Chat Assistant

Ru is the Runnit Chat Assistant. Ru can answer questions, inspect work you can access, find assets, update authorised project data, and generate or edit media. Open Ru from the Chat icon in the global header. On a desktop, Ru opens in a floating window that stays in front of your current page, and you can also dock it to the right edge as a full-height side panel.

Chat from the current page

The side panel starts with the page you opened it from. Each page and record has its own conversation, so project A, project B, My Tasks, and the dashboard do not share one drifting thread. Returning to a page restores its previous Ru conversation.

Each message carries what is currently on screen. When the page shows a specific brief, project, client, or task, Ru reads that item within your access before answering. You can ask “what is the budget in this brief?” or “what is at risk?” without restating where you are.

On My Tasks, Ru also receives the selected filter and the exact task order shown on the page. “What is the top task?” means the first card you can see, and “mark the top task complete” targets that card’s exact task record rather than a different result from a general task search. After Ru changes a task, the page refreshes and the visible order is updated for the next message.

For questions about Runnit itself, Ru treats the official Runnit User Guide as the product authority. Where a guide page matches the screen you opened Ru from, Ru can read that exact public article before answering instead of assuming that a setting or feature is unavailable. For example, from Admin > Agent Integrations, Ru knows that administrators can register Remote MCP servers and fixed HTTP APIs as managed tools.

When you ask where a project is up to, what its current status is, what has happened on it, or what someone is waiting for, Ru reads the discussion as well as the records. That means the project’s own chat, where the team writes notes such as “waiting on the client to come back on the latest concepts”, together with the comments on the project’s tasks and the project’s activity feed. Ru quotes the relevant messages and says who wrote them and when, so a status note that only exists in project chat is not missed.

The assistant also resolves vague references itself instead of asking you to be more specific. Asking for “the latest Aurora project” looks up your projects and picks the most recent match. When several items genuinely match, it asks one question that lists the real candidates so you can reply with a name or number.

When a creative request is missing details such as the objective, audience, or style, the assistant asks once and proposes a default for each gap. Telling it to go ahead (for example “just do it” or “use your defaults”), or confirming the defaults it proposed, is final: it states the assumptions it is using and does the work without asking again.

Use the controls at the top to:

start a new conversation;
expand the current conversation into the full chat workspace; or
close the panel and return to it later.

You don’t need to keep the panel open while Ru works. If Ru finishes by asking you something (for example, to pick between numbered options or confirm a detail) and the conversation isn’t on screen, Runnit sends you a Ru needs your input notification and shows a short on-screen alert quoting the question. Choose Open chat on the alert, or select the notification in the bell inbox, and Runnit returns you to the page the conversation belongs to and reopens it so you can answer. The notification clears once you reply, and the link still works after signing back in or on another device.

If you start a new conversation and open the full chat before sending a message, the full workspace keeps the page, selected item, and query selection where you opened Ru. Your first message creates the conversation against that saved context. Moving to another page does not move or replace the source page’s conversation.

Drag the Ru title area to move the window. Drag the three-line grip in its bottom-right corner to change its width and height. Runnit saves that position and size in your browser. You can also focus either handle and use the arrow keys. Press Home on a handle to return the window to its starting position and size. On a smaller screen, the window uses the full screen and the move and resize handles are hidden.

Select the panel icon in the Ru header to dock the window to the right edge as a full-height side panel. The page shifts left to make room, so nothing is covered while you chat. Drag the panel’s left edge to change its width, or focus the edge and use the arrow keys (Home restores the starting width). Select the same header control again to float the window; it returns to the exact position and size it had before you docked it. Runnit remembers your docked or floating choice, and both sizes, in your browser. It also remembers whether the chat was open: reloading the page brings Ru back exactly as you left it, docked, floating, or full screen on a phone, and closed stays closed.

On a desktop, the page behind Ru stays interactive during normal conversation. You can keep navigating, typing, and updating your work while Ru answers a question or reads information. The chat window does not dim or block the rest of Runnit for those requests.

When Ru starts changing workspace data, Runnit briefly pauses interaction with the page behind the window to avoid conflicting edits. A notice explains that Ru is updating the page. The pause ends as soon as the write finishes, while Ru reviews the result and writes its answer.

Select Interrupt Ru in the progress card or the page-update notice whenever you need control back. This releases the page immediately and stops the current request without deleting or ending the conversation. Changes Ru completed before you interrupted remain in place.

The full chat workspace shows a prominent Back action for the page where that conversation started. Select it to restore the same selection (for example the collection or tab that was open) and automatically reopen the side panel with that page’s conversation. Its conversation list is sorted by latest activity and grouped under Today, This week, and Older. Drag the divider beside the chat list or the Assets/Sources area to resize those panels. Runnit saves both panel widths in your browser. Focus a divider and use the arrow keys to resize it; press Home or double-click to restore its starting width.

While a request is running, the panel shows a progress message. A response is only presented as complete after the assistant has either performed the needed action or returned the result it already has; it does not intentionally leave a request at “I’ll check that now”.

When Ru changes project or task data, active information on the page refreshes while the request runs and once more when it finishes. This also works when the request finishes while the panel is closed or you are in the full chat workspace: reopening the conversation refreshes the pages behind it. You should not need to reload the page to see completed writes.

Ru also keeps the exact successful changes it made within the conversation. A follow-up such as “undo that” or “change the task you updated earlier” refers to the most recent relevant change Ru performed, including the exact record it changed. Ru saves that action as soon as it succeeds, before it finishes writing the normal reply. If you send another message while the first one is finishing, Runnit handles the requests in order so the second message can use the first request’s completed actions.

This action history records the operation Ru actually performed, not the words it used in its reply. If you ask what it changed, challenge an outcome, report that the page doesn’t match its answer, or ask for an undo, Ru checks that history before replying. Confirmation previews and failed attempts aren’t treated as completed work. Ru must distinguish operations such as archive, deactivate, revoke, and delete. It must not claim that a record was deleted unless the completed action says it was deleted.

For task changes, Ru also keeps the task state from immediately before the update. It uses that state when you ask for a reversal, and checks the live task or its project history when needed. It does not treat the updated state as proof that nothing changed. Failed or permission-denied attempts are not saved as completed actions. A reversal still follows your normal permissions and confirmation rules, and some destructive actions cannot be reversed.

For task-editing problems, Ru checks your live permissions and the exact failed write before explaining the cause. A project target date can mark scheduled work as overdue, but it does not itself block a start-date or estimated-hours change. If the available evidence does not establish a blocker, Ru says that the cause is not yet known instead of inferring one from the current schedule.

A single message can use a limited number of tool-use steps. When a large request reaches the limit, the assistant stops and asks you to send a follow-up so it can continue from where it stopped; your progress is kept. Organisation administrators can raise the limit in Organisation Settings > Projects and AI planning > Chat assistant step limit. See Organisation Settings.

Upload documents and screenshots

You can give Ru files to read and work with. Use the paperclip in the chat composer, paste a screenshot, or drag files onto the message box. Attached files appear as chips above the composer until you send the message.

Supported files include PDF, Word (.docx), Excel (.xlsx and .xls), PowerPoint (.pptx), CSV, plain text and Markdown, HTML, RTF, and common image formats (PNG, JPEG, GIF, WebP). Each message can include up to 6 files, each up to 25MB, with up to 40 uploads per conversation.

When you send a message with files attached, Ru reads them before answering:

documents are read as text, including spreadsheet rows and the text on each presentation slide;
images and screenshots are described and transcribed by a vision model; and
scanned PDFs without a text layer are transcribed when the configured vision model supports it.

Older .doc and .ppt files can be uploaded and saved to collections, but Ru cannot read their contents. Re-save them as .docx or .pptx, or export a PDF.

Ask Ru to do real work with what it reads. For example:

“Summarise this brief and list the deliverables.”
“Create a brief from this document for Acme.” Ru drafts the brief into a new Brief Builder session, saves the source file with it, and replies with a link so you can review, refine, and hand it off to planning. See Brief Builder.
“Update Acme’s brand assets from these documents.” Ru reads each file, saves the originals into the client’s brand collections, and creates or updates the summary documents kept there.

Text inside an uploaded document is treated as content to analyse, not as instructions to the assistant. If a document contains text that tells the assistant to take an action, Ru points it out instead of complying.

Create files on the page you are viewing

Ru knows which asset collection you are looking at. When you chat from a specific collection page, such as a client’s brand collection or a project’s assets, asking Ru to “create a tone of voice summary in here” saves the file directly into that collection, and Ru confirms the destination by naming the collection in its reply. If the file would overwrite one that already exists in the collection, Ru names the file and asks before replacing it, unless you explicitly asked to update that exact file.

Combined views such as the All files view in Admin > Organisation Library span many collections, so they cannot receive files directly. When you ask Ru to create something from a combined view, it first asks one question listing the real collections as a numbered list, with a recommended fit, and only saves after you pick. This means files never end up somewhere you cannot find them. If a file still lands in the wrong place, ask Ru to move it and it reports the file’s new home.

Saving this way follows your normal asset access. AI Directives collections are the exception: their files steer the assistant itself, so they are changed through the directive editing flow instead. See AI Directives.

Directives steer how Ru works

Ru follows your organisation’s AI Directives: editable markdown playbooks that describe how each kind of work should be done. Before starting a matching piece of work, such as copywriting, strategy, or a social media calendar, Ru reads the matching directive and follows it. A client’s own directive files take precedence when the work is for that client.

If you hold the Manage AI agents permission, you can also ask Ru to create or update a directive, for example “add a directive for how we build launch plans”. Ru reads the current file first, confirms the change before overwriting, and reports what changed. Without that permission, Ru declines directive changes and suggests asking an administrator.

Ru can also report a file’s change history. Ask “who last changed resource_planning.md?” and Ru lists each version, who made the change, when, and whether it was made through chat or manually. This works for any file you can view, not only directives.

Check availability and block out dates

When Ru checks who is free for work, it evaluates real bookings across all projects and excludes weekends, organisation closure dates, and each person’s blocked-out dates, counting only the remaining hours on partially blocked days and reporting any blocked dates so unavailability is explained rather than hidden.

You can also ask Ru to list blocked dates, block out dates for you (whole days or a number of hours per day), or change or remove an existing entry, for example “block out next Monday to Wednesday, I’m on leave” or “cancel my leave block next week”. Ru can always manage your own dates; managing another person’s requires the schedule management permission. If a new block or a stricter change conflicts with scheduled work, Ru reports the affected tasks and asks whether to replan them or leave them flagged before it acts. See Blocked-Out Dates.

Build kanban boards and pipelines

If you hold the kanban management permission (managers and administrators by default), Ru can set up and change kanban boards for you: “create a sales pipeline board with Enquiry, Qualified, Proposal, Won and Lost”, “add a Handover column to the delivery board”, or “rename the Qualified stage to Shortlisted”. Ru reuses stages your organisation already has, creates missing ones with the right behaviour, and explains what it built; renames apply everywhere because stages are shared across boards. Ru can also limit a board to specific people for you (“only the sales team should see this board”), and will remind you that a board audience organises navigation rather than hiding the projects themselves. Deleting a board or a stage always requires your explicit confirmation, and stage deletions move the affected projects to a stage you choose. Without the permission, Ru can describe your boards but not change them. See Kanban & Gantt for how boards work.

Assets and generated media

The assistant grounds asset answers in actual search results. It only links to files returned from your accessible library and shows generated images or video from real generation results. Media created earlier in the same conversation can be used again, for example, ask it to animate or edit the image it just made.

If a generation fails or no generation tool runs, the assistant reports that instead of claiming a file exists. See Asset Search & Intelligence for indexing and filters.

Submit product feedback

You can report a bug, request a feature, or send an improvement from chat. If Ru’s checks establish that product behaviour is a bug, Ru also offers to draft and send a report even when you didn’t ask to file one.

Describe the issue or request, or let Ru investigate the problem you found.
Review the draft title, category, body, and acceptance criteria.
Explicitly confirm that you want it sent.
Ru sends it automatically after your confirmation and reports whether the submission succeeded.

Runnit includes your name, email, role, organisation, and current page so the engineering team can understand the report. It does not send at the draft stage, and it does not claim the report was submitted unless the submission completed. If delivery fails, Ru tells you and offers to retry.

Organisation administrators can extend Ru with trusted external tools. See Agent Integrations to connect a fixed HTTP API or remote MCP server.