Workspaces: organize projects and collaborate

A workspace is the top-level container in Genie that brings your projects and people together in one place. Instead of working in isolation, your team shares a single environment where every project is visible, activity is tracked, and everyone works from the same context. You can belong to multiple workspaces — for example, one for your company and another for a side project — and switch between them instantly from the sidebar.

​

Create a workspace

​

The workspace home

When you open a workspace you land on the workspace home — your team’s dashboard. It has three main areas:

​

Sidebar

The left sidebar is the primary navigation hub for your workspace. Workspace settings and your personal settings share one navigation, reached from the account row at the bottom of the sidebar: choose Workspace settings (General, Members, Roles, Integrations, Secrets, Knowledge access, Auto, Governance — which also holds PII Detection, Guardrails, Activities, Tracing and Spend — and, for admins and owners, Billing and the Ontology) or Personal settings (Appearance, Sign-in methods, and Notifications). Opening either replaces the sidebar with a settings navigation and a back arrow that returns you to the workspace. See Invite and manage workspace members for roles, and Account settings for your personal preferences.

​

Workspace profile

The General settings page is where an owner or admin edits the workspace itself: its name, description, a logo, and an accent palette — saved together and shown everywhere the workspace is named, from the sidebar switcher to the mobile drawer. The palette is a full set of colors, not one swatch: it paints every time-of-day gradient throughout the workspace, not just the switcher badge. Leave a logo or palette unset and the badge falls back to the workspace’s initial on a default background, with gradients following the default time-of-day rotation, same as before.

​

Projects page

Projects lists everything the workspace builds — Genie Apps, Genie Agents and Genie Tools — in one grid. Each card carries a badge naming its kind, so colour is never the only hint. One toolbar line holds the search and three menus: type (All, Apps, Agents, Tools, each with its count), scope (All in workspace, Favorites, Mine, Archive, Unlisted) and sort. You only see the kinds your role may view. The old Genie Agents and Tools pages now open this page with their type preselected, and an agent’s editor or a tool’s page work as before. Create in the page header opens a menu with one entry per kind you may build. Each starts a new Genie AI chat that asks what you’re starting from and guides you into building it. Genie Apps are still in beta, so that entry carries a Beta marker.

​

Recent projects

Below the greeting and the Genie AI prompt box, the home page shows a few of your most recently edited projects, plus a card to start another one. It’s a preview, not the full list — click View all (or Projects in the sidebar) for the complete, searchable, filterable, paginated list of every project you have access to. If you have no projects yet, you’ll see an empty state instead, with a way to create your first one. Below that, two more read-only previews follow the same shape: your most recent Genie Agents, and a few of your workspace’s Knowledge documents — each with its own “Ask Genie AI” shortcut and a link to its full page. Neither peek can run an agent or manage a document itself; they’re for a quick look on your way to the page that owns that. Both an agent’s card and a document’s row carry a cover picture, generated automatically from its name and description so a list of agents or documents is no longer a wall of identical cards. It’s generated the moment you create one, or the first time you open an older one that predates the feature — nothing to trigger yourself. If you own projects that aren’t in any workspace, an Unlisted pill appears on the Projects page; tapping one of its cards adds that project to the current workspace instead of opening it, and the pill disappears once none are left. Genie AI can list and assign them too. Star a project from its card (or ask Genie AI to do it) to keep it close: favorited projects get their own shelf above the Projects grid, lead this dashboard preview, and can be filtered for with the Favorites pill on the Projects page. A favorite is personal — teammates keep separate shelves — and the feature only switches on once a workspace holds three or more projects, since below that the grid already shows everything at once.

​

Activity log

If the activity log is enabled by an admin or owner, a recent-activity panel appears below the recent projects. It shows a live feed of workspace-wide events — new projects, member changes, setting updates — with relative timestamps. Click View all to open the full activity log page. If you’re the only member of the workspace, this area shows your plan and credit usage instead of an activity feed. The full activity log page adds an insights panel above the event list: a KPI row and a trend chart, both re-cut instantly by a metric dropdown (events, active members, or a specific category) and a time range (7, 30, or 90 days). Below that, a free-text search and a category filter search the whole log on the server — not just the rows currently on screen — so a search for a colleague’s name finds everything they’ve done, however far back it goes. A Security alerts panel sits alongside it, summarizing recent blocks from your PII rules and Guardrails side by side — visible to anyone who can already see that underlying log. From the same panel, anyone with that access can subscribe to get emailed the moment a new block happens; it’s off by default. You can also export the whole activity log as a CSV or PDF from View all.

​

Workspace AI

Every workspace includes a Genie AI that understands your workspace — your projects, your team, your credit usage, and your history. You can ask it questions and give it instructions in plain language, without switching to a project chat.

​

What you can ask

• “Which project is using the most credits this month?”
• “Send a follow-up prompt to my dashboard project: add a dark mode toggle”
• “Create a new project: a customer support ticketing tool with a Kanban board”
• “Invite my colleague as a member”
• “Show me the latest workspace activity”
• “How much does the Genie Pro plan cost?”
• “I’m seeing a blank screen after checkout — has this happened before?”
• “What database migrations has my dashboard project shipped?”
• “Does the checkout page look right on mobile?”
• “My app just threw an error — can you check what happened?”
• “Show me what the dashboard would look like with a warmer color palette”

The workspace AI can create and update projects, manage members, query credit usage, answer pricing questions, search past conversations, check whether a bug you describe has a known fix, look at your project’s screenshots or search its live console logs to see what’s actually happening, show you a design mock-up of a change before you build it, review a Genie Cloud project’s database migration history, and manage your MCP integrations — all through natural language. If you send that follow-up prompt from inside the project’s own chat, Genie starts the build right away — there is no confirm button to press.

​

Genie AI queues a busy project’s prompt

If you ask the workspace AI to update a project that’s already building, your prompt no longer just fails and asks you to try again — Genie lines it up in a queue and sends it automatically once the current build settles, spacing sends out by a short cooldown so a project isn’t hit with back-to-back builds. The card in the chat tells you plainly whether your prompt is starting a build right now or joining the queue — and, if it’s queuing, roughly when it will go out. Ask the workspace AI anytime to check, pause, reorder, or clear a project’s queue.

​

Send a message while Genie is still answering

You don’t have to wait for Genie AI to finish. While it’s answering, a queue button appears beside the stop button as soon as you type — press it, or press Enter, and your message waits above the input as a greyed-out bubble. It goes out as the next turn the moment the current one finishes (or stops), even if you’ve closed the tab in the meantime. Only one message can wait per chat, and everyone viewing the chat sees it. Its author can un-queue it before it’s sent to get the text back in the input. A message that would be refused — because of your workspace’s PII rules or guardrails — is refused right away instead of after the wait.

​

Is it still building?

Every place you talk to Genie AI — the in-project Genie AI tab, the standalone chat page, and the command palette — shows a small status badge for any project connected to that conversation, so you can tell at a glance whether it’s currently building, running an automatic fix, or has prompts waiting in its queue. Nothing shows once a project is idle.

​

Genie suggests what to build next

The moment a build finishes, the status just above your message box turns blue with an idea for what to build next. Tap it to see the suggestions as a checklist, tick the ones you want, and Apply puts a ready-to-send prompt in your message box — it doesn’t send it for you. Dismiss it and it folds away without being lost.

​

Genie AI can see and debug your project

Ask “does this look right?” or “why did my app just throw an error?” and the workspace AI can check for itself instead of guessing from your description. It can pull up your project’s live screenshots — the deployed app’s default view, or exactly what was on your own screen if you were the one looking at it — and it can search your live app’s own console output, errors, and crashes to check whether a bug you’re describing really happened, or whether a fix landed.

​

Genie AI answers with the actual code

Ask “how does checkout work in my app?” and Genie AI now reads the real code behind its answer instead of guessing from file and function names — so it describes what your code actually does, not just what its name suggests. For the files most relevant to your question it now reads the whole file rather than a short snippet, and you can also ask it to open a specific file directly — “show me App.tsx” or “what’s in the login page?”

​

Genie AI can search the web

When a question isn’t answered by Genie’s own knowledge or your workspace — a third-party API’s current behavior, a competitor, a price Genie can’t otherwise know — Genie AI now runs a live web search, or reads a specific page you point it to, instead of guessing or answering from what it already knew at training time. A web search shows up in the chat as a compact card: what was searched for and up to three sites it drew on, each a link, with the rest a tap away. Back-to-back searches share one strip instead of stacking.

​

Genie tells you when it notices something wrong

Genie watches your live preview for real problems — a crash, a thrown error, something logged as broken — and raises a calm alert right above the Genie AI composer: “Genie noticed something in your app,” together with a plain-language explanation instead of a raw error message. There’s one action, Fix with Genie AI, which hands the issue straight into your conversation so Genie can check the actual logs — and ask a question if it needs one — before proposing a fix, rather than guessing from the error text alone.

​

Genie AI can show you a design mock-up

Ask “what would this look like with a warmer palette?” or “make the dashboard feel more like Linear” and Genie AI can show you, before touching any code. It screenshots your real app, has an image model repaint it with the change you described — up to five pages at once — and shows you both pictures side by side. Genie AI is shown the same two pictures, so it describes what actually came back rather than just restating your request. React to what you see and Genie AI iterates — refine the same mock-up, or start over from your app’s real screenshot. Once you’re happy with a direction, ask Genie AI to build it and the approved picture joins the build alongside the change list, the same way an attachment you upload does — so the build starts from the mock-up you actually saw, not just a written description of it. When a draft covers more than one page, swipe between them — or use the arrows once you’ve opened one at full size — the same way you’d flip through photos. The tabs above the picture stay in sync with whichever page is on screen, and opening one at full size lets you page through the whole set, before-and-after included, without closing and reopening the viewer. Building something brand new? Describe the app you want and Genie AI drafts the key screens as a concept before your project even exists — no existing app or screenshot needed. Approve a direction and Genie AI carries it straight into the project it creates for you, shaping the very first build around what you picked.

​

Genie AI can theme your app with two colors

Ask for something like “yellow buttons and a navy navbar” and Genie AI applies both at once: your primary color drives buttons, links, and other action elements, while a second, darker color anchors navbars, headers, sidebars, and other dark surfaces. Before, only one color could be set at a time, so a two-color request would come out with half of it missing.

​

Genie AI can read and update your project’s design guidelines

Ask “what are this project’s brand colors and fonts?” and Genie AI reads branding.md — the style guide the build agent checks before writing any frontend code, covering colors, typography, component specs, and dos and don’ts. Tell it a lasting design decision — “our brand green is #0f9d58, headings are Poppins, never use pure black” — and Genie AI saves it, creating the guidelines document if the project doesn’t have one yet, so what you decide in chat is remembered by every future build instead of living in a single build brief.

​

Attachments carry into your build

Attach an image or PDF to your message when asking the workspace AI to create or update a project — a mockup, a reference photo — and the file itself goes into that build, not just a text description of it. Beyond images and PDFs, you can attach plain text, Markdown, CSV, JSON, Word documents, and spreadsheets too — Excel workbooks and Apple Numbers files both come through with their cell references and formulas intact, not just the calculated values — and an iPhone HEIC photo is converted automatically so its preview displays correctly. Before you send, click any attachment chip to check exactly what Genie will see: an image opens in a lightbox you can page through, an uploaded PDF opens in a new tab, and a document shows the exact text that will be sent, its length, and whether it was clipped.

​

Genie asks before it builds

For a clear, specific ask — “a one-page site that just says hello” — Genie starts building right away, at most with a single quick follow-up if one detail would change the result. For a broader or more ambitious idea, Genie first asks what it’s really for and where you want to take it, batched into one short card exchange rather than a back-and-forth interview. Your answers shape the very first build: the must-haves become the initial feature set, and anything mid- or long-term becomes a tracked goal for later instead of getting lost.

​

Genie checks before guessing

If what you ask could reasonably mean two different things — which project you mean, or whether “add” means a new section or replacing what’s there — Genie AI offers the readings as a quick pick instead of building the wrong one. And if you ask for the same thing again — worded differently, in the other language, or just “that still isn’t right” — Genie recognizes it as a repeat instead of sending an identical prompt a second time, and stops to check what happened last time before trying again.

​

Guided toward what Genie builds well

When you ask the workspace AI to create a project or shape a prompt, it steers you toward the kind of app Genie is proven to build reliably — a React/Tailwind frontend backed by Genie Cloud, with AI text, vision, and image generation already built in — rather than something likely to stall partway through. If what you asked for isn’t a good fit, Genie suggests the closest thing it can actually ship instead — for example, a request for real-time video effects becomes AI photo editing you can generate on demand.

​

Interactive answers

Genie doesn’t always reply with plain text. When it makes more sense to offer you a choice than to write a paragraph, it renders a clickable card right in the chat: a button to confirm a single action, a list to pick exactly one option from, or a checklist where you can tick as many choices as apply. Either kind of choice card always has a free-text field under the options too, so you can type your own answer if none of them are quite right instead of having to cancel. Tap a card to answer — if you send a new message instead without answering, the card is cleared automatically so it never lingers out of date. If a card needs an answer before Genie AI can continue — confirming an action, or reviewing a project it’s about to create — you can minimize it instead of answering right away or canceling it outright. Tap it again anytime to bring it back; nothing about the request is lost while it’s collapsed.

​

Open a Genie Tool or Agent beside the chat

On the full chat page, building or opening a Genie Tool or Genie Agent no longer just scrolls past in the transcript — its card docks to a canvas on the right, with the conversation keeping the left. If more than one is open, a button on each card’s header switches the canvas to show it instead; closing the canvas (or that button on the docked card) folds it back into the conversation. On a phone or a narrower window, cards stay inline in the transcript as before. Any Genie Agent action Genie AI takes in the chat — not just building one — opens that agent on the canvas. A Genie Agent with no branching draws top to bottom on the canvas — start, each step, result, one per row — instead of the sideways grid the editor uses, so it reads cleanly in the canvas’s tall, narrow column. An agent with a branch, or one you’ve hand-arranged in the editor, keeps the grid. When the chat has more than one thing on the canvas — a Genie Tool, a Genie Agent, the knowledge model — the canvas’s bar shows one tab for each, in the order they arrived; switch with a click or the arrow keys. A Genie Tool or Agent that Genie AI is building or editing appears on the canvas while it happens. Knowledge on the canvas. When Genie AI works with your knowledge model in the chat, the entities it names or changes are drawn on the canvas as a small chart with their closest surroundings, following the newest step as the chat goes on. What Genie AI staged in a draft is drawn in orange — dashed for entities, dotted for links — so you can tell it from what is already live. A layer picker on top of the canvas switches which layer the chart shows, and the canvas only opens when you click it. A sheet that stays out of the way. Click a card on the canvas and its details open folded, as a slim pill with the entity’s name, class and values, so the chart you clicked into stays in view. Click the card again, or drag the grip, to unfold it to half the canvas or nearly all of it; Esc steps back down and then closes it. Entities that aren’t linked to anything are drawn as a grid of larger cards, two columns on a wide canvas and one on a narrow one, each listing up to six values. Tabs you have already opened keep their place — the chart, a PDF page or a table’s search is exactly as you left it when you switch back. Open a document beside the chat. Ask to see the slides or where a document says something and Genie AI opens the file itself on the canvas, with the passage highlighted. On a phone, the card’s View opens it instead. Clicking a highlighted name in an answer opens that entity on the canvas on a desktop, with its questions and Not this at the top of its panel. A Genie Agent shows on the canvas as a short list of steps, like an automation, with what each step uses shown as chips. Charts and flowcharts. Ask Genie AI for a chart — bars, lines, areas, pies, donuts or scatter plots — or a flowchart of a process or an approval path, and it draws one straight into the chat. It can chart rows you paste, a workspace table, the numbers behind a Genie Tool, a Genie Agent’s latest run, or what the knowledge model holds (entities per class, process stages, a class’s values), always within what you may see. On the full chat page every chart stacks on the canvas beside the conversation, and each one downloads as PNG or PDF. When a chart covers only part of the data, it says so under the figure. Press an entity — in the chart, or from a chip under it — and its card opens as a panel under the chart, in your light or dark theme, folded to one row with its name, its class and its main values; press the row to unfold the rest — its sources, its process and, if you may edit the model, Edit and Layer. The chart gives up the room it takes, and Ask Genie AI buttons on the panel write into this chat’s message box instead of opening the command palette. When the entities the chat named have no links among them, they show as detail cards instead of an empty chart. The canvas also follows a draft being published, so it shows what is live once the run finishes. It draws up to 1,000 entities at a time. The entity you pick wears your workspace’s accent colour, and a card Genie AI created or changed in this chat carries a small new or changed badge; press it to see every step, with old and new values. A click opens the panel unfolded and a click away folds it to one row. Workspace tables on the canvas. When Genie AI reads or changes one of your workspace tables, the table opens as its own tab on the canvas — 50 rows a page, searchable and sortable, with the rows it just changed highlighted. Press a row, or Add row, to edit it right there. A table’s name in an answer is marked, and pressing it opens the table (in a sheet over the chat on a phone). A Genie Agent’s table step shows its table’s entries the same way in the agent editor. Close and reopen canvas tabs. Each tab has its own ✕, and the canvas has one more to close it all. Closed tabs stay closed the next time you open the chat, and Undo brings one back. A closed tab comes back when Genie AI has something new for it, when you press a mark for it, or from Show the canvas in the chat header. When a chat is shared, each tab is shown only to those who may see what is in it. Calculations you can check. When Genie AI states a figure it worked out — a sum, a share, an average — press the figure to see the arithmetic step by step and where each input came from. A figure only quoted from a document or table is a citation instead. Genie AI follows up on runs it started. When Genie AI starts a Genie Agent from the chat, it comes back on its own when the run ends: a small note says the run finished or failed, with Open run, followed by a report of what the run did, which step failed and why, and what to do next. If you closed the tab, the usual “Genie AI is done” notification tells you. Take another perspective. Press Work from this perspective on an entity card — often a person — or ask Genie AI to, and the chat answers from that entity’s point of view, for example “what did Anna own?” A pill over the chat shows it and clears it. It is a focus, not an identity: it never widens what you can see, and nobody’s mailbox or connections are used. Hand over a departed person’s work. When someone leaves, open their entity card and press Hand over to a successor — or ask Genie AI, “Hand over Anna’s work to Ben”. Every confirmed responsible for, leads and works on link that Anna holds moves to Ben, keeping its history, sources and values; the same links Ben already has, and links that can’t move, are skipped with a reason. It works in steps of ten links, so a large handover continues until it is done, and you can ask for a preview first that counts what would move without changing anything. Proposed links still waiting in the review queue stay where they are. You need permission to edit the knowledge model. Genie AI explains instead of going quiet. Every answer comes with text, and when Genie AI can’t or won’t do something, it says why. A chat turn that cost limits held back says so too.

​

In-app links open as chips

When Genie AI points you to a page in your workspace — a run, a document, a tool — the link shows as a small chip with the destination’s icon and a readable name instead of a long address. Click it to open the page. Links to other sites look as before.

​

Fix an integration from the chat

Ask Genie AI to reconnect or repair an integration and it shows the integration as a card in the chat: its live status plus the same buttons as its row in settings — Connect or Connect again and, for owners and admins, Start over, Check it works and Pause / Resume. If you can’t take a step yourself, the card says who can. See Manage connected servers.

​

Chat templates

A chat template starts a Genie AI chat that already knows what you want to do. Open the Templates tab in the Library and press Start chat on a card: Genie opens with a short message and a choice card with the template’s options, then carries on from your pick. The first template, PowerPoint presentation, asks whether you want to start a new deck, edit one from your knowledge base, or upload one. The page has two sections: From Genie, templates available in every workspace, and This workspace, your own. Members create and edit their own by asking Genie AI (“make a template for weekly status reports”) — each template gets a generated cover picture, and can be translated so every member sees it in their language. Deleting one asks you to confirm in the chat. The same templates can be listed and managed through the Genie MCP Server.

​

Chats name themselves

A new chat names itself as soon as you send its first message: a short title in your language, plus an icon and a colour, appear in the chat history and the chat header within a few seconds — usually while the first answer is still streaming. Rename the chat or pick another look at any time (click its badge), or ask Genie AI to do it. Your own choice is never overwritten.

​

Typing bubble and step list

While Genie prepares an answer, a typing bubble with three moving dots holds the place of the reply. Once Genie starts working, a steps control shows what it did in plain language (“Checking your team”, “Looking inside your app”) with a loading, done or error state for each, and it stays collapsed so the answer stays in front. When Genie AI remembers or forgets something for you, a small card says what.

​

Get notified when Genie AI is done

If you close the tab or look away while Genie AI is answering, you get a push notification when the answer is ready: “Genie AI is done — Your answer in Acme is ready.” Nothing is sent while you’re watching the answer arrive, and on a second device you’re not looking at, the notification still arrives. It’s on by default; switch it off under Personal settings → Notifications, or ask Genie AI to. See Account settings.

​

Always replies in your language

Genie AI answers in whatever language you write to it in — regardless of your account’s interface language — and keeps writing in that language for the rest of the conversation, including any cards it shows you and the project briefs it authors on your behalf. If you switch languages mid-conversation, Genie follows along from your very next message.

​

Genie remembers your preferences

As you chat with the workspace AI, Genie remembers things you share about yourself — your industry, your team setup, the kinds of apps you build — and uses that context in future conversations automatically. You don’t need to repeat background information in every chat. You can also ask it directly — try “what do you remember about me?” — and Genie searches everything it’s picked up, not just the most recent items. If something’s outdated or you’d rather it didn’t know, just ask Genie to forget it. Genie saves lasting facts on its own as you chat, and keeps what only matters for one workspace — a client’s team size, say — in that workspace, visible only to you there. General facts about you, such as your role or your preferred tone, follow you into every workspace. Genie AI only suggests next prompts from the general ones, since a project’s suggestions are shown to everyone with access.

​

Genie remembers your project

Beyond what it knows about you, Genie AI also keeps notes that belong to the project itself, shared with everyone who has access to it — a standing requirement you’ve stated (“every screen must work on mobile”) or a confirmed detail about a third-party service you’re integrating (“the webhook payload nests under data.object”). Once you’ve told Genie something like that, you don’t need to repeat it in the next conversation, and it’s there for your teammates too. Ask “what do you remember about this project?” to see what’s recorded. If a requirement changes, just restate it — Genie updates the existing note in place instead of piling up a contradicting one. Genie never stores credentials or secrets this way, only confirmed details and standing instructions.

​

Knowledge base

Beyond what Genie AI remembers from conversation, your team can file the material Genie should build with — brand guidelines, tone of voice, an API schema, a data dictionary — as knowledge documents, at two scopes:

• Workspace — applies to every project.
• Project — applies to just the one project it’s filed under.

Open Library in the workspace sidebar to browse and manage documents at both scopes, or open a project’s Settings → Knowledge tab to manage just that project’s — either surface can also see and file the other kind, and moving a document between scopes keeps its content and search index intact. Workspace owners and admins can file a document into any project in the workspace, not only ones they’re a member of. The browser is shaped like a familiar file manager: folders and unfiled documents sit at the root, opening a folder shows what’s inside it, and a breadcrumb walks you back up — nothing is ever listed in two places at once. Folders and files share one list, and your connected cloud drives sit at the bottom of it. Search drops the folder view entirely and searches everything at once, so finding a document never depends on remembering where it was filed — and each result shows small chips saying why it matched: the entities it is cited for, its name, its text, or a related topic. Switch between a list or grid layout (remembered for next time), sort any column, and drag a document straight onto a folder to file it — or onto the Knowledge breadcrumb to take it back out — instead of opening the move dialog. A search in the Library also looks through the knowledge model: the classes, entities, process stages, relationship types, and layers your query names appear as the first rows of the results, followed by the entities directly linked to them. Upload plain text, Markdown, CSV, JSON, YAML, XML, HTML, source files, Word (.docx), Excel (.xlsx), PowerPoint (.pptx), or PDF files — up to 50 MB each, with up to 1,000,000 characters of text read per file. Pictures (JPEG, PNG, WebP, TIFF, GIF, BMP), older Office files (.doc, .xls, .ppt) and PDFs that mix typed and scanned pages are read too: a scan or a picture is read with text recognition (OCR) instead of being rejected, and a mixed PDF keeps its typed pages as they are. You can also upload a whole folder at once — Genie recreates its structure as knowledge folders, skipping anything it can’t read rather than failing the whole upload. A workspace holds up to 50,000 documents and 5,000 knowledge folders. Email files (.eml, .emlx, .mbox, .msg, .oft) are read too: a mailbox is read message by message, oldest first, and a file that only carries a mail’s name but isn’t mail is read as the text it is. Upload page. Upload in the Library opens a page of its own. It shows how far the whole upload is — files waiting, uploading, filed and failed, with a progress bar — and lists every file as a row you can search, filter by status and sort. The upload keeps running if you go back to the Library, and a running upload shows on the New button. Folder uploads skip system files such as ._*, .DS_Store and Thumbs.db. Preview. Double-click a document (or choose Open preview) to see the file as it looks, not only the text Genie read: a PDF page by page with zoom and the cited passage highlighted, a Word file with its headings, tables and pictures, a PowerPoint deck slide by slide with speaker notes, and an Excel workbook as a tab per sheet. Macro-enabled and template flavours open too (such as .docm, .pptm, .xlsm and .xltx), and so do older .xls workbooks and Apple Numbers files. The Text tab shows the extracted text, and the side panel has the document’s details, Download, and Ask Genie AI. Citations, chat cards and search results open the same preview. A very large file says where its preview was cut. Use the find box in the preview to search the document: matches are highlighted in the page, slide or sheet you’re looking at, and the names the knowledge model knows are marked in the text and listed in the side panel. Pictures show in the preview, and video and audio files play with the browser’s own player; a picture has no text, so search covers only what was read from it. Delete a folder. Deleting a folder asks what should happen to what’s inside. By default its documents just become unfiled; switch on delete contents to remove them too, together with the knowledge-model entities that only those documents support. A preview lists what would go before you confirm. Genie AI can do the same on request. Network drives. Besides Choose files and Choose a folder, the upload page offers Network drive. An owner or admin connects an Amazon S3 bucket (or one folder in it) once with an AWS access key that can only list and read it; Genie checks the key against the bucket before saving it, and shows a read-only policy to create one. From then on every member can browse the drive, pick files to import, or keep a folder in sync like a cloud drive. Because everyone in the workspace can browse a network drive, connect only a folder every member may see — a synced folder from it can’t be filed into a restricted access area. Genie AI can read the drive and open the connect form in the chat. Each document shows its own search-index status — indexed, keyword-only if the semantic index failed to build, or not yet indexed — with a Retry indexing action right there when it needs one. Each document also shows the knowledge model class it belongs to as a small badge, and owners and admins get a Class filter chip alongside the scope filter to narrow the list to one class. The browser’s folder, search, sort, and page all live in the page’s address, so a link or a bookmark reopens exactly where you left it. Before building or updating a project, Genie AI automatically checks the knowledge base that applies to it — the workspace-wide documents plus anything filed just for that project — and carries what it finds into the build. When it does, its answer marks exactly which passage each statement came from with a small citation — click one to open the source document at that passage, or open the Sources list under the answer to see (and download) every document it drew on. Genie AI can also find a document by its filename or description alone, not only by what’s written inside it — useful for a document whose content is mostly figures or a scan, like a spreadsheet of numbers or a signed contract. Ask it to show or open one directly and it renders as a card right in the chat, with actions to view or download it, instead of just naming it in the reply. Each document also carries a date it is about, shown beside it wherever it’s listed or cited, separate from the day it was uploaded. Genie works this out automatically — a full date in the file name or title, or else a connected drive’s last-modified time — and, since a document’s name doesn’t always carry one, a small model also reads the document’s own text for a date it’s about during indexing, falling back to the upload day only when none of those find one. Correct it yourself from the document’s Rename or move dialog, or ask Genie AI to read the document and set it for you; your correction sticks through every later sync or edit, until you switch it back to automatic. The same is available through the Genie MCP Server and the workspace API. Within either scope, documents can also be sorted into folders, which can now hold sub-folders of their own — a filing layer for finding things, separate from scope, which decides what Genie AI can use. A “Brand” folder can hold both a workspace-wide document and one project’s own copy deck, and filing something into or out of a folder never changes what Genie finds when it searches. Create, rename, or delete folders from the chip rail above the document list, or drag one into another to nest it; each holds up to 200 documents directly, and deleting a folder never deletes what’s in it or below it — those documents just become unfiled. Ask Genie AI to update a document and it can rewrite its full text in place, not just retitle or re-describe it — the new text is re-indexed immediately, so search and the next build see it right away. The same is available through the Genie MCP Server and the workspace API. A document mirrored in from a connected cloud drive can’t be rewritten this way, since the next sync would overwrite the edit — change it at the source instead.

​

Cloud drives

A Cloud drives panel at the bottom of the Library page shows where each member stands with Google Drive or Microsoft 365 (OneDrive and SharePoint) — not connected yet, or connected and ready to browse. Once you’ve connected your own account (see Available preset integrations), it appears as its own place inside the Knowledge browser: search across the whole drive — your OneDrive, or every SharePoint site you follow, for Microsoft 365 — open folders, and click Import into knowledge on any file Genie can read to file a copy as a knowledge document, citable like any other. Beyond importing files one at a time, you can also keep a whole folder in sync — a Drive folder, or a OneDrive/SharePoint library or a folder inside one: pick it (or just the sub-folders you want) once from the Synced folders list, and Genie mirrors it into your knowledge base automatically from then on — new and changed files are brought in on a Sync now whenever you want it, and on its own every few hours in between. Move a file between two folders you sync and it stays the same knowledge document instead of being filed again; remove one from the source and its document isn’t deleted either — it’s marked Removed from the drive, drops out of search and Genie AI’s answers, and comes back as itself if the file does too. It stays that way for 7 days in case the file returns, then it’s deleted for good. Sync tab. The Library’s Sync tab gathers your cloud and network drives, your synced folders and what reading them into the knowledge model has done. Open a synced folder’s document count to see which of its files made it into the knowledge base and why the others didn’t — No text, Too large, Unsupported type, Failed or Not selected — and the same chips appear on files as you browse the drive. Before you start a sync, the setup dialog estimates what it will cost in credits, folder by folder. Genie reads a synced folder’s whole tree, however many sub-folders it has, and first indexes the files by type with a time estimate for the rest. Each synced folder has a progress page — open it from the Sync tab — that shows what is in, what is waiting and why files were left out. Large PowerPoint decks on an Amazon S3 drive are read without downloading the whole file, and PDFs, pictures and older Office files sync like uploads. Owners and admins can switch on sync alerts in the Sync tab: you get an email and a push notification when a synced folder needs attention or recovers (off by default). The progress page breaks every file count down by type and size, charts how fast the folder has been syncing, splits Still to import by file type, and always says why and when a run ended. Ontology reading progress. Under the Sync tab, a progress page for the knowledge model shows how many documents are read, being read, waiting or failed, how fast Genie is reading, and when the rest should be done. Owners and admins can Pause reading and Resume reading there; documents already being read finish first. Ask Genie AI to pause or resume it for you. Why files failed, and trying again. The sync progress page and the knowledge model’s reading page now say which step went wrong — fetching, reading the text, recognising text in a scan, saving or indexing; model, database or PII rules while reading — and list the latest failed files. Try failed files again starts a run that retries them, and Try all again does the same for the knowledge model; owners and admins can use them, and Genie AI can too. Each synced folder has a Generate cover images switch and an Include images switch. Turn Include images off for a drive full of logos and photos: pictures are left out (listed as Images off in the file list), documents already imported stay, and switching it back on reads them. The file list also shows each file’s type, size and date. Pause a synced folder to hold its sync, and Resume to carry on; if the current run is slow to notice the pause, Stop now ends it. Anyone who can pause a sync can turn images off; only its owner can turn them back on. The Generate cover images switch: Turn it off to skip the generated cover picture for every new document the folder files, which for a large folder is often the biggest cost of the first import; covers already drawn stay. System files such as ._*, .DS_Store, Thumbs.db and Office ~$ lock files never show up in a drive listing or a sync.

​

Excel workbooks

Genie AI can build and edit real Excel workbooks in chat, not just read them. Ask it to “build me a budget workbook with a formula column for VAT” and it hands back a working .xlsx — sheets, data, and formulas calculated immediately — instead of a markdown table you’d have to retype. Give it a workbook you’ve already uploaded and ask it to add a column, insert or delete rows, sort a sheet, or reformat a range, and it edits the file in place and recalculates every formula that depends on what changed; ask it a question like “what’s the total for March?” and it evaluates the formula against the workbook’s current values without needing to save anything. A workbook is stored as a knowledge document, so it’s filed, searched, and shared the same way any other one is — see Knowledge base above.

​

Convert a workbook into your workspace

The Excel conversions page turns a workbook into the classes, records, and links your workspace runs on, one step at a time. Upload an .xlsx file and Genie first surveys its structure without AI — where the tables are, which column identifies a row, and which columns look up another table. You then see what reading the regions with AI will cost in credits before you start it. Review the resulting map, correct what Genie is unsure of, and choose Approve the map. Nothing in your workspace changes until you approve a step, and a conversion left idle for 30 days expires. The next steps create the classes, import the rows as dated records and link them, turn ledgers and flows into workspace tables, turn status columns into processes with the imported records placed in their stages, and propose Genie Tools and calculators from the workbook’s formulas, each created at its own approval; an owner or admin has to approve what needs those permissions. Use Ask Genie AI on the page for help checking a region.

​

PowerPoint presentations

Genie AI can read, create, and edit PowerPoint decks, not just open them. Attach a .pptx to a chat and ask it to “summarize this deck” or “turn it into a Genie Tool”, Give it a deck you’ve uploaded and ask it to fix the typos on slide 4, reorder slides, add a table, or edit speaker notes: every request is saved together or not at all, and the result shows as a document card with the actual file to download. Text, tables, speaker notes, chart data, and picture descriptions are all read, so a deck is searchable and citable like any other knowledge document. Decks can be up to 25 MB in a chat or edit (10 MB when imported from a cloud drive). The same tools are available through the Genie MCP Server. Decks can also be designed: Genie AI draws background pictures, illustrations, or whole designed slides, then places them with shapes, cards, and styled text, and can set a deck-wide colour palette and fonts. Pictures count toward a daily per-workspace limit. Ask for a new presentation — “build a five-slide pitch” — and Genie AI builds it as an app: a web slide deck with its own link, not a .pptx file. That link opens for anyone who has it unless you lock the app with private mode, and private mode only guards sign-in and the app’s data, not text built into the app. So Genie AI tells you before a deck carries confidential workspace content, and won’t offer private mode as protection for it.

​

Knowledge model

Beyond search and citations, Genie AI can also build a structured model of what your knowledge base actually talks about — the entities in it (a person, an organisation, a contract, a decision) and the relationships between them, each one traceable back to the exact passage it came from. The Graph is a tab of the Library, so every member can open it. The Knowledge page switches between its views of it — Documents and Graph — from a header of its own (owners and admins open it from Workspace settings → Ontology; members browse the documents in Library). Turn on Read documents automatically at the bottom of the Documents view (owners and admins only, off by default) and Genie reads every document in the workspace and proposes the entities, values, and links it finds — new and changed documents are picked up automatically from then on. Everyone in the workspace can browse the result in the Classes and relations panel further down the same page: the kinds of things being tracked (people, organisations, decisions, terms, and any custom class an admin adds), how many properties each has, and how many entities you can see. Open a class to see its entities, each marked Proposed until confirmed, and Partly hidden where a document you can’t read says more about it. New entities, links, and classes don’t apply themselves — they wait in the Review queue below the classes panel until someone with permission decides. Confirm or reject a proposed entity or link one at a time, or use Confirm all to accept everything currently listed at once (it runs in the background, and the queue shows exact counts) — or pick one kind from its menu, such as only entities or only duplicates; a possible duplicate is offered as a merge (with a reason), and a class or relationship Genie thinks your workspace needs is offered as a suggestion you can create or dismiss. Admins can also create and edit classes and relationship types themselves — name, description, properties, and which classes a relationship connects — from the same panel, and can re-read every document from scratch or reset the model entirely if it needs a clean start. Search the model from the Library, or press any entity in the Graph, a chat answer or a list, and it opens on its own page: its names, properties and every source document that mentions it, and beside them its surroundings drawn as a map, where a neighbor opens its own page. Drag the date scrubber to see what was true on an earlier day, as links close and values change over time; it answers as you drag. Each value shows a small icon and color by its name, and admins can adjust them per class in the class editor. A file’s card says how it was added (created by Genie AI, synced from Google Drive) and who edited it last. Every confirmed entity also carries a one-sentence description under its name, written for you from what everyone in the workspace may see; edit it yourself or ask Genie AI to, and your wording is never replaced. When a new name looks like another place at the same street, such as Tiefgarage Neumühlen 1-5 next to Neumühlen 3, Genie offers a possibly related link for you to confirm or reject, never a merge. Saved views: ask Genie AI to show how things connect and it composes a view — the entities that answer, a note on each, a title — and saves it under the Library’s Saved views tab. A view is private until you share it, and downloads as PNG, PDF, JSON or Mermaid. Move a card on a saved view and the arrangement is kept in the view, for you alone unless you share it. Graph draws the same information as a live, interactive map — every entity a dot sized by how connected it is, relationships as quiet lines between them — that updates on its own as the model changes. Pan with a drag, zoom with the wheel, a pinch, or the zoom bar, and point at or pick an entity to see its name and its neighbors’ names. A flowchart (cards that show each entity’s class and its first links without a click) and a list are one switch away. Both link back to a source document on the exact passage that supports what they show, and respect the same access areas as the Documents view — an entity or link only a folder you can’t reach cites simply isn’t there for you. Ontology tab and model details. The Library’s Ontology tab shows the knowledge model at a glance: entity, class, relation and process counts, entities per class, and how classes connect. A class can be a kind of another class and then carries its parent’s properties too. An entity can keep its former names with the days they were used, so a document that still uses an old name resolves to the same entity; Genie AI can set and clear them. Each fact card also shows dates, review status and process history, and a file the model knows appears as its own row on its card. When Genie finds a possible duplicate it asks a decision model to rate it before proposing a merge. The Graph’s flowchart follows your theme — a light stage with white cards in light mode, a dark one in dark mode — wherever it appears, including the chat canvas. Processes. A class can carry a lifecycle — the stages its entities move through, like a contract going from Draft to In review to Signed. An admin defines the stages and the allowed moves between them (a move can ask for a reason, and can be limited to people who manage the model) in the class’s editor; an entity then shows its current stage as a chip and a stage panel with its history. Each move is recorded with who made it and when it took effect, and a move can be dated in the past to record how things actually went. Genie AI can move an entity along its lifecycle for you. A process view shows how many entities sit in each stage right now, how long they typically stay, and how many reached an end; select a stage or a move to see its details, or scope the view to one layer. A move can start a Genie Agent — press Create a Genie Agent on a move. Link values and tenses. A relationship can carry values of its own — “holds shares in · 50 %” — defined on the relationship type and set by you or Genie AI, each traceable to its source. A relationship type can also have past and future wording (“worked for”, “will work for”), so a link that has ended or not begun yet reads correctly on the Graph’s flowchart: ended links struck through, upcoming ones dashed. Press a link on the flowchart to open the links it stands for, with their dates and sources. Conditional links. Some relationships only hold in company: “Anna likes Weißwurst, but only with mustard”, or “Carl drinks coffee, but never with milk”. A link can carry a condition — it holds only with certain other entities, or never with them — and the entity’s card, the review queue, and the Graph show it (the flowchart draws these links with long dashes). Two links between the same pair can both stand when their conditions never apply together, such as different values with and without mustard. An entity’s card also lists what it makes possible and what it rules out. Owners and admins can correct a condition; Genie AI can set and check conditions and tell you which links hold in a given situation. Genie also reads conditions from your documents: when a passage states one for a link that already exists, Genie doesn’t overwrite it — it raises a suggestion in the review queue showing what the link holds under now and what the documents say, and you choose Set condition or Keep as is. A condition you rejected isn’t suggested again by later documents. You can also add conditions yourself or ask Genie AI. On the flowchart, a companion’s effect on a link is drawn as a line to the link’s label — dotted green where it makes the link possible, dashed red where it rules it out. Link dates. A link Genie reads from a document starts on that document’s date, so the date scrubber no longer shows it before the document existed; a document with no date of its own leaves the start open. If an earlier document states the same link, the start moves back to that day, and a passage dated after a link ended proposes the pair as linked again from that day. Links Genie recorded earlier are dated in the background. A new or reset workspace starts with a set of common relationship types — works for, reports to, customer of, part of, and so on — so the very first reading already records links. Anything they don’t name stays a suggestion in its own words, which is what new relationship types are proposed from. Confirm all now confirms everything waiting in the review queue, not only the rows you can see, and can be stopped part-way. It also accepts the suggestions the queue lists: possible duplicates are merged, suggested classes, properties and relationship types are created, and conditions documents state are set. Only the leftover rows of a split are left out, since each needs a place. The button sits at the top of the queue and its dialog lists what it will do. Layers. A large company’s knowledge doesn’t need to be one tangle. A layer is a named area of the model — Sales, Engineering, HR — so a team can look at its own part or at the whole while links still run between them. It is still one model: classes, entities, and links are filed in a layer (or in the shared core everyone builds on), and a view scoped to a layer lists its own entities first, plus the links that reach into other layers. Search says how many matches it left out of the layer and offers Search all layers. Who is who never depends on a layer: a new record that looks like one in another layer is offered as a merge instead of quietly standing beside it. Owners and admins manage layers; Genie AI can file things in them. The Graph runs the full width of the page. Pick an entity and its card floats over the right side of the map instead of taking a permanent column; close it with ✕ or Esc. Use the fullscreen button to give the map the whole screen, and Hide unlinked to drop entities with no drawn link and lighten the view. It opens with up to 5,000 entities and can show up to 1,000,000, loading the rest with a status line while the map stays on screen; the list view reads up to 10,000. Hide detached clusters keeps only the largest group of linked entities and drops the chains and stars around it. A search on the Graph page is fast and also finds entities by meaning — “who supplies our tomatoes” finds the company cited in your supply plan — fading what didn’t match in the flowchart. An entity with no links offers Why no links?, and each fact on an entity’s card lists the passages it came from. An entity’s card on the graph lists its own values, and an entity linked only to one hub folds into that hub’s stack card. The page has no header band — controls float over the map as icon buttons (views, filters, legend, refresh), and anyone allowed to edit the model can change an entity’s names, values, and look straight from its card. Web and email addresses in an entity’s values are clickable, and Markdown documents open in a source’s citation sheet fully formatted, with the cited passage highlighted. Entities in answers. When Genie AI answers in a workspace chat, the people, projects, companies and codes from your knowledge model that the answer names are marked in the text. Hover a mark, tap it, or focus it and press Enter to see a card with what the workspace knows about it — and ask about it or open its full details. A mark only appears when a name points to exactly one entity you may see: a name two people share is left plain, and what you can’t open names nothing. Teammates in the same chat can see more or fewer marks, since each person’s view follows their own access. Table rows. Notes, tasks and logs in your workspace’s tables are read for the names they mention, with nothing to set up. An entity’s card lists them under Mentioned in, and Genie AI sees the links when it reads the table. A grid in a Genie Tool gets a Mentions column once its rows hold free text: press Not this one on a chip to take a name off a row, or ask Genie AI to link words it couldn’t decide — both stay put on later reads. Editing this takes permission to edit the knowledge model. What a value names. Open the version history of a text value on an entity’s card to see which other entities that text names, with the link to each — dashed while it is only proposed. Look up (or Look up again) reads the value now and offers a link to every entity it names that isn’t linked yet; Genie AI can do the same. A value that names a linked entity ranks above one that doesn’t, and an address value such as Hauptstr. 2, 22763 Hamburg offers a link to places the model records at that address, like Bürohaus Hauptstraße 2. Big merges run in the background. Merging or splitting an entity with a great many links no longer has to fit in one step: the decision is made at once and the links move in the background, up to 200,000 rows. The entity’s card shows a Merging · 61% chip while it runs, and an entity being moved can’t be merged or split again until it is done. Classes as a table. Classes and relations lists classes as a searchable table that loads a page at a time, so it stays quick with hundreds of classes. Each row shows how many properties the class has, how many of its entities you can see, and its process as a row of stages. Press a row to open the class in a side sheet with its properties, process, relationship types and entities. The Ontology page in Workspace settings now holds settings only. Names that mean the same. Genie also compares names by meaning, not just spelling, and proposes a merge when two names in one class clearly refer to the same thing. Like every duplicate, it waits in the review queue for a person to decide. On the Ontology settings, owners and admins can pause the AI duplicate check and give it a daily budget; once the budget is used up the check waits until the next day. Confirm from the card. Press Confirm on a proposed entity’s own card, in the app or on the chat canvas, to accept it without opening the review queue. With drafts on, the confirmation lands in your draft. Link yourself to your own Person. Questions like “who is my manager?” need a starting point. Ask Genie AI “who am I in the knowledge model?”, pick your card from the suggestions and press Link to me. Only you can set or see this link, and nobody else can set it for you. Genie AI never assumes a card is yours just because it carries your email address. Better answers from your documents. Genie AI’s document search now re-ranks its best matches, so the passages that actually answer the question come first. When it answers from an entity’s page, the claims carry the same numbered source markers as document answers, and a marker opens the document with the passage highlighted.

​

Drafts

In a workspace with staging on — the default for new workspaces; owners and admins switch it in the knowledge page’s extraction panel — changes to the knowledge model don’t land on the model everyone reads. They go into your draft: you, and Genie AI when it edits for you, see the live model with your changes on top, while everyone else keeps seeing it as it was. Reading documents, integrations and other automatic processes keep writing to the live model, called main.

• Pick where changes go with the draft switcher in the knowledge views: Main, My draft, a shared draft or New draft…. A shared draft has a name (“Q3 reorg”); its owner adds members as editors or viewers.
• See what changed. A banner shows the draft’s name and how many changes and conflicts it holds, and entities, links and classes carry New in draft, Changed in draft, Removed in draft or Conflict badges. An entity’s card lists each changed field, main’s value beside the draft’s. The Drafts page lists your drafts, drafts shared with you and — for people who can publish — drafts awaiting publication.
• Check and publish. Check against main tries a draft without changing anything. Members with the Publish knowledge changes to main permission publish a draft, which applies its changes to main one by one, each as its author; everyone else uses Request publish. A publisher can also pick Main to edit live, for eight hours at a time.
• Conflicts. When main changed something your draft also changed, you’re told and choose Revert own change, Swap to mainline (take main’s version) or Keep mine.

Genie AI stages into your draft like you do. Anything that drops changes — reverting, swapping or discarding a draft — asks you to confirm on a card first. When Genie AI shows a draft on the chat canvas, it draws the draft’s changes in orange, and a publish card waits for the publish to finish and tells Genie AI how it ended. Where a document came from. A document’s details name how it got into the workspace and for whom — for example Synced from Google Drive by Anna · 9 Oct 2026 — its last edit, and, for a synced drive file, the folder path and file name. A member who can’t see a fenced folder never sees its path.

​

Access areas

By default, every workspace member can see every knowledge folder. If a folder holds something more sensitive — HR records, one client’s own files — an owner or admin can fence it into an access area from workspace settings → Knowledge access: create an area, decide whether it’s open to every member by default or only to people you grant, then add the roles, groups, or individual members who should reach it and put the folder into it. Anyone outside an area’s reach doesn’t just lose access to the folder — they don’t see that it exists at all. Each area is one card on the Knowledge access page, with a one-line summary that opens onto three questions: who can see it (a Private / Everyone switch, every grant as a row you change in place, and one picker for groups, roles and people), which folders are in it, and what the knowledge model learned from them — the entities cited from the area, by class, and how many of their cards are invisible outside it. Creating an area asks only for a name, who can open it and an optional description. Genie AI can read the same summary. An area can grant read access (see and use what’s inside) or write access (also add, move, and delete documents in it). Moving a folder into or out of an area, or changing who reaches it, applies to everything inside it, including sub-folders. If every document an entity is cited from moves into an area that isn’t open to all members, its card follows: it becomes visible only to people who may read those documents, and anyone with permission can widen it again. This only ever tightens — a card is never opened up automatically, and one you set yourself is left alone. Personal areas. Some things belong to one member alone — their own synced mailbox, for example. Those land in a personal area that Genie creates for that member: no other member can read, change or add to it, and it can’t be shared or relabelled. The workspace’s owners and admins can still read it, but they can’t write in it. When its owner leaves the workspace, the area is held, never opened to anyone else. A mirror of your own Google Drive or OneDrive is private by default too: when you set up a synced folder, the choice of who may see it starts at Only me, and sharing it with every member or an access area is a choice you make. A private mirror stays private; to share it, stop it and set up a shared one.

​

Talk instead of type

Don’t want to type your prompt? Tap the microphone icon in any Genie chat composer, or next to the Get started button on the workspace dashboard, to dictate instead. Your words appear in the composer as you speak, rather than arriving all at once when you stop, and dictation recognizes Genie- and product-specific terms — like “Genie Cloud” or “MCP” — instead of mishearing them. The microphone stays available even after you’ve started typing, and even while a response is still streaming in or an earlier conversation is still loading, so you can dictate a correction or an addition onto a draft without clearing the box first; tap it again to stop recording, or it stops on its own after five minutes. Genie transcribes what you said so you can review or edit it before sending — nothing is submitted automatically. Voice input is available in the workspace chat, a project’s chat panel, the workspace dashboard, a Genie Tool’s own form fields, and the command palette’s Ask Genie, on any browser that supports microphone recording.

​

Workspace chats are private by default

A conversation with the workspace AI is visible only to the person who started it — nobody else in the workspace, including admins, can open or write to it by default. Share it from the chat’s settings menu to bring others in, either with Everyone in the workspace or with Only selected people, picked from a member search. In a shared chat, every message shows who sent it — name and avatar next to their turns — so a multi-author conversation stays readable. Browsing your Recent chats still only shows your own conversations by default, whether or not you’ve shared any of them, so this doesn’t flood your list with everyone else’s threads. If you ever need help from Genie support with a conversation, they can access it to assist you, even if it hasn’t been shared. Delete a chat. Choose Delete chat in a chat’s settings menu and confirm. You can always delete a chat you started; deleting one a teammate started — and shared with you — needs the permission to delete chats, which owners and admins have by default. A teammate’s private chat can’t be reached at all. Deleting is permanent and removes the thread for everyone it was shared with, and an answer still being written in it is stopped. Drag any chat from the sidebar onto the Pinned dropzone at the top of the list to pin it — drop a pinned chat there again to unpin it — or use Pin chat from the same settings menu. Pinned chats lead the list under their own heading, ahead of the usual date groups, everywhere your chats are listed.

​

The command palette

Press ⌘K (Mac) or Ctrl+K (Windows) from anywhere in Genie to open the command palette — a fast way to jump to a project, run an action, or start asking Genie a question without leaving the page you’re on. Inside a workspace, the palette also searches your Library as you type. It lists matching entities from the knowledge model and up to five documents — each with the passage that matched — and Show all results in the Library opens the full search. Picking a document opens it right there, without leaving the page you’re on. Choosing Ask Genie switches the palette into a chat window in place. If you close the palette mid-conversation, nothing is lost — the conversation is saved and reappears under Recent chats at the bottom of the list. Reopening the palette shortly after chatting picks the conversation back up automatically, so you can dismiss it without losing your place.

​

Related pages