Multi-Agent
A bigger task needs more than a longer prompt.
Give a connected goal to a coordinated workflow. Specialised roles investigate the project, prepare the approach, implement subtasks and check the result – without making you manage every handoff.
Multi-agent orchestration requires Ultra and available credit. Tool approvals and project quality checks still apply.
One goal. Separate responsibilities.
- Plan
Turn project context into concrete subtasks.
- Implement
Work through the agreed changes with the project tools.
- Verify
Review the result and run relevant checks.
How orchestration works
Specialist agents work on one shared task.
- 01
Break the goal into subtasks
The Planner investigates the goal and project context, identifies affected areas and creates concrete subtasks.
- 02
Implement the changes
The Implementer works through subtasks with the project tools. Results are collected within the same run.
- 03
Review independently
The Reviewer checks changes and relevant verification results. Open points return to implementation as follow-up tasks.
- 04
Bring the result together
Receive the changes, verification results and open points for the shared task. You decide what to accept.
Autonomy with boundaries
Keep the scope visible. Keep a cost limit.
-
Set the budget.
The default run budget is €20. Use --max-cost to set a different total limit before starting.
-
Choose the mode explicitly.
--mode plan prepares an artifact; --mode agent implements. --mode auto follows the current CLI mode.
-
Check the outcome.
A completed run is not a correctness guarantee. Keep tests, human review and your release controls in place.
Start with a bounded task.
Run these slash commands inside a Lurus Code chat in your project. Choose a clear goal and review the plan before handing over a larger change.
Read the documentation- Prepare a plan without source edits
/orchestrate --mode plan "Add a settings page" - Implement with a €20 run budget
/orchestrate --mode agent --max-cost 20 "Add a settings page"
Your code stays yours.
Source code and request context are processed with Zero Data Retention. No use for model training. Providers and processing regions are disclosed transparently.
See privacy and processing detailsA few things to know before you start.
Which plan includes the orchestrator?
Does Plan mode implement changes?
Is a completed run ready to merge automatically?
Multi-Agent
Give the next larger task a connected workflow.
Choose Ultra for multi-agent orchestration. Start with a well-defined task rather than handing over the entire project.
Multi-agent orchestration requires Ultra and available credit. Tool approvals and project quality checks still apply.
Want to start with your team? Talk to us.