Two lanes, and why a run waits
Simple tasks are answered directly, with no sandbox. Work that needs code, a shell or files gets a sandbox of its own. A run that waits always says why.
Updated 2 min read
Answered directly
Most messages are answered directly: conversation, planning, and the tools we run for the agent, such as your connections, Rykon Mail, knowledge, approvals and your trusted browser once you have granted it. No sandbox is made, so the first reply comes quickly. The run says Answered directly, and its details show how long the first reply took.
Working in a sandbox
When a task needs code, a shell, files or packages, the agent asks for a sandbox and says why, for example "run the tests". The request is checked against the agent’s rules and kept on the record. The sandbox work is a run of its own: the conversation shows Working in a sandbox to and the reason, with a link to that run, and the sandbox run links back. Its details say how it was isolated.
Strict isolation
If you would rather every run, plain replies included, worked in a sandbox, an owner or admin can choose Strict under Settings > Run isolation, for the whole account or for single roles. Replies then take a few seconds longer and use more compute, because every run starts a sandbox; the run’s details say Strict isolation. Strict is available on paid plans, and can be part of an enterprise contract, in which case Settings shows it as part of your contract and it cannot be turned off there.
Why a run is waiting
A run that has not started says why, in plain words, and what to do if anything. It also shows who has to act (an owner or admin of your account, or us), its place among your account’s waiting runs, how long it has waited and, only when we can tell, when it is expected to start.
| What the run says | What to do |
|---|---|
| Getting ready, or preparing the sandbox | Nothing. It starts in a moment; the first run on a new setup takes a little longer. |
| Your runner host is offline | Start the runner host again, or check its network. The run starts when it connects. |
| No place to run sandboxes yet | An owner or admin sets up a runner host for the account. See Where runs execute. |
| Running as many tasks as the account allows | Nothing, or stop a run you no longer need. It starts when one ends. |
| Every sandbox is in use | Nothing. It starts as soon as one is free. |
| Cannot reach an AI model | Check Spend: the budget or today’s limit may have been reached. |
| The sandbox could not be prepared | Nothing on your side; we have to fix it. Write to support@rykon.ai with the run’s id if it lasts. |