Agents
An agent is the core execution context in Hoziron. It wraps an LLM conversation with a persistent identity, a set of competencies, memory, and a lifecycle you control. Agents are the unit you deploy, monitor, and scale — every task your platform handles runs inside one.
You create agents when you need a stable, addressable AI worker: a claims processor that handles incoming documents, a compliance reviewer that checks policy updates, or a customer-facing assistant that lives behind a channel integration.
Agents don't do much on their own — they gain capabilities through competencies (what they can do) and connect to the outside world through skills and data integrations (what they can access).
What you'll find here
| Guide | Description |
|---|---|
| Creating agents | Declare agents via manifests, templates, or the API — including auto-load at startup |
| Lifecycle | The state machine that governs agent execution: created → running → suspended → stopped |
| Sending messages | Interact with agents through chat, one-shot invocations, the REST API, or channel bridges |
| Scheduling | Trigger agents on cron schedules or in response to external events |
Common tasks
- Spin up a quick test agent → Creating agents § From the CLI
- Gracefully restart an agent without losing context → Lifecycle § Suspend and resume
- Send a one-shot prompt and get a structured response → Sending messages § One-shot mode
- Run an agent every morning at 8 AM → Scheduling § Cron expressions
- Understand what happens when an agent errors → Lifecycle § Error states
How agents relate to other concepts
Related
- Competencies — what gives agents the ability to act
- Workflows — orchestrating multiple agents together
- Providers — the LLMs that power agent reasoning
- Data integration — connecting agents to external systems