Skip to main content

Autonomous Agents

Autonomous Agents in Arrotech Hub are stateful, intelligent entities that execute workflows based on triggers (schedules, events, or manual calls). Unlike simple scripts, Arrotech agents can observe their environment, reason about the best course of action using LLMs, and use tools to achieve a goal.

Core Concepts​

1. The Agent Persona​

Every agent can be configured with a Persona. This is a system-level prompt that dictates how the AI reasons. For example, a "Sales Agent" might prioritize lead conversion, while a "Security Agent" might prioritize data validation.

2. Execution Orchestrator​

The Orchestrator is the engine that drives agent behavior. It follows a loop:

  1. Perceive: Read input data or the current environment state.
  2. Plan: Decide which tools or workflow steps are needed.
  3. Act: Execute a tool call (e.g., Send a WhatsApp message).
  4. Learn: Observe the result and adjust the next step.

Agent Lifecycle​

  1. Deployment: Use the /agents/create API to deploy an agent from a validated workflow.
  2. Activation: An agent is typically "Passive" until a trigger occurs.
  3. Running: During execution, the agent's status changes to active.
  4. Monitoring: All agent actions are tracked in real-time, providing metrics on success rates and token usage.

API Integration​

To programmatically manage agents, refer to the following endpoints in our API Reference:

info

Agents respect Universal Guardrails. No matter how an agent is configured, it cannot perform actions that violate your workspace's security policy.