action menu in the list view.

Monitoring and Performance

The Analytics tab provides per-agent metrics:

• Task completion rate and average duration
• Token consumption over time
• Error frequency and common failure reasons
• Conversation volume trends

Agent Concepts Learn about the app-agent-skill hierarchy and how agents fit into the bigger picture.

Store Agents Browse pre-built agents available in the Horizon Store.