Changelog

Voice agent in meetings, sharing end to end

• NewVoice agent in your meetings. Tana's voice agent joins an internal meeting as a voice participant so you can talk things through hands-free. In a solo meeting it appears as a 1x1 voice control tile in the meeting grid. In a multi-participant meeting it lives in the meeting overflow menu as "Add voice agent" so it does not crowd the toolbar. External-meeting captures hide the voice tile so the external-capture widget owns the slot.
• NewWake words for the voice agent. In a solo meeting you can start the voice agent just by saying "Hey Tana" or "Hi Tana", with localised equivalents in Norwegian, Swedish, French, Spanish, German, Italian, Portuguese, and Dutch. Saying "Bye Tana" or "Goodbye Tana" stops it.
• NewThinking sound while the agent reasons. Between your input and the voice agent's first audio reply, a subtle progress loop plays so the gap does not feel dead. The sound is tied to the "Call Sounds" toggle in Settings, Advanced.
• NewAI proposes access changes on any document. The AI can now propose changes to a document's access: restrict it to specific participants with editor or admin roles, open it back up to everyone in its space or event, and toggle the Guest Link (anyone with the link can view). Access changes go through the same proposal approval flow as the rest of the AI's edits, and the proposal preview shows a clear access callout with visibility and per-participant role changes before you accept.
• NewAI can look up access and metadata. A new metadata tool lets the AI read current access mode, participants and roles, guest link state, owner and location, timestamps, and external IDs (Linear, GitHub, and so on) for any document the user can see, without loading the document body. Useful for questions like "who can see this doc?" or "which Linear issue is this linked to?".
• NewGuests see the live meeting digest. Guests joining a meeting via a link now see the live Meeting Digest in a middle column, mirroring the host's layout in read-only form. No pause control, no transcript dropdown, but every section the digest writes during and after the call is visible.
• NewGuests see pinned docs and approved outcomes. Guest meetings now have a surface column on the right that shows public documents pinned to the event and approved meeting outcomes (bugs, tasks, ideas, and any other type). Cards carry the same coloured type badge that the host sees, plus image thumbnails for artifacts. Clicking a card opens the doc in the public viewer. The column stays hidden until there is something to show, so the digest takes the full right rail in empty meetings.
• NewHand off context to a coding tool. A new "Open in coding tool" built-in skill packages the current document or chat context as a self-contained prompt and hands it off to Claude Code, Cursor, Codex, GitHub Copilot, Lovable, or v0. Triggered by phrasings like "create a PR", "open a pull request", or "send this to Cursor". When more than one tool is enabled it asks you to pick. When the repo or project is not named in the source material, it asks where the change should land instead of guessing.
• NewAI usage in your user settings. Settings, User now has a Usage section showing your AI spend against the monthly cap and the rolling 5-hour cap, as percentage progress bars with relative reset captions ("Resets in 3 days"). No dollar amounts, just percentages. Where usage caps are not enforced, a gray info callout above the bars notes that the bars are informational.
• ImprovedVoice agent auto-starts in the welcome meeting. When you finish onboarding and land in the welcome meeting, the voice agent starts itself so you can speak with Tana right away. Subsequent visits do not re-trigger.
• ImprovedGuests can see and hear the voice agent. When a Tana user activates the voice agent in a meeting, guests see the agent's tile and hear its audio, and they get the same auto-mute toast and manual-unmute override that other participants get. Guests cannot activate the agent themselves.
• ImprovedVoice agent participant tile polish. The voice agent's participant tile is a proper square that matches the rest of the grid, shows a "Tana" name overlay, dims slightly while the agent is thinking, and collapses to a small tile when a presenter takes the stage.
• ImprovedResizable guest meeting layout. The guest meeting view now uses a resizable three-column layout: meeting on the left, digest in the middle, surface on the right. Drag handles between the columns let guests adjust the layout in session. Sizes are not remembered across reloads.
• ImprovedWelcome meeting opens with talking points and a how-to doc. The onboarding welcome meeting now opens with a populated Talking Points doc and a How to use Tana doc already pinned to the event. You see how a real meeting surface looks before you start typing. The private AI chat in the meeting starts empty and lazy-creates on your first message.
• ImprovedUseful messages when an AI run fails. When an AI run fails, the chat shows a red callout with a one-line, actionable reason: provider rejection (e.g. "image exceeds 2000 pixels on the longest side"), idle timeout, too many tool steps, repeated tool errors, stalled response, or a generic "Something went wrong" fallback. "Cancelled" and "Usage limit reached" keep their pill style.
• ImprovedSkill chat tile reshapes with its lifecycle. Running a skill from the meeting Skills button now pins a single tile that reshapes as the skill runs: a compact streaming tile (skill name plus progress indicator) while the AI works, then an expandable group stack of the proposals the skill produced. Clicking the tile in either state opens the full conversation. Both states have a More menu and right-click context menu with Open chat and Unpin from event.
• ImprovedCapture tile shows what is being captured. When the meeting's context already knows the concept being captured, the in-progress capture tile uses its title (for example "Capturing iOS scroll bug...") instead of a generic "Capturing..." label. Manual captures from the floating popover keep the generic label.
• ImprovedRefined proposals update in place. When the AI refines an existing meeting proposal (e.g. you elaborate on a captured item and Tana updates the same item), the proposal tile re-renders in place with the new content. The refinement stays silent on the surface, so the proposal you were looking at keeps its position.
• ImprovedMove action across more panels and menus. The Move action (and a proper overflow menu with Copy link and Delete) is available on the meeting event overflow menu, the inline document popover, and the Image, Asset, Audio, Call, and Transcript panels. So you can change where a doc lives from any panel, including docs at the org root where the toolbar's Location chip does not apply.
• ImprovedReport a meeting from the overflow menu. The meeting and event overflow menu has a Report item that any user can use to submit a bad-recap report, the same way you can report a bad chat reply.
• ImprovedPublic doc viewer shows every assignee. Shared link pages render the full assignee list as an avatar stack, matching the in-app view. Embed cards inside the viewer also show all assignees.
• ImprovedBrowser tab titles for shared docs. Shared link doc pages set the browser tab title to the document's title once the page loads, so shared links are recognisable in tab bars and bookmarks.
• ImprovedClickable embeds in shared doc pages. Embeds inside the public document viewer are clickable. Left-click navigates to the embedded doc's public view; middle-click, Cmd-click on macOS, Ctrl-click on Windows and Linux, and right-click behave like any other link. Embeds for items without a public page render as plain item cards.
• ImprovedSign-in call to action on private shared links. Opening a shared link that has been turned private shows a Sign in call to action that returns you to the authenticated view, so the document is one step away if you have access to it.
• FixedRegion-scoped error boundaries in the sidebar. A single bad item in the Spaces list, Custom Agents list, or Activity link can no longer blank the entire app. The affected region greys out in place, while the rest of the app keeps working.
• FixedIteration chats stay with the doc. The wrapper chats created by "Open in Chat" on a doc are owned by the doc, so they do not show up in the main chats list alongside real conversations. Direct links to an iteration chat still work.
• FixedAI error messages do not over-promise reports. AI error messages no longer claim a failure has been reported when no report was filed. If you want to report a problem, use the Report action explicitly.
• FixedChat-attached images resized before sending. Images attached to chat messages are resized to a max 1568 pixels on the longest side before being sent to the model, so very large images no longer get rejected mid-run.
• FixedScreen capture survives brief GPU hiccups. Brief graphics hiccups during screen capture do not end an in-progress recording. The capture keeps going through the blip and resumes the next frame.
• FixedDesktop sign-in survives sleep and wake. On the desktop app, the sign-in session now pauses when the machine sleeps and resumes when it wakes. Long sleeps no longer leave the app stuck on a stale session.
• FixedMeeting capture start and stop race fixed. Quickly starting then stopping a meeting capture no longer leaves the recorder in a bad state. Start and stop signals now resolve in order.