Keep Workspace assistant answers direct
Completed Workspace assistant answers no longer end with an unnecessary offer or follow-up question.
Completed Workspace assistant answers no longer end with an unnecessary offer or follow-up question.
Workspace assistant answers no longer repeat the same links, exact total, and table breakdown.
The Workspace assistant benchmark now includes the original question about items across both Idea Workspaces.
Detailed searches and summaries now keep their filters, exact totals, scope, and bounded previews as saved results.
Workspace assistant model and reasoning changes can now be compared against the same tool-selection questions.
Workspace assistant deployments can now set a supported OpenAI reasoning level without changing code.
Exact product-area counts now remain visible as saved results in the conversation.
Workspace assistant record searches now keep their exact totals and bounded previews as saved results.
The Workspace assistant can use a saved canvas selection for later item searches and counts.
Workspace assistant searches now send only supported filters and use the correct invoice status fields.
The Workspace assistant can search canvas items and calculate exact totals across selected Idea Workspaces.
Workspace Assistant tool failures now return safe recovery messages while retaining diagnostic details in protected logs.
Workspace Assistant Reader feed results now link directly to the matching subscription.
Workspace Assistant feature results now use member-area links that open the relevant Magpie page.
Links returned for Notes now open the selected Note in the Notes editor.
The administrator Assistant Reference now lists the returned fields, limits, and link behavior for each feature read.
The workspace assistant can now retrieve bounded Markdown content from an accessible Note.
Store catalog metadata is now available to the workspace assistant.
The workspace assistant can now find useful records across the rest of your Magpie workspace.
Every assistant question was failing straight away. The tool list it sends with each question contained a malformed entry, which caused the whole request to be rejected before the question was read.