Operator

The Operator is Horizon’s central intelligence — an AI orchestration layer that sits above individual agents and coordinates their work toward your company’s goals. While agents execute specific tasks using skills, the Operator manages the bigger picture: deciding which agents should work on what, when to escalate issues, and how to optimize resource allocation.

What the Operator Does

Agent Coordination

The Operator routes tasks to the most suitable agents based on their skills, workload, and department assignment. When multiple agents could handle a task, it selects the one with the best availability and relevant context.

Goal Alignment

Every decision the Operator makes factors in your company goals. It prioritizes work that moves the needle on active objectives and deprioritizes tasks that do not contribute to stated targets.

Altitude Checks

The Operator performs altitude checks — structured assessments of your operations at strategic (30K ft), operational (15K ft), and tactical (ground level) depths — and surfaces actionable insights.

Anomaly Detection

The Operator monitors agent performance in real time and flags anomalies: unusual error rates, token consumption spikes, connection failures, and workflow bottlenecks.

How the Operator Thinks

The Operator maintains an internal model of your organization that includes:

• Company profile — your industry, size, description, and timezone.
• Goals and key results — what your organization is trying to achieve.
• Department structure — how your teams are organized and what each is responsible for.
• Agent capabilities — what each agent can do, based on its skills and connections.
• Historical performance — past task outcomes, error patterns, and throughput trends.

This model is updated continuously as agents complete tasks, connections change, and you update your company configuration.

Operator Decisions

The Operator makes several types of decisions throughout the day:

Task Routing

When a new task arrives (from a user request, a scheduled job, or a webhook trigger), the Operator evaluates:

1. Which agents have the skills needed for this task.
2. Which of those agents are available (not overloaded).
3. Which agent has the most relevant context (e.g., recent conversations with the same customer).
4. Whether the task aligns with any active goals (if so, it gets priority).

The selected agent receives the task and begins execution.

Escalation

If an agent encounters an error it cannot resolve, the Operator decides how to escalate:

• Retry — if the error is transient (e.g., a rate limit), retry after a delay.
• Reassign — if the error is agent-specific, assign the task to a different agent.
• Alert — if the error requires human intervention, notify the appropriate user via the configured notification channel.
• Pause — if the error indicates a systemic issue (e.g., a connection is down), pause related tasks until the issue is resolved.

Resource Optimization

The Operator periodically evaluates agent utilization across departments and may recommend:

• Moving underutilized agents to busier departments.
• Spinning up new agents when task queues are consistently long.
• Consolidating agents with overlapping skill sets to reduce token costs.

Note

The Operator never makes changes to your workspace automatically. It recommends actions and waits for a human administrator to approve them. You maintain full control at all times.

Viewing Operator Activity

Navigate to Company > Operator to see the Operator’s activity log. This page shows:

• Recent decisions — a timeline of routing, escalation, and optimization decisions with explanations of the Operator’s reasoning.
• Active recommendations — pending suggestions that require your review.
• Performance summary — metrics on how the Operator’s decisions have affected task throughput, error rates, and goal progress.

Configuring the Operator

You can adjust the Operator’s behavior from its settings panel:

• Decision aggressiveness — control how proactively the Operator redistributes agents and escalates issues. Options range from Conservative (minimal intervention) to Aggressive (frequent optimization).
• Goal weighting — adjust how heavily company goals factor into task prioritization.
• Escalation thresholds — define how many retries should occur before the Operator escalates to a human.
• Quiet hours — set time periods during which the Operator will not send non-critical notifications.

Tip

Start with the Conservative setting and increase aggressiveness as you gain confidence in the Operator’s recommendations. You can always review the decision log to understand why the Operator made specific choices.

Operator and Altitude Checks

The Operator is the engine behind altitude checks. When a check runs, the Operator:

1. Gathers current metrics from all agents, departments, and connections.
2. Compares them against your goals, historical baselines, and configured thresholds.
3. Generates a report with findings and recommendations.
4. Distributes the report to the configured recipients.

For more details on altitude checks, see Altitude Checks.