Skip to main content
First identify the problem: an unsupported answer, a failed generation, a permission block, or an uncertain action result. Repeating the same prompt is not always the right recovery, especially if it could create an issue, comment, or other shared record.

​
The answer is wrong or lacks evidence

Check the sources cited in the answer and whether they apply to the current question. If a detail is unsupported, verify it manually or ask a clarifying question rather than sending it as fact. Ask a workspace admin to check Evelin AI → Knowledge sources:
  • The Help Center source must be enabled, and the relevant article must be published. An unpublished draft or pending review proposal is not published knowledge.
  • A website source must have usable, prepared pages. Check its status and refresh state; simply entering a URL or sitemap does not make content ready.
  • Update contradictory or outdated guidance at its source. Avoid adding duplicate copies of the same Help Center content through a website source.
Style guidance and example replies change wording; they are not factual evidence. Use Playground to check a representative product question after correcting sources. A successful Playground answer does not verify every live conversation or integration action.

​
Generation fails or stops

Read the visible error and distinguish a network problem from a session-expired message. If chat history failed to load, use Retry loading chat. After reconnecting, inspect the existing turn before sending it again. Check any displayed email-verification or AI-credit warning. Ask an admin to inspect the relevant workspace AI settings and Usage when access or balance is the issue. Do not assume all failed requests are free or that a missing answer proves no action occurred. If the problem persists, use a focused question with the necessary context. Preserve unsent work before refreshing or signing out. For a customer reply that cannot wait, write and verify it manually.

​
Linear is connected but an action is blocked

Ask a workspace admin to open Evelin AI → Integrations → Manage and check:
  1. Whether Evelin access is enabled.
  2. Whether the specific requested permission is enabled and saved.
  3. Whether the issue belongs to a team authorized in Linear.
  4. Whether the connection reports that authorization must be renewed.
Use Reconnect Linear when authorization has expired, been revoked, or the integration explicitly requires reconnection. A paused permission is not necessarily an authorization failure; reconnecting alone does not correct every denied action. If you cannot manage the integration, ask a workspace admin. Specify the issue key and intended action without sharing tokens or API keys.

​
A change may have completed

Check the target record, action receipt, and any available activity before retrying. Treat an unknown outcome differently from an explicit “nothing changed” result. Do not repeatedly ask Evelin to create an issue or add a comment while the result is uncertain. For stale proposals, review the latest record and request an updated proposal. For an unintended completed change, use Undo only where it is actually offered, or the target system’s supported correction flow. Stopping the answer does not reverse an already committed action.

​
Report a reproducible problem

Share the time and timezone, the operation, visible error, expected result, and any run/reference ID the product displays. Include the issue key or a redacted screenshot when relevant and permitted. Do not send credentials, private Evelin history, customer message bodies, or full network logs unnecessarily. Separate what you observed from what you inferred: “the request failed and the issue exists” is more useful than assuming that nothing happened. Check the final composer before sending any text recovered from an earlier result.