Primary agents

Primary agents are the agents you select for normal OpenCode collaboration. They share the same core safety and configuration blocks, but each one is tuned for a different style of work.

Inventory

Professional name	Mythic role name	Stable routing key	Model handler	Best for
`alpha`	`atlas`	`primary/alpha`	`primary`	First-line development help, architecture discussion, debugging, security review, and practical consultation.
`architect`	`daedalus`	`primary/architect`	`primary`	Architecture, RFCs, decomposition, system design, and solution shaping.
`developer`	`hephaestus`	`primary/developer`	`primary`	Focused implementation with verification discipline.
`orchestrator`	`sisyphus`	`primary/orchestrator`	`coordinator`	Longer-running coordination, decomposition, and execution planning.

Persona and display names

The primary.persona setting changes the role framing used in primary-agent prompts:

"professional" uses alpha, architect, developer, and orchestrator framing.
"mythic" uses atlas, daedalus, hephaestus, and sisyphus framing.

The primary.useRoleName setting controls whether display names follow the persona. Keep it false if you want stable visible names in the UI.

See Primary-agent configuration for the full set of primary.* fields.

Model routing

For model routing, prefer stable routing keys in models.agents:

{
  "models": {
    "agents": {
      "primary/alpha": "@provider/main",
      "primary/orchestrator": "@provider/coordinator"
    }
  }
}

Most users should route primary agents through handlers instead:

{
  "models": {
    "handlers": {
      "primary": "@provider/main",
      "coordinator": "@provider/main"
    }
  }
}

Use per-agent overrides only when one primary agent needs a different model from the others.

What is not a primary agent

beta runs in primary mode internally so /beta can inspect the current session, but it is not a selectable primary agent. It is a hidden slash-command agent with its own routing behavior. See Slash-command agents.

On this page