Skip to main content

​
Contents

​
Redesigned app shell and dark mode

The page, the AI chat, and the item details panel are now separate rounded panels on a shared background, with the navigation sidebar sitting on that same background rather than behind a divider. Settings uses the same layout. Opening the details panel floats it over the page instead of narrowing what’s beneath it, so tables, trees, and the Reader keep their full width while it’s open. Drag handles now sit in the gap between panels, which makes them easier to grab, and the sidebar’s view picker spans the column so long view names truncate rather than resize the column. Dark mode has a cooler, clearer hierarchy: backgrounds, cards, popovers, panels, and text share a coordinated indigo cast, secondary controls stand above the page, and Tree matches Table’s background. Light mode picks up a slightly quieter neutral shell behind the sidebar and workspace panels.

​
Boltline BOM reads and work plan write-back

Ask the agent for the engineering BOM below a part revision, the as-built tree below a serial number, or the operations and steps of a work plan, and it reads them from Boltline one level at a time. Tree results say when a depth or size limit stopped the read, so the agent can narrow the question instead of treating a partial tree as complete. Turn on the new Boltline write-back toggle on the integration page (an API key with write permission is required) and the agent can create a work plan template from ordered operations and steps, or open a new draft version of a published template and add operations to it. New plans stay drafts unless you ask the agent to publish. Writes go to Boltline immediately and aren’t undone by discarding a branch; a write that fails part-way reports the operations it already created.

​
Email MFA for password sign-in

Organization administrators can now require an emailed verification code after email and password. SSO sign-in is unchanged. Email delivery must be configured before enabling the setting.

​
Viewers and external recipients can submit reviews

Read-only workspace members can approve or request changes and later update or clear their decision, while proposal state, engineering data, suggested edits, and merge actions remain read-only. A recipient of a share grant can also open a review for a branch covered by the grant, discuss it, reply, and approve or request changes. Changed items and diffs stay limited to the exact items the grant reaches, while ready state, edits, suggestions, and merge controls remain unavailable.

​
The agent knows which entity you were looking at

When you send a chat message with an entity open in the details panel, that entity travels with the message. Questions like “why is this stuck?” or “change this one” now resolve to the open entity instead of the agent asking which one you mean. The reference is recorded when you hit send and never changes afterwards, so navigating away doesn’t rewrite a message you already sent, and the agent sees nothing else about your screen.

​
Sharing a view from anywhere

The Share dialog now puts invitees first, lets you pick the exact item, branch, and saved view destination, and keeps branch reach and expiry tucked under one options row. Share is also available from table-row and details-panel item menus, alongside the existing header and navigator actions. The find box in a view now matches rows by any value global search would find: typed text fields, numbers, dates, URLs, and IDs (human ID, legacy ID, or entity id), not just name and description. It follows the same minimum-term rule and shows the same hint for a term it can’t serve. Results also start arriving sooner, and searching inside a focused folder no longer slows down with the size of the folder.

​
Improvements

  • Editor-group changes in branch activity. Assigning, changing, or clearing the user groups allowed to edit a branch now records an entry in the review page’s Activity feed, naming who made the change and which groups were added or removed. A new Editor groups column on the branch listing shows the current assignment beside Owner and Reviewers and can edit it in place — add the column from Display options → Choose columns…, then double-click a cell to tick the groups allowed to edit that branch.
  • Admins can add reviewers to any branch. Project admins can add reviewers to a branch someone else owns.
  • Linear teams show by name. The Linear integration page lists connected teams as “Name (KEY)” instead of ids, and the assistant sees team names on Linear projects, so it can create an issue in the team you name.
  • Toggles. A switch that’s off now sits on a solid gray track instead of a faint tint, so it reads as a control at a glance on any light background.
  • Welcome greeting on the agent page. The agent page now opens on a welcome greeting under the Flow mark, with the composer centered beneath it.
  • Copying an item’s ID from the details panel. The ID in the panel header is ordinary text you can select and copy by hand, and a dedicated copy button beside it puts the whole ID on your clipboard. Clicking the ID itself no longer copies it.
  • Older imported lists. Bulleted and numbered lists from older imports now render through the lightweight path in tables, trees, and the Reader.
  • Imported requirement documents. Descriptions imported from ReqIF now render through the lightweight path in tables, trees, and the Reader, so those documents open and scroll faster.
  • Export link columns. The Export dialog offers one link field per data model instead of separate entries for each direction. Choose Up+Downstream, Upstream, Downstream, or Cross-links from the field’s dropdown.
  • Orphaned items in search. Orphaned items are hidden from a view’s rows but remain findable through global search and Find in view.

​
Fixes

  • Reviewing someone else’s change. The merge action in the Changed Items status popover is no longer squeezed off the edge for reviewers who aren’t the branch author.
  • AI chat names. The name the assistant gives a new conversation stays put when the reply finishes, instead of reverting to “Chat” until the page is reloaded. A chat named from a short opening message is now given a real name instead of a paragraph explaining there was nothing to summarize.
  • Assistant tool reuse. A tool the assistant already used or looked up earlier in a chat, including one prepared before you accepted its plan, now works the first time on later turns instead of failing and retrying.
  • Assistant entity counts. Asking the assistant to count items now shows the per-type and per-value breakdown with type names again, instead of a blank or missing card.
  • Stopping a document read. Stopping the assistant while it reads a document now stops the read itself, instead of leaving it running through every remaining page. Pages already read are kept.
  • AI conversation summaries. Long conversations retain more detail when compacted. Incomplete summaries continue automatically, and existing context is preserved if summarization can’t finish.
  • Side panels in a narrow window. Shrinking the browser window past the point where the AI panel or the agent page’s right-hand panel meets the sidebar now narrows that panel to its minimum width instead of pushing its controls off screen. The width you set is kept, so widening the window restores it.
  • Details panel tags. Multi-value tag fields now wrap onto additional rows without squeezing or breaking the text inside each tag.
  • CSV and JSON export requests. Invalid column or filter plans now return a clear request error before a download or job starts.