Grafana Assistant Workspace
Grafana Assistant Workspace is a full-page Assistant experience. It keeps your conversation list, active chat, and canvas visible so you can review history, add context, run investigations, and work with results across multiple conversations in parallel, all in one place.
What you’ll achieve
With Workspace, you can manage multiple Assistant conversations from a dedicated page instead of the narrow sidebar. You can open recent chats, add context from Grafana resources, review canvas output, and work with investigation reports without leaving the conversation.
Before you begin
Before you use Workspace, make sure Grafana Assistant is enabled for your deployment and Workspace is available for your stack.
For investigation workflows, you also need access to Investigations. To create a dashboard from canvas output, you need permission to create or write dashboards.
Open Workspace
Workspace is designed for longer conversations and side-by-side analysis. Use it when you want more room than the core Assistant sidebar provides. Workspace is a regular Grafana page, so the Grafana navigation menu and top bar stay available while you work.
To open Workspace, use one of these options:
- Click Workspace in the Grafana top bar. If a conversation is open in the Assistant sidebar, Workspace continues it.
- In the navigation menu, select AI > Assistant Workspace.
- Search for Assistant Workspace in Grafana search.
- To open an existing conversation, open the conversation menu in the Assistant sidebar and select Open in Workspace.
When Workspace is open, the Assistant sidebar closes, the conversation moves into the full-page layout, and the navigation menu collapses to give Workspace more room. To open the navigation menu again, click the menu icon in the top bar. The top bar shows Exit Workspace instead of Chat and Workspace.
Leave Workspace
Click Exit Workspace in the top bar to return to the page you were on before you opened Workspace. If the Assistant sidebar or the navigation menu was open there, it opens again. If you opened Workspace directly, for example, from a link, Exit Workspace takes you to the Grafana home page. You can also leave Workspace like any other page, from the navigation menu or search.
When the Assistant opens a Grafana page for you, for example, a dashboard it created, you leave Workspace for that page and the conversation continues in the Assistant sidebar.
If you open an Assistant link with a prefilled prompt in your browser, review the draft and send it when you’re ready. Opening the link doesn’t send the prompt automatically.
Navigate the layout
Workspace uses a three-column layout so you can keep navigation, chat, and visual output in view at the same time.
- The left column shows your conversation list. Your open conversations come first, and Working and Settled sit below them. Search keeps the filter you picked. Without a filter, search finds conversations of every type, archived ones included. To show only conversations, all investigations, your own investigations, the investigations of your teams, or archived conversations, click the filter button next to the search field. Your teams includes investigations that alerts start for your teams. Without a filter, the list also shows your teams’ investigations that need someone: they run, or they had activity in the last 14 days and wait for a reply, failed, have a fix or pull request waiting, or investigate an alert or incident that isn’t resolved. The button next to it starts a new conversation. You can pin, rename, settle, or archive a conversation from its row, and archive all unpinned conversations from the filter menu. You can toggle this column’s visibility with the Open/Close sidebar toggle at the top of the column, or by pressing option/alt+s.
- The center column shows the active chat. This is where starter suggestions, follow-up questions, queued prompts, and shared conversation controls appear.
- The right column is the canvas. Workspace opens it when the conversation produces panels, diagrams, or investigation output, and you can collapse or reopen it as needed. You can toggle this column’s visibility with the Open/Close Canvas toggle at the top of the column, or by pressing option/alt+c.
The New action follows the conversation-list filter. Filter the list to investigations to start a new investigation, or to chats to start a new chat, regardless of the conversation currently open.
The chat header also lets you share or download a conversation. Shared links stay inside your Grafana organization and represent a snapshot of the conversation. If the conversation changes later, update the shared version to include the latest messages and its current title.
Use Download conversation to save the current conversation as a JSON file. The file uses the grafana-assistant-chat format and includes visible user and assistant messages plus limited conversation metadata, such as the conversation ID, name, modified time, and export time. Hidden, system, internal, and transient streaming state are excluded. If a conversation is still updating, the download captures the messages that have loaded when you click Download conversation.
Hand off a conversation
To continue your work in another AI assistant, open the conversation menu and select Hand off conversation. Choose a handoff method, then copy the generated prompt into the other assistant:
- gcx / coding agent: Use an assistant that can run
gcx. The prompt includes commands to retrieve the transcript and send follow-up messages to the original Grafana Assistant conversation. - Grafana Cloud MCP: Use an assistant connected to the Grafana Cloud MCP server. The prompt asks it to retrieve the transcript with
get_assistant_conversationand use that context when callingask_assistant. The generated prompt doesn’t explicitly resume the original conversation.
Handoff is available for existing conversations outside trial mode and embedded Workspace.
Keep your conversation list focused
Workspace lists your open conversations first, with the newest activity at the top, so a conversation comes back to the top when its work does. Views without sections, such as Conversations, Investigations, and search results, list running conversations of any date under Working, above the date groups Today, Previous 7 days, Previous 30 days, and Older. Finished work moves to its own section, so the list stays short while Slack conversations, Automations, and alert investigations keep arriving.
- Your open conversations are at the top, without a heading. The ones that need you keep full contrast: a conversation that asks you a question, one that waits for your approval, and one that changed after you last opened it. The others are dimmed until you point at them.
- Working lists the conversations where Assistant is still running, below your open conversations. This includes conversations whose reply is done while an agent it started, such as a coding agent, still works. The section is collapsed by default, and the conversation you have open stays visible under its header. To keep running conversations at the top of the list instead, click the filter button and clear Working section. Workspace remembers this choice in your browser.
- Settled holds finished work, with the newest activity first. The section is collapsed by default.
An icon at the end of a row shows its state, for example, a question mark when Assistant asked you a question, a shield when it waits for your approval, or an alert sign when an investigation failed. Point at the title to read the state in words and when the last activity was. A running conversation shows a spinner and how long it has run, and other conversations show the time of their last activity. An unread dot marks a conversation that changed after you last opened it. A conversation whose reply waits for your browser to run a step shows a dashed circle, and its title tooltip says what it waits for. When it’s your own conversation, the row stays at full contrast, because opening it lets the step run. In views without sections, a check mark shows the conversations you settled.
Settle conversations
Point at a row and select Settle to move a conversation to Settled, or select Settle conversation in the menu of the conversation you have open. You can do this in every view except Archived, and a notice with Undo appears. Settle several conversations one after another and the notice counts them, so one Undo brings them all back. You can also press Cmd+Z on macOS or Ctrl+Z on Windows and Linux while the notice shows, unless you’re typing in a text field.
Each row also has a menu with the rest of its actions: select More actions or right-click the row to pin, settle, rename, or archive it. To rename a conversation, you can also press F2 while it has focus. Press E on a conversation that has focus to settle it, or to reopen a settled one. After a settle, focus moves to the next conversation, so you can work down the list.
To act on several conversations at once, hold Cmd on macOS or Ctrl on Windows and Linux and click them, or Shift-click to select every conversation between two. With the keyboard, press X on a conversation that has focus, or Cmd+A or Ctrl+A to select every conversation the list shows. Pinned conversations stay out of a selection. The bar at the bottom of the list says how many you selected and offers Settle and Archive for them. Either action moves the whole selection at once, with one Undo. Press Esc to clear the selection.
Workspace also settles a conversation when one of these is true:
- It has no activity for 7 days, or for 14 days when it waits for your reply. A conversation where Assistant is still running doesn’t settle for inactivity.
- It’s an investigation that completed, and you opened it after it completed.
- Everything linked to it finished: no pull request is still open, every incident and alert group resolved, and at least one pull request merged or one incident or alert group resolved. A pull request closed without a merge doesn’t count as finished work on its own.
Otherwise, a conversation that’s pinned, waits for your reply, still runs an investigation, or has a fix waiting for your approval or for a pull request doesn’t settle. An investigation counts as running only while Assistant works on it, so one that’s paused or waits for your next message settles like any other conversation. A conversation comes back from Settled when someone sends a message in it, when Assistant asks you a question, or when an investigation fails. An Automation run that finds nothing doesn’t bring it back.
Settled is personal: settling a shared investigation doesn’t change your teammates’ lists. Select Reopen to bring a conversation back yourself. The Conversations, Investigations, and Your teams views and the Needs your attention section of the Assistant home page use the same rules.
Undo an archive
When you archive a conversation, it leaves the list right away, and a notice with Undo appears for five seconds. To archive every unpinned conversation in the current view, click the filter button and select Archive all…. It asks you to confirm and states how many conversations it archives. It keeps pinned conversations and conversations that other users own.
Search titles and messages
Search matches conversation titles and the text of your saved messages. Message matches appear under In messages with the text around the match. Select a match to open the conversation at that message.
Add context to a conversation
Workspace keeps the same resource-picking workflow as the core Assistant chat, but makes it easier to keep selected context visible while you work.
Type @ in the prompt field or select Add context to choose resources. From the Workspace picker, you can add data sources, labels, metrics, dashboards, entities, skills, and notebooks. After you add a notebook, you can ask the Assistant to summarize or edit it. Selected items appear as tags above the prompt so you can confirm what the Assistant will use.
Workspace also shows context window usage in the prompt area. This helps you see how much room remains in the current conversation context before you send another prompt.
In standard chat conversations, you can attach files and images to a prompt. Use the attachment control, paste files into the prompt, or drag files into chat. Assistant supports PNG, JPEG, GIF, WEBP, CSV, JSON, YAML, Markdown (.md or .markdown), and plain text (.txt) attachments. You can attach up to five files per message, with a maximum size of 10 MiB per image and 10 MiB per selection. Assistant resizes a larger PNG, JPEG or WEBP image, up to 50 MiB, to fit, so you don’t need to compress screenshots first. Text files must be 500 KiB or smaller unless the built-in Grafana Cloud coding sandbox is available for your chat, which permits text files up to 5 MiB each. If a file exceeds the available limit, attach a smaller excerpt or split it into smaller files. In investigation mode, you can paste or attach images to add visual context to your prompt.
Continue with a suggested prompt
After the Assistant finishes a reply, it can suggest a follow-up in the empty prompt field. Press Right Arrow to fill the field with the suggestion, then edit or send it when you’re ready. Accepting a suggestion doesn’t send a message. You can also ignore it and type your own question.
Suggested prompts work in both Workspace and the Assistant sidebar.
Send follow-up messages
If you queue several follow-up messages while Assistant works, it sends each one as its own message, in the order you queued them. When Assistant finishes replying to one, it sends the next.
Sending a queued message immediately sends only that message. The other queued messages wait and follow it in order. In standard chats on stacks where chat tiers are available, sending a queued message immediately interrupts the current reply and waits for it to stop before starting the new reply. Clicking Stop leaves any remaining queued messages waiting for you to send them; they don’t start automatically.
Copy a response
Click Copy message on an Assistant response to reuse it elsewhere. When you paste into a plain-text editor, the response keeps its Markdown formatting, including lists and code blocks. Rich-text editors receive the rendered formatting. References to Grafana resources use their readable labels in plain text.
Get a reply in Slack
When a reply takes longer to finish, Workspace and the Assistant sidebar can offer to send it to Slack. Your stack must have Slack connected, and your Grafana account must be linked to your Slack account.
Click Notify me to receive the final reply and a conversation link in a direct message. Keep the browser tab open until the reply finishes. This choice applies only to the current reply and doesn’t send a notification while Assistant waits for your approval.
Slack preserves the reply’s Markdown formatting. If the reply is too long for Slack, the message contains a shortened version; click Open conversation to read the full reply.
On stacks where chat tiers are available, sending a priority follow-up keeps an enabled Slack notification for the replacement reply. Clicking Stop turns the notification off.
Click Cancel to turn off the notification. Hiding the notice doesn’t cancel the notification. If delivery fails, click Retry. In the conversation menu, use Show Slack notification prompt to turn these offers on or off. Turning on the prompt doesn’t enable a notification by itself.
Respond to a safety notice
If the model blocks a request through its safety guardrails, Workspace and the Assistant sidebar show a notice instead of ending the reply without an explanation. Start a new conversation to continue. If you think the request is blocked incorrectly, contact support.
Choose a chat tier
On stacks where chat tiers are available, use the tier picker in Workspace or the Assistant sidebar to choose the capability level for your next reply. Auto lets the Assistant’s selection policy choose a tier. Turn Auto off to select Small, Medium, or Large, from the lower-resource preset to the highest capability preset offered.
You can change the tier between replies. A tier doesn’t change your permissions or approval requirements, and a higher tier doesn’t guarantee a faster or more accurate answer. Token consumption depends on the work performed. For usage details, refer to Manage Grafana Assistant pricing, usage, and limits.
Chat tiers aren’t available in every region or conversation type. If your chat doesn’t offer the tier picker, Assistant uses the configuration available for your conversation.
Recover an interrupted reply
On stacks where chat tiers are available, Workspace and the Assistant sidebar retry once automatically if a reply stream closes unexpectedly before the Assistant starts responding. Cancellations and explicit errors aren’t retried automatically.
After a server restart, Assistant can also retry once if no tool work has started and no approved action is being resumed. If automatic recovery could repeat an action, or the retry fails, a message asks you to retry your request. Completed tool results remain in the conversation. Review those results before retrying; an interrupted or cancelled reply doesn’t undo completed actions.
Use the canvas
The canvas is the main feature that separates Workspace from the standard Assistant sidebar. Instead of opening visual output in a separate view, Workspace keeps it beside the conversation.
Panels and Mermaid diagrams appear in the canvas as the Assistant produces them. When the Assistant references a panel in chat, you can focus that item in the canvas directly from the conversation.
When available, panel and query output menus include Download CSV so you can export tabular data from Assistant results. Assistant chat can also show Download CSV for exportable tool output.
The canvas also collects the named outputs a conversation produces, such as reports, dashboards, and query panels, as tabs across the top. Select a tab to open an output, then use its menu to view or download it. When a conversation produces more outputs than fit, scroll the tab strip or use the canvas summary to open an output.
When Assistant creates a report artifact, the chat card includes Open report. Click it to open the report tab directly in Workspace. Report exports use the Workspace export menu, so you can download the report as Markdown, PDF, or another available format.
Investigations promote their output into dedicated tabs:
- Report shows the investigation findings and recommended next steps.
- Hypotheses appears when an investigation produces a hypothesis plan.
- Sources appears when the conversation produces panel-based query output.
During an investigation, the Assistant annotates its responses with inline source citations. Numbered chips, for example, Source 1, appear directly in the chat text. Click a chip to open the referenced panel in the Sources view, or hover to preview it.
If you have the required permissions, you can also build and edit Grafana dashboards in the canvas. For more information, refer to Build dashboards in Workspace.
Arrange canvas tabs
You can reorder, close, and reopen canvas tabs such as Report, Hypotheses, Sources, or an artifact tab. The canvas remembers your arrangement for each conversation, including in the Assistant sidebar.
To reorder tabs, drag a tab to a new position. When a tab has keyboard focus, you can also press Alt+Shift+Left arrow or Alt+Shift+Right arrow (⌥⇧← or ⌥⇧→ on macOS). After you move or close a tab, the canvas keeps that order and adds new tabs at the end.
Closing a tab only hides it. The output stays in the conversation, and a reopened tab returns to its previous position.
You can also close Subagents and Changes. These tabs keep their positions at the start of the strip.
To close a tab, hover over it or focus it with the keyboard to reveal the close icon, middle-click the tab, or press Delete while the tab has keyboard focus. Right-click a tab to close it, close every other tab, or close the tabs to its right. In Workspace, you can also press Alt+W (⌥W on macOS) to close the active tab.
To reopen a tab, select it in the summary card. When the summary card is hidden, click Show canvas summary beside the canvas icon and select the tab under Recently closed. If a closed tab has changes you haven’t seen, for example, because the Assistant updated it after you closed it, the tab stays closed and shows New changes in that list. Opening an output from the conversation, for example, clicking a source citation or Open report, also reopens its tab. In Workspace, press Alt+Shift+T (⌥⇧T on macOS) to reopen the most recently closed tab. When every tab is closed, click Reopen all tabs in the canvas.
Split the canvas
You can split the canvas into two panes to compare outputs, for example, a report beside its Sources. Each pane has its own tabs, and the pane you last worked in keeps its active tab highlighted.
To split the canvas, drag a tab onto an edge of the canvas, or right-click a tab and select Split right or Split down. You can also drag an item from the canvas summary onto the canvas: drop it on a pane to open it there, or on an edge to open it in a new pane. You need at least two tabs to split. Terminal, Subagents, and Changes stay in the main pane.
To move a tab to the other pane, right-click it and select the Move to option, or drag it onto the other pane. Drag the divider between the panes to resize them, and double-click it to make them equal. When a pane’s last tab closes or moves, the other pane fills the canvas again. The split lasts until you reload Workspace, and it isn’t available on narrow screens.
Run investigations in Workspace
Workspace is the main place to start and review Assistant investigations. For a fresh conversation, switch the mode to Investigation before you send the first prompt.
If you don’t see Investigation mode, ask an administrator to grant the Assistant Investigation User role or the grafana-assistant-app.investigations:create permission.
When an investigation is active, Workspace shows progress in chat and builds the report beside the conversation. Hypotheses can appear as a summary card in the conversation and as a dedicated view in the canvas. You can keep steering the investigation from the same conversation while the report is still being assembled.
Before you start an investigation, you can choose a private scope or add one or more Grafana teams. After an investigation starts, you can continue with follow-up questions, pause or resume it, and open the report in a full-page investigation view.
Build dashboards in Workspace
Workspace can create and edit Grafana dashboards in the canvas, so you can build a dashboard in the same place you talk to the Assistant.
Switch the mode to Dashboarding, then describe the dashboard you want. The Assistant adds panels, variables, and layout to a dashboard that renders live in the canvas. You can keep refining it through the conversation, for example, change a visualization, adjust a query, add a variable, or rearrange panels, and the canvas updates as you go. You can also ask the Assistant to load an existing Grafana dashboard into the canvas to edit it.
If you know the service but not the exact telemetry, Assistant can search existing dashboards and alert rules to find useful queries and thresholds. When Knowledge Graph memory is enabled, it can also search infrastructure metadata. Assistant verifies the results against your data sources before it builds panels.
To build dashboards, you need permission to create or write dashboards (dashboards:create or dashboards:write). If you don’t see Dashboarding mode, ask an administrator to enable it for your stack.
Save a dashboard to Grafana
The dashboard in the canvas is a working draft until you save it. When you’re ready, select Save to Grafana and complete the dialog:
- Title: The dashboard name. This field is required.
- Description: An optional summary shown in dashboard search results.
- Folder: The folder to save into. The default is the General (root) folder.
- Commit message: A note recorded in the dashboard’s version history.
Select Save to Grafana to create a new dashboard, or Save changes to update one you loaded or saved earlier. After the dashboard saves, select Open in Grafana to view it.
Understand how Workspace differs from the core Assistant
Workspace is a conversation surface, not a replacement for every Assistant page and control.
- Use the Assistant homepage and Settings for setup and configuration tasks such as rules, quickstarts, skills, integrations, tokens, and other tenant or user settings.
- Workspace focuses on standard chat, investigations, and dashboarding. It doesn’t expose every mode from the core Assistant surfaces, such as learning mode.
- Workspace doesn’t include the tutorial entry that is available from the core Assistant menu.


