# Troubleshooting > Identify whether a failure belongs to inputs, execution, provider access, files, saving, or an external client before retrying. Source: https://kaitoi.io/docs/reference/troubleshooting/ Section: Account and help Written for: Anyone diagnosing a failed or unexpected workflow. Last reviewed: 2026-10-24 --- Start with the first useful error and the operation that produced it. Inspect existing results and save state before changing several settings or starting another run. ## Build a useful report from the first failure Record the first failing action, exact error, selected workspace and relevant project or run identity. Inspect saved or retained work before retrying. Preserve unsaved browser work before resetting settings. Use Report a problem with the appropriate category, review the submitted details and keep the returned reference. Remove keys, sensitive inputs and unrelated private information. The fictional retrieval error keeps original run r1 available for inspection. A separate saving branch preserves project A’s unsaved edits. The report artifact removes credentials while keeping the expected behavior, actual error and relevant identity. *Figure: Preserve the existing work and collect specific, cleaned evidence.* ## A feature or control is missing Check the selected **Personal** or team workspace, the restriction shown beside the control, and your team role or seat. Then check the required provider, connector or local service. Adding credits does not supply a missing feature capability. Use **Settings → Account → Profile** and **Plans**, when offered, for account context. See [Plans and availability](/docs/reference/plans/). ## Nothing runs or no result appears Check the selection before **Run Graph**, required inputs, compatible connections and execution state. Adding a node does not run it. A completed node can have an optional output that was not produced; check the selected output rather than assuming every preview is current. Read the node or run error and inspect any retained result before retrying. Keep the run ID and output/file identity. A thumbnail or temporary signed URL does not establish that the full file is still available. Use supported Station **Resume** or API idempotency controls for their intended recovery cases. They do not provide universal rollback or free retries. ## A provider or paid request is refused Check **Settings → Account → Providers** for managed access, **Variables** for eligible personal keys, and **Billing** for available, held and owed amounts or the remaining team allowance. A missing managed price, provider content refusal and timeout require different repairs. Repeated calls cannot create a missing price. A managed-provider `SERVICE_UNAVAILABLE` funding refusal can include an opaque reference; keep that reference and investigate the service rather than assuming customer debt. For local work, test the service on its own machine, then its connection to Studio, then the required model or recipe. For connectors, check the connected account, service scopes and resource permission. ## A file, save or recovery action fails Reopen an authorized file view to request a fresh expired link, and check whether the underlying file remains available. For offline or failed saving, preserve the tab and browser data while inspecting save status and backups. Read [Saving and recovery](/docs/studio/saving-and-recovery/) before resetting or clearing storage. For API guarded saves, follow the exact [error recovery path](/docs/core/errors-and-retries/#recover-a-guarded-project-save). **Reset All Settings** is not routine first aid: its confirmation describes discarding unsaved drafts, closing projects, deleting chat history and clearing local settings. Saved projects and backups are preserved, but unsynced work may not be. An incomplete reset can already have partly cleared state and may wait on original work. Do not repeatedly reset to get past that condition. ## An external client or integration fails Check the actual environment endpoint, credential type, scopes, origin policy and account capability. Server REST keys are not browser-origin keys. An MCP server saved in settings may still expose no tools to the current conversation. For a live Studio attachment, check Agent access, the intended session and the current revision. Do not switch silently to another graph after an attachment failure. Use [MCP troubleshooting](/docs/core/mcp-troubleshooting/) or [CLI troubleshooting](/docs/cli/troubleshooting/) for the relevant client. ## Report a useful reproduction 1. Open **Help & feedback**, then **Report a problem**. **Open Penny** is also available for help with the app. 2. Choose **Bug**, **Feedback** or **Idea**. Describe what you expected, what happened, and the smallest steps that reproduce it. 3. Keep relevant project, Station version, run and transaction IDs. Remove credentials, private inputs and signed URLs from your own text. 4. Review the disclosure and submit deliberately. Keep the returned **Reference** for follow-up. The form attaches bounded app, page, project, error and session context; the service also associates your authenticated identity and browser information. It does not automatically attach your graph or files, but error text or your description can still contain private material. Inspect what you include. A saved feedback receipt establishes that the report was recorded, not that a notification email reached support. Station visitors can use **Error details → Copy for Station author** to prepare its bounded diagnostic; copying does not send it.