# How Lykeion works

Lykeion connects a shared research workspace to computers that execute agent work. The hosted dashboard lets you describe work and inspect its history. Execution depends on a connected machine and its agent CLI, so a browser session alone does not mean research is running.

## Trace the handoff

1. A researcher creates a study with a question, inputs, assignee and scope.
2. Lykeion schedules execution for the selected agent or Team on an eligible runtime.
3. The machine's daemon carries out that work through its configured agent CLI.
4. Runs write progress, hand-ins and other outputs back to the study.
5. Required stage work and review determine whether the study can continue or finish.

Use [runs](/docs/runs/) to distinguish queued work, active execution and an ended attempt. Keep the selected machine online and its CLI signed in while work needs it.

## Read state at the right level

A finished run means that execution attempt ended. It does not by itself mean that the whole study is Done. A study may have more configured stages or required review outstanding. Read its stage, activity and Review surface together before treating the result as complete.

The demonstration study used for these guides produced real Grounding hand-ins but waited for scoring. Its screenshots show that intermediate state, rather than a successful end-to-end validation.

## Verify one handoff yourself

Open a study, find its latest agent card and identify the reported outcome. Check Files or Notebook where relevant, then inspect Review. You should be able to point to an actual record or artifact that supports what the agent says it delivered.

If no run starts, check [machine connection](/docs/connect-machine/) before changing the research question. If a run ended with an error, read the error and address its cause before retrying. [Evidence and results](/docs/evidence-and-results/) explains the inspection surfaces.
