Skip to main content

Command Palette Rewrite, Workspace Search & Deep Research Plan Cards

· 9 min read
Copera Team
Copera Team
Product Team

This week ships a complete rewrite of the Command Palette with 91 commands now accessible to both keyboard users and Bankai voice, plus a new workspace-wide search that finds documents, boards, transcriptions, meetings and people in one place. Deep Research gets a plan-and-confirm card with depth tiers and live stats on completion. Virtual Office gains real guest admission with full privacy isolation, and Bankai rolls out per-user vocabularies plus onboarding polish.

New Features​

Command Palette — Complete Rewrite with 91 Commands​

The Command Palette has been rebuilt from the ground up as the universal action surface for both keyboard users and Bankai voice (#2899):

  • 91 commands across four categories — navigation (open pages, join channels), creation (new docs, tasks, rows, messages), actions (send messages, schedule, share), and utility (toggle theme, zen mode, sidebar)
  • Workspace-wide search — typing instantly searches across documents, boards, rows, transcriptions, meetings, people, and direct messages with smart ranking
  • Single-character shortcuts — prefix a query with #, @, doc:, mail:, board:, row:, file:, or meeting: to narrow your search to a specific scope
  • Send via the palette — type > send @user message or > send #channel message to dispatch a chat or DM without opening the channel
  • Create via the palette — type > create row [board] or > new doc to kick off creation flows directly
  • Join via the palette — type > join [meeting channel] to enter a huddle with one command
  • 10-second undo — after any send action, a toast lets you undo within 10 seconds before the message is actually delivered
  • Bankai Command mode — every palette command is also available by voice when you activate Command mode

Workspace Search — Find Anything in Copera​

A powerful new search API and index behind the Command Palette that covers your entire workspace (#2890):

  • Search everything — documents, boards, rows, transcriptions, meetings, meeting channels, direct messages and people are now indexed and searchable from one place
  • Strict permissions — results are filtered to only show content you have access to, per channel, board, and workspace
  • Public search API — developers can now query workspace search programmatically via /public/v1/search with a Personal Access Token
  • Workspace info endpoints — the public API also exposes workspace info, members, and teams for integrations

Deep Research — Plan Cards, Depth Tiers & Completion Stats​

Three major improvements land this week to bring deep research to parity with leading AI research tools (#2875, #2876, #2878, #2880, #2882, #2884):

  • Plan-and-confirm card — when the AI decides a query needs deep research, it now shows an editable plan with an AI-generated title, sub-questions you can add/remove, a depth selector, and a 60-second auto-start countdown you can pause while editing
  • Three depth tiers — Standard (5 sub-questions, 2 iterations), Deep (8 sub-questions, 3 iterations, smart chunk filtering), and Very Deep (12 sub-questions, 4 iterations, source curation, recursive follow-ups)
  • Completion notifications — get a bell, desktop, and mobile notification when your research finishes so you can switch tabs while it runs, matching ChatGPT's Deep Research UX
  • Click-to-deep-link — clicking a notification jumps you straight to the chat and scrolls to the specific result
  • Auto-dismiss on view — if you open the chat directly instead of via the notification, any matching unread alerts clear silently
  • Completion stats — completed research now shows "Research completed in 2m 15s · 22 citations · 333 searches" so you can see at a glance how deep the investigation went
  • Persistent elapsed timer — the running timer survives page navigation so you don't see it reset when you switch views
  • Research chip protection — the Research button is now disabled while a research is already running so you can't accidentally start multiple at once
  • Failed research status — if a research is cancelled or errors out, the card now clearly says "Research failed" instead of staying stuck on "Research started"
  • Inline Very Deep warning — selecting Very Deep now shows the warning right on the card instead of a second confirmation dialog, so you can proceed with a single click

Virtual Office — Real Guest Admission & Privacy Isolation​

The Virtual Office guest flow has been hardened with real access control and full privacy isolation between rooms (#2889, #2891):

  • True audio/video isolation between rooms — when you walk from one virtual room to another, people in the old room no longer hear you, and you no longer hear them (previously audio was only muted locally)
  • Screen share isolation — screen shares are now strictly scoped to the room they started in, so they never leak to other rooms or guests
  • Guests are truly isolated while waiting — guests knocking on a locked room can no longer hear the conversation inside until a member admits them (previously they were connected with full permissions during the waiting period)
  • Position persistence — your last position in the Virtual Office is remembered across refreshes, so you respawn exactly where you were
  • Faster joiner hydration — new joiners now see existing members in their correct rooms on the first frame, with no flash to the lobby center
  • Invite guests from huddle view — a new "Invite guest" button in the huddle header lets you copy the invite link without leaving the huddle
  • Smart guest spawn — guests joining via the same invite link now spawn at slightly different positions inside the room instead of stacking on top of each other
  • Map-wide guest badges — the waiting-guest indicator now shows on any room that has pending knocks, not just the one you're standing in, so members can admit guests from anywhere

Bankai — Per-User Vocabulary, Onboarding & Analytics​

Bankai gets a polished onboarding wizard, personal vocabularies, and rich usage analytics (#2865, #2872, #2888, #2892, #2893):

  • Per-user custom vocabulary — your dictation dictionary is now personal to you rather than shared across the workspace, so you can teach Bankai your names, jargon, and acronyms without polluting your teammates' dictation
  • Usage analytics — a new metrics card on the Voice History page shows your words spoken, time saved, session count, streak, and per-mode breakdown
  • Interactive training wizard — a fully redesigned onboarding tutorial measures your typing speed against your voice speed, then walks you through each mode hands-on with clearer step-by-step guidance
  • Milestone celebrations — toast notifications celebrate when you hit 1K, 10K, or 50K words, your first hour saved, or a 7-day streak
  • Mode discovery nudges — after 20+ sessions, weekly nudges surface modes you haven't tried yet
  • Platform-aware shortcuts — the wizard now shows the correct key combo for your operating system (Fn on Mac, Ctrl+Win on Windows) throughout every step
  • Release gesture activation — the wizard's final "Enable Bankai" now activates on the voice-release gesture itself instead of a button click, making activation feel thematic
  • Language preservation — Portuguese dictations now stay in Portuguese when your workspace language is set to "auto", fixing a bug where they could be reformatted into English
  • Smarter TTS voice selection — Chat mode now picks the right voice based on what the AI actually said, so English "sim card" no longer gets read with a Portuguese accent
  • Failed-recording recovery — recordings that failed to process on the server now still show up in your voice history with play, download, delete, and retry actions so nothing is ever silently lost
  • Desktop microphone picker — choose your active microphone directly from the desktop settings
  • Mute toggles — new "Mute system audio during recording" and "Mute dictation sounds" toggles in desktop settings
  • Trigger word fix — Omni mode now properly requires the "Bankai" trigger word before firing commands, preventing accidental triggers during dictation

CLI OAuth & Public API — Programmatic Access​

New entry points for developers and power users (#2894):

  • CLI OAuth flow — a new /oauth/cli page walks you through creating a Personal Access Token for a CLI tool, with workspace selection and copy-paste token display
  • Board API description field — the public board row endpoint now returns the row's rich text content as a description field for easier integration

Bug Fixes​

  • Inbox — the search bar no longer drops keystrokes as you type; your full query is preserved character by character (#2874)
  • Inbox — the notification bell count now matches the list of notifications you can actually see, instead of showing more than it can display (#2874)
  • Inbox — meeting bot transcription shares now actually notify the recipient, with a bell notification and click-through to the transcription (#2874)
  • Inbox — the blue "needs reply" dismiss button now appears on every thread that has the indicator, not only on the filtered need-reply view, and the tooltip now clearly says "Mark as no reply needed" so you know you're dismissing the marker, not the email (#2871)
  • Board — fixed default-value options on non-workflow columns so they behave correctly (#2877)
  • Bankai — "Copy to Clipboard" toggle restored in desktop settings after a naming mismatch that broke it (#2870)
  • Bankai — the desktop settings page now shows a clear "No microphones detected" error in red when no microphone is connected (#2895)
  • Desktop — the desktop app's language now stays in sync with the language you set in the web UI, both on mount and whenever you change it (#2896)
  • Email — embedded images no longer carry a synthetic identifier that could confuse some email clients

Improvements​

  • Speech-to-Text — every microphone across Copera (Ask Copera AI, Command Palette, email composer, task comments, chat inputs, board AI, formula assistant, and more) now runs through a single unified pipeline with automatic fallback, giving you consistent behavior and more reliable recognition across every surface (#2879)
  • Dictation everywhere — microphones have been added to 11+ new surfaces including the AI Formula Assistant, RRule Generator, Support Chat, Prompt Library, Workspace Invite messages, Approval comments, Email composer body, Transcription AI chat, task comments, doc comments, and direct messages (#2879)
  • Deep Research zero-stats handling — older research runs that don't have valid stats now fall back to a plain "Research complete" line instead of showing "0 citations · 0 searches" (#2884)
  • Deep Research accurate citation counts — the citation count now reflects how many sources the final report actually cites, not how many were considered, so the number matches what you see in the sources panel (#2884)
  • Voice-to-text reliability — the voice transcription system now includes more accurate internal metrics for every microphone in Copera, enabling better monitoring and more reliable recognition (#2879)
  • SEO — several search indexing improvements across the app, with proper canonical links on login and signup pages and correct robots rules to keep private pages out of search results (#2885)