What are Agents?
Understand how Overseer agents encapsulate instructions, model choice, and tools.
Agents are Overseer's focused workers. They are not the whole system, and that is the point.
What an Agent Is
In Overseer, an agent is a reusable unit of behavior with a specific job, a model, a prompt, and an attached set of tools. The orchestrator chooses when to use an agent. The supervisor decides how that work should execute. The agent itself should stay narrow and understandable.
What an Agent Owns
- A name and role that make routing intent obvious.
- A model choice appropriate for that role.
- Prompt instructions scoped to one responsibility.
- A bounded tool list instead of universal access.
Operator Rule
When to Create One
Create a new agent when you need a stable reusable role such as release analysis, workflow summarization, desktop execution, or report synthesis. Do not create an agent just because a prompt got long once.
Anti-Patterns
- One "do everything" agent with every tool attached.
- Prompts that describe personality but not execution boundaries.
- Using an agent to hide orchestration logic that belongs in the orchestrator.
- Marking success without evidence from a tool, trace, or supervisor result.
Next Steps
Continue with Creating Agents to design one that the orchestrator can actually route to reliably.