Documentation Index
Fetch the complete documentation index at: https://kawax.biz/llms.txt
Use this file to discover all available pages before exploring further.
Fleet Mode
Fleet Mode is an experimental orchestration feature that runs multiple sub-agents in parallel from a single parent session. When work can be split into independent units, the parent session assigns scopes and aggregates the results of each sub-agent. In Laravel, you can callsession.fleet.start from the typed RPC layer.
Good use cases
- Refactoring that can be split per file, package, or language SDK
- Investigation tasks that examine multiple modules separately
- Reviews scoped per diff, alert, or documentation page
- Migration-style tasks where each scope can be verified independently
Start Fleet Mode
prompt is optional. If omitted, it starts with only the runtime-side Fleet Mode instructions.
Start from Plan Mode
If you handle the Plan Mode UI yourself, you can advance from an approved plan into Fleet Mode by selectingautopilot_fleet in onExitPlanModeRequest.
autopilot for a single autonomous worker, interactive for an operation where the user stays continuously involved, and autopilot_fleet for running independent work units in parallel.
Sub-agent coordination
Fleet Mode coordinates through explicit work units rather than implicit shared memory. The parent session splits the task, each sub-agent processes only its own scope, and returns a completed or blocked state. When designing work units, make the following clear:- The scope a single worker is responsible for
- Which files or directories that worker may edit
- The deliverable or verification result to return on completion
- Dependencies on other tasks
Notes
- Fleet RPC is an experimental feature of the generated RPC layer. When using it, match the versions of the Copilot CLI and the SDK
- Avoid conflict-prone file edits and clearly separate scopes
- Assuming the parent session aggregates the results, including each worker’s report format in the prompt makes it easier to handle
Related docs
For the latest updates, see the GitHub repository.