Comment on Artifacts, Workspace Stories & a Redesigned Copera

You can now click any element inside an Artifact and leave a comment right there, Stories bring short workspace updates to your sidebar, and Copera gets a full redesign — a new Midnight default theme, per-area accent colours, and rebuilt AI Workforce, Automations and Ask Copera screens.

New Features

Artifacts — Point at Anything and Comment on It

Artifacts get the review flow they were missing, and the frame around them gets out of the way (#3497, #3504):

• Click to comment — arm the pointer in the toolbar, hover to highlight, and click any element inside the Artifact. The comment box opens right where you clicked, with the element already named.
• Pins that live in the page — your comment lands as an initials chip pinned to the element itself, so it scrolls with the content instead of floating over it. A pin whose element has been rewritten away is quietly skipped rather than left pointing at nothing.
• Threads at the pin — click a pin to read the thread, resolve it, or (if you wrote it) edit and delete it. Anyone who can see a remark can mark it dealt with; only the author can change or remove their own words.
• "Ask Copera to fix N comments" — a bubble in the corner counts your open remarks and opens the list, with one button that hands them all to Copera as a real message. Resolved remarks are left out.
• Just the Artifact — the header bar and card border are gone. The Artifact now runs edge to edge with a single quiet line underneath: "Copera artifact · Version 4".

Artifacts — Faster to Build, Easier to Live With

• See it while it builds — an Artifact used to take 22–27 seconds to reach you. It now appears in about three seconds and finishes assembling in the background (#3484).
• Edits revise, instead of quietly making a second one — asking for a change to an Artifact now updates that Artifact and adds a version, rather than creating a duplicate and telling you it updated the original (#3480).
• Simple pages are valid pages — "give me a simple landing page saying Jr Rocks" used to be refused for having too few words. A short, deliberate page is a design decision and is now accepted (#3485, #3489).
• Just write it and Copera works out the rest — you no longer have to spell out how an Artifact should be packaged; Copera derives that from the files themselves, which removes the single largest cause of failed Artifact creation (#3480).
• Room to actually look at it — the chat panel now opens from the toolbar and starts closed, so the Artifact gets the full width, and the desktop / tablet / phone size buttons finally have the space to work (#3504).
• A "Switch artifact" menu — replaces the old tab strip, listing your Artifacts with "All artifacts" at the bottom. Duplicate and Delete move into a "⋯" menu, out of one stray click's reach.
• Download files — save the working files of an Artifact to your machine, including edits not yet frozen into a version.
• The version is a handle — the version number is now a pill that opens the history, rather than a banner announced above your work.

Stories — Short Workspace Updates, in the Sidebar

A new way to share a quick update with your workspace (#3462, #3465):

• Rings in the left menu — unseen stories carry a coloured ring, seen ones go grey, and you can dismiss one on hover.
• A fullscreen viewer — tap either side to move between stories, hold to pause, and hold a video to watch it at double speed. There's a pause/play control at the top, and progress is remembered so a story you have watched doesn't come back as new.
• Create from a Spark or an upload — turn a recording you already made into a story, or upload media directly.
• Choose who sees it and for how long — pick the audience and the expiry when you post. People who joined the workspace after a story was published don't see it.
• See how it landed — a management page in settings lists your stories with view counts and per-person analytics.

A Redesigned Copera — Midnight, Area Accents & Rebuilt AI Screens

The largest visual change in months (#3477, #3472):

• Midnight is the new default — a deeper, calmer canvas, with the top bar, sidebars and right rail sharing one surface so the app reads as a single frame. The old "Dark" theme is now called Navy; if you had picked a theme yourself, you keep it.
• Area colours that actually reach everything — picking an accent for an area used to repaint the top bar and nothing else. Buttons, checkboxes, switches, tabs and selections now all follow it, and Ask Copera gets its own copper accent.
• A quieter top bar — the bar is a neutral surface and only the area you're in carries the colour. The fully-coloured bar is still available in the Customize drawer.
• Small print you can read — captions, table headers and chart labels were too dim to meet accessibility contrast requirements in every theme. They are now compliant across all four.
• Tabs that aren't glued together — 23 of the app's tab rows had no spacing at all and rendered as one run-on string. Fixed everywhere at once.

AI Workforce — A Real Roster, and Hiring That Tells You Things

• A header that says how you're doing — hired, working and "needs you" counted at the top, with filter chips (All / Working / Needs attention / Paused) that genuinely filter (#3477).
• Three tabs instead of one long scroll — Roster, Agent store, and Built by you, which is now one screen for the whole authoring lifecycle: still building, published, in review, why a submission was rejected, install counts and ratings. You can hire your own creations from there.
• Hiring is a decision again — Browse → Connect → Done, with the disclosure on the first step: what the agent does, what it will ask of you, what it can reach, what other hirers say, and what it costs. Nothing is created until you press the button.
• Say what an agent needs in words — the "what it needs" block used to print raw internal identifiers. It now shows the creator's own sentence for each requirement, and groups an agent's reach by category and strength rather than listing functions.
• Store cards show adoption — how many other workspaces run an agent, plus its rating, right on the card.
• Grant capabilities and access — admins can now grant an agent specific tool capabilities, and "Add access" on an agent profile finally opens a grant dialog instead of the wrong screen. Automation-authoring capabilities stay ungrantable by design.
• Scheduled jobs that say what happened — a job that didn't run used to report an internal status code in green. Every outcome now has plain wording and an honest colour, plus when it next runs.

Automations — Cards, Filters and a Usage Screen That Adds Up

• A header with the state of the workspace — "N automations · N enabled · N failing" — and the ability to create an automation from this page, not just browse templates (#3477).
• Cards instead of a data table — health, name, what it does and the on/off switch together, with run history as a quiet strip beneath.
• Filters that mean something — All, Enabled, Draft, Failing, and Owned by an agent, each with a real count. "Failing" means broken right now, not "has failed at some point".
• It says where a rule lives — a card used to say "Board"; it now names which board.
• A rebuilt Usage tab — the period and how much of the workspace ran, four totals with what they're made of and how they moved, spend and runs by day, where the spend goes, and the runs that didn't finish grouped by reason. Exportable to CSV for whatever range you're looking at.

Ask Copera — Settings as a Rail, and a Rebuilt Setup Page

• Settings is a section rail — one section at a time beside its own navigation, instead of everything stacked in one scroll behind hash links (#3477).
• Prompts and scheduled calls moved in — both were top-level destinations you only ever visited to set something up. Old links still work.
• Current setup, rewritten — memory, guidance, skills and connectors each get their own card and their own live state, plus a ranked answer to the question people actually have: what wins when instructions disagree.
• Action review is a real control — the highlighted row is the mode new chats actually start in, and choosing one saves it.
• Memory as rows — your remembered preferences are a list you can edit in place, not a screen of labelled form fields.
• Skills as collapsed rows — name, on/off, and what it loads on, with editing in place. The skill library becomes its own full screen with a category rail, search and sort, instead of a dialog paging fifty cards at a time.
• Connectors listed with their status — what Ask Copera can currently reach, how many tools each exposes, and a warning on one whose last connection failed.
• A better empty state — filled suggestion cards, and a row offering the agents your workspace has actually been talking to.

Agentic Chat — Queue, Canvas & Honest Failure States

• Queue your next message — send while a run is still going and it waits its turn, showing its position, with the option to take it back out (#3498).
• Canvas — HTML and SVG the assistant produces render inline, with a preview/code toggle.
• States that tell the truth — "Stopped", "Not approved" and "Interrupted before it finished" each explain themselves, instead of everything unknown being reported as a failure (#3495).
• The clock keeps time — the elapsed timer keeps ticking across a reconnect, and pauses while a run is waiting on you.
• Branch switching without losing your place — the full message tree is held, so switching branches doesn't tear the conversation down and rebuild it.
• Your permission mode sticks per conversation — reopening a chat restores the mode that chat was actually running in, rather than silently reverting to "ask before acting" (#3464).

Sparks Editor — Presentation, Inserts and Cutting by Transcript

• Presentation settings — choose an aspect ratio (16:9, 9:16, square or source), a background (Botanical Shadow, Midnight Studio or Soft Paper) and how much padding sits around your recording (#3493).
• Auto Zoom — Copera zooms toward your clicks automatically on recordings that captured them, and tells you plainly when a recording has no click data to work from.
• Insert another Spark — drop an existing recording into your timeline at a chosen moment, with a preview before you commit.
• Cut by transcript — click a word or drag across a phrase to cut it. Struck-through text is skipped in the preview and removed on export, and a cut line can be restored.
• Mark start and end — set your trim points directly from the transport controls (#3486).
• Smart edit that reports back — removing silences and filler words now tells you how many ranges went and how much time you saved, and says "nothing new to remove" when there's nothing left.
• Alerts that offer a way out — every editor failure now says what happened and gives you a retry, and your local draft is kept rather than discarded.

Sparks AI — Manage Your Languages

• A language list for subtitles — add a language, watch its progress, then select it for both the transcript and the closed captions (#3486).
• A language list for dubs — add audio for the original language or any ready subtitle language, then pick it to listen in the player.
• Transcript language — the original transcript's language is now stated, so you always know what you're translating from.

Inbox — Counts on Your Folders

• Unread and draft counts — folder tabs now carry the number of unread messages, and Drafts shows its total (#3475).

Developer — Tempo, Agenda and Inbox in the Public API

• New surfaces — calendar events, Tempo calendars and shelf items, mailbox messages, and booking-host management are all available through the Public API, each behind its own dedicated permission (#3501).
• Tools for your assistants — the Tempo, Agenda and booking-host surfaces are also available as connected tools.

Bug Fixes

• Personal token permissions — the edit dialog was missing six permissions the create dialog offered, so editing a token could silently narrow it (#3502).
• Board owners losing admin rights — a board owner assigned to a team with a weaker role was demoted by it. Owners now stay administrators, and a stronger permission granted to you directly outranks a weaker one inherited from a team.
• Workspace deletion — a deletion that got stuck can now be resumed, a workspace being deleted disappears from your list immediately, and it can no longer be signed into while it's on the way out (#3473, #3476).
• Artifact preview cut off — the preview was clipped at a fixed height with the rest of the screen left black, and it restarted from scratch every time you switched back to the window (#3478).
• AI Settings pushed off the sidebar — turning Virtual Office on pushed the Settings row past the bottom of the menu with no way to reach it (#3479).
• Replies in the wrong language — an English request from an English account could come back entirely in Portuguese. Copera no longer guesses at your language where you typed your message; it simply answers in the language you wrote in (#3485, #3489).
• Turns that did the work and then failed — an Ask Copera turn that streamed a complete answer and made real changes could still die at the last step with an unexplainable error (#3464).
• Rows created from templates — cells belonging to deleted or changed columns are now skipped rather than failing the whole row, and linked-record limits are checked before the row is created (#3474).
• Long messages in the composer — the box now grows properly before it starts scrolling internally (#3474).
• Voice invites — you no longer get invited to a channel you're already in, and a call that failed to connect can be rung again (#3474).
• Board columns on older boards — table columns saved before some options existed no longer lose their colour or order (#3483).
• Editor stability — fixed crashes around slash commands, stale block menus, emoji lookups and task lists (#3481).
• Voice note cleanup — a recording that timed out while wrapping up no longer leaves the note stuck (#3481).
• Mailbox connection — a dropped connection now keeps retrying with a growing delay instead of giving up (#3481).
• Local storage blocked — the app no longer reports an error when your browser refuses it local storage (#3481).
• Document AI reading — asking Copera to read a document it can't reach is now a plain "couldn't read that" rather than a hard failure.

Improvements

• Drive opens instantly — folders are fetched before you click into them, and the previous listing dims rather than flashing empty while the new one arrives (#3471).
• Calmer chat lists — a single channel event no longer re-renders every conversation in your sidebar, timers tick more smoothly, and your agenda stops refetching itself every few seconds (#3471).
• Switching areas is instant — changing sections no longer rebuilds the whole theme behind the scenes (#3472).
• Fewer wasted attempts when creating an Artifact — Copera now accepts the shapes models actually send instead of rejecting correct work over its packaging, and when something genuinely can't be built it says what to change (#3480, #3467).
• A quieter, thinner AI sidebar — Settings stays pinned while everything above it scrolls, section labels take a distinct label face, and the Automations, Bankai history, AI Workforce and Artifacts icons are all redrawn.