Agents

Bring the agents you already use.

Any of them can be lead, coder or reviewer, and you can mix kinds and models in one team: a small fast model for simple milestones, a stronger one where reasoning is the hard part. These kinds ship in agents.default.toml. Run orch agents to see which are installed on your machine, and orch agents sync after installing a new one.

Only one of them installed? That is enough. Open one pane and start the plan there: the lead runs orch team <kind> to build coders and a reviewer from the same tool, and orch team up opens their panes when you say go. See One tool, whole team.

AgentLaunch commandPick its modelStart it as leadStatus tracking
Claude Code
claude
claude --permission-mode acceptEditsorch agents launch claude --model <id>/orch-plan <feature>Live (idle, working, blocked)
Codex
codex
codex --approve-for-meorch agents launch codex --model <id>$orch-plan <feature> (or: use the orch-plan skill: <feature>)Live (idle, working, blocked)
Amp
amp
ampInside the tool; write it in the planuse the orch-plan skill: <feature>Live (idle, working, blocked)
OpenCode
opencode
opencodeorch agents launch opencode --model <id>/orch-plan <feature>Live (idle, working, blocked)
Hermes
hermes
hermesInside the tool; write it in the planuse the orch-plan skill: <feature>Live (idle, working, blocked)
Antigravity
agy
agy --mode accept-editsInside the tool; write it in the planuse the orch-plan skill: <feature>Live (idle, working, blocked)
Any other agent
Add it to ~/.config/herdr-orchestra/agents.tomlSet model_flagRun `orch lead plan` and follow it. Feature: <feature>Set herdr_detect

The lead is whichever agent you start the plan playbook in. It plans, dispatches and decides, and never codes or reviews in its own run. Agent setup and adding your own → · Models and tiers →

Agent marks are trademarks of their owners and are shown only to identify compatible tools.