Mission Modes

Every mission runs in one of three modes that control how much autonomy Kavi has during planning, execution, and follow-up handling. You choose the mode when creating a mission, and it determines the balance between speed and operator oversight.

guided_autopilot

Kavi plans and executes autonomously. Follow-up recommendations generated by agents are applied automatically when there are no operator review blockers. The operator is notified of progress through checkpoints but does not need to intervene at each step.

Best for:

• Well-defined tasks with clear scope
• Work where you trust the agents to proceed without review at each step
• Iterating quickly on implementation tasks

In this mode, Kavi will:

1. Decompose the prompt into an execution plan via a planning task
2. Route and execute each task in the plan according to dependencies
3. Automatically apply follow-up recommendations from agents
4. Run acceptance verification when all tasks complete
5. Transition the mission to ready_to_land on success

inspect

Kavi plans and executes tasks, but pauses at key decision points for operator review. Follow-up recommendations require manual application rather than being auto-applied. This gives you visibility into what the agents are doing while still offloading the planning and routing work.

Best for:

• Complex tasks where you want to verify agent decisions
• Work spanning multiple agents where coordination matters
• Tasks where you want to review intermediate results before proceeding

In this mode, Kavi will:

1. Decompose the prompt into an execution plan
2. Route and execute tasks, but pause when follow-ups are generated
3. Surface recommendations for operator review before applying them
4. Run acceptance verification when all tasks complete

manual

Full operator control. Tasks must be individually approved and routed. Kavi does not auto-apply any follow-up recommendations, and the operator drives each step of the workflow.

Best for:

• Sensitive changes (security, infrastructure, data migrations)
• Exploratory work where the path forward is unclear
• Learning how Kavi works by observing each step

In this mode:

1. The operator approves and routes each task
2. All recommendations require manual review and application
3. The operator controls the pace and direction of execution

Mode Comparison

Choosing a Mode

As a general rule:

• Start with inspect if you are new to Kavi or working on a project for the first time. It lets you observe how Kavi decomposes and routes work while retaining control over follow-ups.
• Move to guided_autopilot once you are confident in your routing rules and agent capabilities. This is the fastest path from prompt to landed code.
• Use manual when the stakes are high or the problem space is ambiguous. The extra control is worth the slower pace.

The mode applies to the entire mission. If you need different levels of control for different parts of the work, consider splitting the work into separate missions with different modes.