Skip to main content
When Kato cannot finish a task, inspect what already happened before retrying. A failed reply or unfinished setup can leave useful resources and submitted work behind. Resume from the recorded state to avoid repeating changes unintentionally. Replace bracketed fields in example requests with your resource names or identifiers before sending.

​
Establish what happened

Return to the Kato conversation and identify the last intended operation. Ask:
Inspect the [workspace name] setup we were working on. List the resources created or reused, their identifiers, any work already submitted and the operations still incomplete. Do not create, start, retry or delete anything.
Open the returned resources and check their actual configuration and work records. Use Find resources and track work with Kato. If a response was interrupted, do not assume that the operation was cancelled or that its effects were reversed.

​
Resolve the specific dependency

Kato cannot grant itself missing authority or manage passwords, sessions, MFA or API keys. How Kato uses context and permissions explains the boundary.

​
Request only the remaining work

Restate the goal, the existing resource identities and the unfinished operation. Keep successful changes in place unless you intend to alter them. For example:
Reuse the existing [workspace name] workspace and [collection name] collection with the identifiers above. Attach only [dataset name or identifier] that was missing. Keep the scope and any exact period we agreed for my data. Do not recreate resources or start collection or processing.
For a retry, specify the affected job/run or binding and permitted scope. Inspect what may repeat, including model calls, collection or external effects. A retry is not a guarantee that every earlier action is skipped. Request cleanup as its own operation; Kato has no general promise of automatic rollback.

​
Verify the recovered state

Open the target and check that the remaining change completed, the intended identities were reused and existing settings still match the goal. Inspect any newly submitted work separately. Record remaining gaps rather than treating partial or empty input as evidence that nothing happened. If work remains blocked, retain the identifiers and diagnostic details for the appropriate resource Owner or administrator. Continue with workspace preparation, ingestion setup or workflow testing from the verified state.