🎼 Orchestrator Mode Is Deprecated — Kilo Code Subagents

What to do instead

• Code for implementation, Plan for architecture, Debug for troubleshooting. Each launches subagents when a subtask benefits from its own context.
• Stop switching to Orchestrator before complex tasks; the agent you're in already has the capability.
• For specialised delegation, define custom subagents (see Kilo's Custom Subagents docs) or custom agents with their own prompt, model and permissions.

How subagents work

1. The agent decides a subtask would benefit from isolation, such as exploring a codebase, a parallel search, or a side task.
2. It starts a subagent with the task tool, for example general for autonomous work or explore for research.
3. The subagent has its own conversation history but works in the same project directory or worktree, so it doesn't isolate file edits.
4. Foreground tasks return before the parent continues. Background tasks let the parent carry on and deliver their results later. Several can run at once.

Kilo Swarm (on by default) lets a session and its subagents share findings on a common board. It works with any agent and doesn't need Orchestrator mode.

Controlling cost and model choice

• subagent_model in kilo.jsonc (or Settings → Models) sets a default model for subagents. If unset, they inherit the parent's model, so an expensive main model means expensive subagents.
• For real isolation of file changes, run parallel work in separate git worktrees (Agent Manager in VS Code, the Agents view in JetBrains) rather than relying on subagents.
• Auto Efficient is another way to keep routine subtasks on cheaper models.