Choose what work to send to Compute

Copy one setup handoff

Copy and paste one setup handoff into the frontier agent you already use. It asks that agent to inspect only workflow information, files, and connected tools you already authorized, so it can help you find safe, repeatable work without creating a new assistant or deciding for you. The prompt text itself is on Copy a Compute setup prompt — this page is about what to do with it once you have it.

Describe one bounded outcome

Ask the frontier agent to name one result that can be checked on its own. A useful candidate states its allowed inputs, forbidden data, output shape, and validation check.

Keep the first candidate narrow. A task that needs your full context or final judgment belongs with your frontier agent, not with Compute — start with Lithi Compute once you have one.

Check the boundary before you approve

The frontier agent reviews the context you authorized, then asks only unresolved questions. It can estimate suitable tasks, a savings range, and the basis for that estimate.

Confirm the input, expected result, check, current capability, quote, and savings basis before approving anything. Do not share credentials or private records to fill a gap. When the boundary is clear, run your first small Compute test rather than a full workload.

Keep the final decision

Lithi handles bounded work while your frontier agent keeps the plan and the final answer. Your frontier agent can keep a task with itself, use Lithi for a candidate, or mark the work unsafe or unsupported — that choice, and every approval after it, stays yours.

This process protects your control at every stage. For the broader data and security scope, read Security.