Navigation
3.10. Find Relevant Tickets and Projects with Smart Search
Search filtered tickets and projects by meaning, review Strong and Possible matches, and cancel or rerun searches as your investigation changes.
An MSP often remembers the problem rather than the words used in its ticket title. Smart search scores tickets or projects against a description of the work, helping technicians find related incidents and previous delivery work. For Northstar Accounting, a query such as “intermittent access to the shared finance folder” can help investigate records with different wording.
Before you begin
Smart search uses the Enterprise implementation and requires an active AI Assistant add-on. Your account needs read permission for the entity being searched: ticket for tickets or project for projects. The server also needs a configured TYPESAFE_API_KEY; this is a server administrator prerequisite, not a key to paste into the search field.
Availability is checked before the list offers smart search. If the button is missing, ask your administrator to check the add-on, permissions, and server configuration. Normal list search remains a separate way to find records.
Search the right candidate set
- Open the Tickets list or Projects list.
- Set the list's filters before searching. For example, narrow the list to Northstar Accounting and the relevant status or project scope. These filter chips define the candidates that smart search will score.
- Enter a description in the search field. The ticket field prompts Search tickets and comments… Enter for smart search; the project field includes projects, tasks, and comments.
- Press Enter or select Smart search. The button requires a nonempty query.
- Watch the progress strip and review matches as they arrive. Open a record using its normal row action to assess whether it answers your question.
The query is used for relevance scoring rather than requiring every candidate to contain those exact words. Tickets supply information including their title, description, and comments; projects supply their name, description, tasks, and comments. Existing access checks still apply to the candidate records.
This makes scope important. A closed incident excluded by your status filter cannot become a result merely because it would be highly relevant. When investigating a recurring issue, consider whether your filters should include closed work before running the search.
Read the results
| Section | How to use it |
|---|---|
| Strong matches | Start here for the records scored as most relevant to your query. |
| Possible matches | Review these for related symptoms or partially matching work. |
| Unlikely matches | Expand this initially collapsed section when you need to inspect lower-scoring candidates. |
| Could not be scored | Inspect records from failed scoring batches separately; failure is not a relevance judgment. |
The Match column displays a relevance percentage. Treat that score as a guide to which record to read, not confirmation that two incidents have the same cause. Rows arrive within their buckets as processing continues; the smart-search tables do not offer ordinary column sorting during this view.
Cancel, change filters, or return to the list
- Select Cancel while scoring is running to stop the search. The progress strip reports how far it got, and already received results remain available.
- After the run settles, select Back to list to leave the results. Clearing the query also exits smart search; Escape in the search field clears it while smart search is active.
- If you change filters after a run, select Filters changed. Run again to score the new scope. Changing the filters does not silently replace the candidate set of the existing run.
- If the strip says Smart search stopped, inspect the message and use Retry after addressing the cause.
The expected result is a set of relevance buckets for the query and filter scope captured when you started that run. If results are unexpectedly sparse, review the filters first, then broaden or clarify the query. A server-configuration or add-on error needs administrator attention rather than repeated query changes.