Updates · What's new
The changes worth knowing about - what each one does for you, and how to use it.
An Interactive HTML block or a 3D model can now have its own link. People who open it see that visualization. The rest of the page stays where it is. You choose a public link or keep it to your organization.
How to use it
Media studio still applies your brand style when you generate a video. For one video you can turn Brand style off. The next video starts with your saved brand on again.
How to use it
Home now opens like a folder: the pages, databases and boards you file under it are listed right under the Home row, in the order you gave them. The rest of your pages stay below, so you always see where each one lives.
How to use it
Every menu now shares one calm look, and on a computer or tablet the ones you open from the sidebar appear beside it instead of covering the pages they act on. Menus work from the keyboard too: Esc closes them and puts you back where you were.
How to use it
The sidebar is calmer: Home is now a row at the top, Reports, Content and Knowledge Base sit above everything else, and the other tools wait under More. You choose what stays on top, so the tools you use every day are one click away and the rest stay out of your way.
How to use it
Ask the chat to give an existing task to one of your connected AIs or to a teammate, and it makes a real assignment, the same one you make by choosing an assignee on the board. Writing one name in a task's Assignee column does the same, and the task's details show who requested it.
How to use it
Your connected AIs, such as Claude, ChatGPT, Codex and Hermes, can now @mention each other in comments and hand tasks to each other, or to themselves, in stages. Each one picks up its part from its own inbox, and a person still accepts the last stage, so nothing is finished without you.
How to use it
Every connected AI now has a page that answers the questions people ask: what it is called, whether it takes mentions and tasks, who can use it, and how it checks its inbox. It also suggests prompts to try, written in your language, starting with one that sets ActVox up with you.
How to use it
Tell the chat “Remember that…” and it now keeps it: what you asked it to remember shows in Memory, where you can forget it again. A web page can go straight into your Knowledge Base too, so answers and drafts can draw on it.
How to use it
Comments on selected text now sit in the margin next to the words they are about, so a page and its discussion read together. React to any comment with an emoji, and resolve a thread once it is settled: it moves behind “n more comments” at the end of the page.
How to use it
Hover any block on a page, such as a table or an image, and comment on it as a whole, with no text to select. A page's comments now fold into one row that shows who took part, how many comments there are and who wrote last, and a long thread shows its first comment and newest reply.
How to use it
On a phone, nothing on a page pushes it sideways any more: long words, links, headings and titles wrap, on pages shared with you too. Long lines of code wrap unless you choose to scroll them, and a table wider than the screen opens full screen so you can read all of it.
How to use it
Every AI app and agent key you connected now appears under Your agents in Connectors, with what is active and when each one last called ActVox, agents that work through the ActVox CLI included. The agent cards say Connected, so you can tell at a glance which apps already work for you.
How to use it
Capture now records voice notes and writes out what you said, keeping the recording with the note. With no connection, your note stays on your device, voice notes included, and becomes a page as soon as you are back online, so nothing waits for a signal.
How to use it
The end of a page now lists the pages that link to it under Linked from, so you can follow a topic back to where it came up. Pages the ActVox agent writes also list their Sources: knowledge-base files open beside the page, and web pages open in a new browser tab, named by their site.
How to use it
The new bell in the top bar collects your mentions, replies, assigned tasks, review requests and approvals under For you, grouped by day and written in your own language. The personal ones also pop up for a few seconds, so you no longer have to check every page to find what is waiting for you.
How to use it
Anyone who can edit a workspace can now connect Claude, ChatGPT, Grok or another agent for themselves. ActVox asks which agent it is, so it gets its own name and @handle, stays private to you until an admin shares it, and never does more than your role allows.
How to use it
Every change to a page's name, icon, cover or place, and to tables, rows and whiteboards, is now recorded with who made it, whether a teammate, an outside agent or ActVox's own agent, and shown as before and after. If something went wrong, restore an earlier version; the restore is itself a change you can undo. Emptying the trash now waits 30 days, so a slip there can be undone too.
How to use it
A motion block can now star up to two cartoon characters who walk, wave, jump and talk in speech bubbles. A coach can explain the workout, a doctor the next step, or a mascot your news, and the character who isn't speaking turns to listen.
How to use it
A page's menu in the sidebar now creates a whiteboard nested under that page, next to a new database. The board sits with the notes it belongs to instead of loose at the top of the workspace.
How to use it
A whiteboard now has the tools of a real wall: a marker, a highlighter and washi tape under the Pen, stamps for a quick thumbs up or your initials, and emoji stickers. Sticky notes come in ten colours and can show who wrote them, so a retro or a review reads at a glance.
How to use it
Every callout used to wear the same 💡 on the same grey. Now you can pick any emoji or upload an image for its icon and tint it in one of nine colours, and shared pages and exports keep the look. In any text, a colon and a couple of letters bring up matching emoji.
How to use it
When you picked an agent in the Assignee column, it got only the row's title and had to go looking for the rest. Now the text on the row's page comes with the task as its brief, and a connected agent checks all your workspaces in one step, without errors for the ones where it can't take tasks.
How to use it
A page can now play a short animated piece: a title, steps that appear one by one, a big number, a quote, a countdown or a breathing circle. It uses your brand colours and plays on shared pages too, so a client gets something to watch instead of a wall of text.
How to use it
Your connected agent can now put a motion block on a page and change it later, so you can describe what you want in a sentence. It writes the scenes, ActVox draws them in your brand colours, and you can still adjust any scene yourself.
How to use it
Your agent can now put an interactive page of up to 16 MB into a block, so a detailed 3D scene, map or dashboard goes in as it is. Before, anything over a few megabytes had to be shrunk or split first.
How to use it
A connected agent only sees what you assigned it when it checks, and nothing in ActVox can wake it. Its connector page now gives you a ready prompt to schedule in that agent, so it works through its mentions and tasks without being reminded.
How to use it
Memory now opens in its own window instead of sharing one with the Knowledge Base, so you can keep both open. Both moved up the sidebar next to your files, and you can hide either one.
How to use it
A page no longer has to choose between comfortable lines of text and room for a big chart, 3D model or table. One block can now spread wider than the text column while the paragraphs around it keep their reading width, on shared pages and in exports too.
How to use it
A page you share to the web now tells you how many times it was opened, and by how many people, over the last 30 days. When an AI agent built the page, readers see a short line saying so under the title, and you can hide it.
How to use it
A table of contents, a formula or a whiteboard used to reach readers of a shared page as a grey box. They now appear as they do in your workspace, so long technical documents read properly outside ActVox and in PDF or PNG exports.
How to use it
Going back to Preview after editing the text in an HTML block used to keep the editing view, so slides and charts looked wrong until you reloaded. Preview now shows the block exactly as it renders, on the same slide you were on.
How to use it
Tables can now keep columns in view while you scroll sideways, total a column in a footer row, and open a row beside the table instead of on top of it. Select several rows to change them all at once.
How to use it
On a phone you can now comment on selected text without leaving the reader, settle an agent's suggested changes, and move a page's blocks up or down. Each action appears only when you can edit the page.
How to use it
Any page can now hold a grid of tiles, so text, images and HTML blocks sit next to each other instead of in one long column. On phones the tiles stack into one column, or two if you prefer.
How to use it
Clearing the trash used to mean pressing “Delete forever” on every item, one at a time. Now one button removes everything in it for good and shows how many items it has deleted while it works. System pages your workspace depends on are kept.
How to use it
When you connect Claude, ChatGPT or another agent to ActVox, the first thing you paste is now a short check. The agent tells you which workspace it is in and what it is allowed to do, and changes nothing. The follow-up prompt has it read what is already there, suggest one small first step and wait for your yes.
How to use it
When the agent linked to a page in its answer, the link opened a new browser tab. Now the page slides up over the chat, already scrolled to the part the link points at, and Back puts you straight back in the conversation. A link to an edit the agent suggested opens the page right on that edit, ready to accept or reject.
How to use it
Images, videos and PDFs in your library now open in a full-screen viewer where you can zoom and move through them with the arrow keys, and HTML artifacts open full screen too. The library also stopped quietly cutting off at a few hundred items — it keeps loading, and search looks through everything.
How to use it
The Share dialog used to say an invitation was sent the moment you clicked, before any email had gone out. Now it tells you the invitation was created and shows the email's real status next to each pending person, so you know whether to wait, resend, or write to them another way. Accepting an invitation also works for people who are new to ActVox, and a link opened from the wrong email address explains that instead of failing.
How to use it
Floor plans and diagrams that agents draw as SVG kept their labels locked: Edit changed the HTML text around them but skipped everything inside the drawing. Now every label in the drawing is a target — a room name, a dimension, a caption — and Save writes it back. Text a script draws keeps your change after the script redraws it.
How to use it
An agent can now build a chart, dashboard or 3D view on a page that reads a workspace database directly and redraws when a row changes. It can use d3, ECharts and three.js, so these are real charts, not hand-drawn shapes. Only people in your workspace see the live rows; share links and exports show the saved data.
How to use it
Agents that build HTML blocks used to guess how they would look inside ActVox. Now an agent can render a block with ActVox's own settings and get back a picture, the block's size and any errors the widget logged, then fix them before you open the page.
How to use it
When you point at a page in the sidebar or at one of your open tabs, ActVox starts loading it, so it is ready by the time you click. A page you haven't opened yet also opens in about half the time it used to.
How to use it
On a phone, a page now opens as a calm Reader: scrolling never drops a cursor or nudges a block. Editing is one tap away when you need it, and ActVox remembers where you stopped on each page.
How to use it
Connector setup pages now copy the way developer docs do. One button copies the entire setup as Markdown, ready to paste into an AI tool or a note, and every command, prompt and config snippet has its own copy button, so you never select text by hand.
How to use it
Any agent or script that can run terminal commands can now work with your workspace through the ActVox CLI. You sign in once in the browser, and a copyable instruction tells the agent how to check the connection and what it is allowed to do.
How to use it
Pages that agents created without naming a place used to pile up at the top of your sidebar. Now you pick one page to collect them under, for the built-in assistant and your connected agents alike.
How to use it
An HTML block no longer has to sit inside a card. Show it without a frame and it takes exactly the height of its content, or turn it into its own page that fills the whole pane — made for a dashboard, a one-pager or a report you built in HTML.
How to use it
Text now sits in a centred column about 75 characters wide at a larger size, and chat replies read the same way. In dark mode, assistant replies, callouts and code blocks are readable again, and the sidebar and tabs step back so the page you are reading stands out.
How to use it
Every code block now has its own copy button — in chat replies, file and Markdown previews, report blog posts, shared pages and Help Center articles. A command that spans several lines copies with its line breaks intact, so it runs exactly as written.
How to use it
Every comment now starts a thread: answer right under it, and resolve the discussion once it is settled — it folds away with its highlight and comes back if anyone replies. Connected agents can join in too: they read threads, reply and mention you, and their comments carry an Agent badge.
How to use it
Claude Code, Codex, Hermes and OpenClaw used to keep running an old ActVox skill or CLI until someone noticed. Now they learn when a newer version is out, ask you once, and upgrade with a single fixed command, so new features reach your agents the day they ship.
How to use it
Tell the workspace what is true — someone's role, a product's price, a decision you made — and it keeps each fact with its subject, its source and how sure it is. Turn on recall and the assistant brings the right facts into every chat, so you stop repeating yourself.
How to use it
Signing in the ActVox CLI used to tie it to a single workspace, so every other workspace needed its own profile. Now you tick several workspaces on the same approval page, switch between them with one command, and anything you did not approve is refused before it is sent.
How to use it
Rewording a heading or clearing a paragraph in Edit mode used to look right on screen while the HTML kept the old words, so Source, shared pages and exports disagreed with what you saw. Your edits now go straight into the HTML, and every view shows the same text.
How to use it
With no tab open, your workspace now opens to its Home page instead of a grid of cards. Ask for anything in the box at the top, run one of your skills, pick up recent work or start a new page — without hunting through the sidebar first.
How to use it
Switching a board to a table, grouping, filtering, sorting or hiding columns used to change the view for everyone who had it open. Those choices are now yours alone — and when your arrangement is the one the team should use, you can make it everyone's default.
How to use it
A public link used to show only the page itself: the pages inside it were out of reach, and embedded databases showed a placeholder. Visitors can now open every sub-page from a list, find their way back with breadcrumbs, and read the rows of embedded databases.
How to use it
One connection used to mean one workspace, so working across your personal space and a team's meant connecting ChatGPT, Claude or Grok again for each. The consent page now lets you tick every workspace the assistant may use, across all your organizations; the first one you tick answers requests that don't name a workspace, and each stays separately revocable in Settings.
How to use it
Connecting ChatGPT, Claude or Grok used to give it read access only, and writing meant a second sign-in from the assistant. The consent page now shows every permission at the first connection: read is always on, and one tick grants write, run and share.
How to use it
If you administer an organization, its workspaces were missing from the consent page unless that organization happened to be active in your session, so an assistant could only reach your personal workspace. The page now lists every workspace you administer, with the organization named under each one.
How to use it
Any page in your workspace can now leave as a file. PDF and PNG are rendered by ActVox with your fonts, images and embedded databases, and interactive blocks print as you see them. Word, Markdown and plain text are made right in your browser.
How to use it
The first row of the sidebar is now a New menu. It creates a chat, a page, a database or a whiteboard from one place, so you no longer hunt for the right section first.
How to use it
The dark palette is recalibrated: tabs, buttons, chat bubbles and hover states keep their contrast, and whiteboards follow the theme instead of opening white. Light mode also gets a softer reading weight for text.
How to use it
Pages, databases, and whiteboards now have a Copy button in their header. One click puts Markdown on your clipboard, ready to paste into a doc, a ticket, or a chat with an AI assistant, and its menu offers plain text too. Code blocks get their own Copy code button.
How to use it
Since late July, a report's Instagram, X, and LinkedIn plans often came back empty, and a report missing one of them could not be applied. Both are fixed, and a report that stalls now fails within about 20 minutes instead of spinning for an hour.
How to use it
Every database now has a New row button in its header, so adding a record no longer means scrolling to the end of a long table. Views switch with tabs centred above the grid, and a database can run at standard or full width, like a page.
How to use it
ActVox now uses Stripe for subscriptions and one-time ActVox Credits. Billing shows whether you are paying personally or through an organization, explains applicable taxes before checkout, and gives you one portal for payment methods, invoices, plan changes, and cancellations.
How to use it
You can now bring still images up to 100 MiB into a Content draft. ActVox converts and resizes the source before upload, keeps the stored file within the server limit, and prepares Instagram images for its 8 MiB publish cap. Sources over 100 MiB or images that cannot be converted are rejected without storing the original.
How to use it
Until now a report whose SEO, competitor or strategy section came back empty could still read Complete, and the gap only showed up when your agent built on it. Such a report now shows Partial: every finished section stays readable, the empty one names what is missing, and Apply report waits until the report is whole. The first-run chat card says “Report needs repair” instead of a false “Report ready”.
How to use it
Bundle 3.1.1 changes how an agent writes Pages through the CLI, SDKs and MCP: creating always makes a new Page, and updating an existing one must be based on a current read of it. A script that reused a Page's id could silently replace its title and body; now that attempt is refused with a divergence message, and the bundled tools send the required check for you. Update the bundle once and existing prompts keep working.
How to use it
Bundle 3.0.0 retires the old research-actvox name and the RESEARCH_ACTVOX_* variables that kept working for one release. A Claude Code, Codex or Hermes install that still uses them stops finding the skill. Reinstalling takes one command: it checks every file against the current bundle, replaces only what changed, and your saved prompts just need to say actvox-workspace.
How to use it
When the strategy step of a report timed out, the Strategy page showed an empty workspace with nothing to press, and the report's Blog section vanished along with its Regenerate button — it looked like the report had simply produced nothing. The failure is now named, the retry is one button, and the step gets twice as long to finish.
How to use it
Handing a Help section to an agent or pasting it into a document used to mean copying the whole guide and trimming it by hand. Each section page now has two actions: Copy page puts just that section on your clipboard as Markdown, and View as Markdown opens the same text on its own, ready to link or feed to an assistant.
How to use it
A product, a part, a venue — a page could show a photo of it, or a link to a viewer somewhere else. Now the page can hold the model itself: upload a GLB or glTF and readers orbit and zoom it in place, with auto-rotate, fullscreen and a poster you capture from the angle you like. Attach a USDZ and readers on iPhone and iPad can place it in their room with AR.
How to use it
Editing an HTML block meant a bare text box and a Save to find out what you had done. Source now opens a code pane beside the block's own preview: the preview follows your draft as you type, a divider gives either side more room, Tab indents and ⌘S saves. Interactive artifacts get a Details table for the title, height, data and fallback text, and any block can go fullscreen or open in a tab of its own.
How to use it
The assistants you already use can now work inside your ActVox workspace, with no API key and no terminal. Each tile walks you through the whole journey — add the connector in the assistant, allow exactly one workspace on the ActVox consent page, then send the first prompt. That prompt onboards the agent: it checks what it can do, reads what already exists, interviews you, and stores your business portrait, its behaviour rules and your sources only after you say yes.
How to use it
The Edit, Image, Carousel and Video lanes used to show four different previews — a bordered card here, a black square around the media there. Every lane now renders the same phone with the post's own chrome around your text and media: an Instagram feed post, an X post, a full-screen Reel or TikTok, a LinkedIn post with the video inside it. Hide the preview when you want the whole pane for editing, and the caption under the phone is counted against that platform's real limit.
How to use it
Generated video concepts were take-it-or-leave-it: you could pick one or write a prompt from scratch, but not change a line. Each concept now opens as an editable script — the hook, the scene, the visual style, the on-screen text, the call to action and the audio — and your edits are saved as you type and marked Edited. The video renders from your version, so a good concept with one wrong line no longer costs a second render.
How to use it
The connector setup used to tell you which commands to run without a way to get the `actvox` command onto your machine. The CLI is now published on npm: pick your package manager, copy one install command, and the sign-in steps that follow work as written. It runs on macOS, Linux and Windows with Node.js 20 or newer, and your credentials stay in your system keychain.
How to use it
An image, carousel, video or text generation used to live only in the window that started it: reload, and it was gone — sometimes failed and refunded while the work quietly finished anyway. Every generation now belongs to the draft itself. Come back from any window and the Studio tab shows how far along it is, or the finished result, and nothing is charged twice.
How to use it
One control now picks the model for images, carousels and video. Auto, Fast and Quality are presets that choose providers for you and fall back between them; picking one provider pins it. The estimated cost sits right beside the choice, and each result records which model you asked for and which one actually ran.
How to use it
Generating a new set of slides used to throw the old one away. It now asks first, tells you what it will replace and what it costs, and keeps the previous set for seven days. If the new set is worse, bring the old one back in one step — and a version list shows every set the draft has had.
How to use it
Every save of a document the assistant wrote is now kept, with who saved it and when. You can read an earlier version before deciding, and bring it back in one step. Restoring adds a new version rather than overwriting anything, so the trail stays complete.
How to use it
Filtering the content list and then stepping away to another tab used to undo all of it: you came back to an unfiltered list at the very top, behind a loading screen. Your filters, sorting, search, layout and scroll position now stay put, along with the draft you had open. Two windows keep separate views, so neither one moves the other.
How to use it
Working in two browser windows on one workspace used to mean they shared a single selected tab: opening something in one window jumped the other window to it, and a reload landed on whichever had written last. Each window now keeps its own selection, and remembers it across reloads.
How to use it
A connected client used to be a name on a list: you could grant it access, but not check what it had done with it. Connections now shows when each client last called and what it verifiably changed, and pausing or revoking one denies its next call on the record. You can export the whole timeline as a single document.
How to use it
Opening one tab too many used to close the oldest one for you — sometimes the one holding text you had not saved, or an image still being generated. Tabs with unsaved changes or a running generation are now protected and stay open. When everything open is protected, you decide what goes instead of losing it silently.
How to use it
Media studio could tell you it had made four images and then show you none of them. The images were real and already saved — the panel was only remembering them until you switched tab. Now it reads them back from your workspace, so they are still there when you return.
How to use it
Generated video concepts were never stored anywhere, so leaving the tab threw them away and you paid to make them again. They are saved to your workspace now, and the one you picked stays picked.
How to use it
A finished report could look hollow — an empty calendar, blank plans, a competitors card repeating the same agent twice. The report was complete; the page wasn't reading it. Now every tab shows what the research actually produced.
How to use it
A five-slide carousel used to give up after a slide or two and mark the rest as timed out. The later slides were simply started too late. Now each slide gets its full generation time, and even a ten-slide set arrives complete.
How to use it
Ask the assistant to move a file or rename a page, and the reply could still come back in English while you work in Russian, Ukrainian or Arabic. Those quick confirmations now follow the language you chose, like the rest of the assistant.
How to use it
Your workspace now speaks MCP, the standard connector protocol. Add ActVox as a custom connector in ChatGPT, approve its powers on the ActVox consent screen, and ChatGPT can read and work in the workspace you choose. You can pause or revoke the connection in Settings at any time.
How to use it
When someone shares a public page with you, you can make it your own. The copy lands in your active workspace, fully editable, while the original stays untouched — and you never get access to the source workspace.
How to use it
Sign up with a website or social profile and the ws-DNA setup interview begins on its own while your company report is still running. It reuses what the report already learned, so it asks fewer questions, and your answers pick up an interview already in progress. Setup now speaks English, Russian, Ukrainian and Arabic.
How to use it
You can keep the same page open in both panes while you work. We fixed a problem where opening a newly created page in a second pane could clear its initial text; both editors now retain the saved document.
How to use it
Until now, asking the assistant to add its research to an existing database ended in an apology: it could only create new things. It now drafts new rows in the database you name and hands them to you as a suggestion to review, so nothing lands without your say.
How to use it
Open a page and ask the assistant to fill it, finish it or add to it, in English or Ukrainian, and it drafts the change as an inline suggestion on that page. You review each change or take the whole suggestion; the page does not change until you accept.
How to use it
A long request used to look like nothing was happening for a minute or more. The typing indicator now names the step in progress — gathering context, drafting variants, ranking the drafts — so you know the work is under way and roughly where it is.
How to use it
Manage Account is now your whole account, not a form with three fields: profile and email addresses, connected sign-in accounts, passkeys, every active session, and account deletion. A passkey lets you sign in with your face, fingerprint or device PIN instead of a code, and the devices list shows where else you are signed in.
How to use it
The Billing page now opens with a strip that names the account being billed and its plan. Inside a team workspace you can switch to your own billing to buy a personal plan or credits without leaving the team, and Team is priced per seat with the monthly total shown as you pick the number of seats.
How to use it
Manage organization now holds everything about the team in one place: the name and logo, members, pending invitations and roles, leaving or deleting the organization, and a Credits page with the shared balance and the seats you pay for. Admins invite; members see the same screens read-only.
How to use it
Pages, databases, decks, generated images and shared links now carry your own colours, type and style instead of ours. You decide when your logo goes on a generated image or deck, and a workspace with no brand captured yet gets a clean neutral look rather than ActVox's. Nothing to switch on — capture your brand once and it follows every piece of work.
How to use it
Every page, deck, interactive page and generated image the assistant makes now carries a thumbs-up and a thumbs-down. Press thumbs-down and you can say in one tap what was off, so the complaint reaches the people tuning the assistant instead of dying in a chat you close. Your rating is saved whether or not you pick a reason.
How to use it
Your workspace now publishes its brand as three files an outside AI tool can read: the written guidance, a ready-made stylesheet, and the brand as plain data. Hand one of those addresses to another assistant and it builds pages and sites in your colours and type without you pasting hex codes. Access is read-only and needs a key you issue yourself.
How to use it
Your workspace now keeps a single library of every file it holds — uploads, generated images, covers and attachments alike. Add files straight into it, and reuse any of them as a cover or attachment without hunting for the original or uploading it twice. And when you delete one, you first see everywhere it's used, so nothing quietly disappears from a page.
How to use it
Give a number column a display format — currency with a real currency code, percent, or grouped thousands with the decimal precision you choose — and a date column a short, long or relative style, with time when you need it. Your stored values stay untouched: sorting, filters and formulas keep working exactly as before, and every view and shared link shows the same formatting.
How to use it
Formula columns now come with a guided editor: insert columns and functions from searchable pickers, or start from a ready template like A × B or A as % of B. A live preview shows the result on one of your real rows before you save, and plain-language messages explain exactly what's wrong instead of failing silently. Renaming a column no longer breaks the formulas that reference it — they follow the new name automatically.
How to use it
Databases now look like the rest of your workspace: an icon and a cover of their own, visible in the sidebar, tabs and shared links. Pick an emoji, an ActVox glyph or upload an image, and frame the cover exactly how you want it.
How to use it
Sometimes the assistant would describe the drafts it wrote for you — and nothing appeared in your workspace. Now every draft it reports is really there, and when you ask for the alternate versions, those are saved too. Links you paste with a request to add them to your knowledge base are ingested and confirmed one by one.
How to use it
Messaging your workspace bot used to end in an apology on every single message. That is fixed: the bot now runs your question through the assistant and replies in the chat, keeping the conversation history per workspace.
How to use it
Your bot no longer answers strangers who find its username — only you and the people you approve. Add a teammate by their Telegram ID, and anyone else gets polite instructions for requesting access.
How to use it
Workspace OS now comes in light and dark, and it remembers which one you picked. The whole shell follows — sidebar, tabs, pages, boards and the phone navigator — so a late session stops being a wall of white. Your reports and public pages are unchanged, whatever you choose.
How to use it
Search used to live down in the sidebar, so finding something meant first going to the place the search box was. It now sits in the workspace header beside notifications, in reach from any page, board or row card. The keyboard shortcut and the results palette are exactly the ones you already use.
How to use it
The use cases page showed four outcomes, which told you what is possible but not whether it applied to you. It now carries 32 workflows grouped by who you are — creator, restaurant, hotel, small business, SaaS — each one naming what goes in, what the work is, and what comes out.
How to use it
Reusing a structure you had already built meant rebuilding it: a new database, the fields typed in again, then every row pasted across. Now you copy the whole thing at once — every row, everything written inside each row, and the way you had the view set up. Copy a page and you get its sub-pages and the databases inside them too. The copy is genuinely separate: edit it, or delete it, and the original does not notice.
How to use it
With seven tabs open, the next thing you opened was refused — a message telling you to go and close something before you could carry on. Now the oldest tab you are not using steps aside and the new one opens. Nothing is lost: everything you had open is still in your workspace, one click away in the sidebar, exactly as you left it.
How to use it
Clicking a row in Board or Gallery gives you a card, which is right for a quick edit and cramped for real writing. Now you can promote that row to a full workspace tab and keep going with the whole page in front of you. You can also add a row straight from Gallery, including when it is empty — no more switching to Table just to create one.
How to use it
The workspace was built for a wide screen, and on a phone it showed: the navigator shrank to a rail too narrow to read a page name in, and the row of tabs ate what was left. Now the navigator slides over your document and back out of the way, the tabs live in a switcher you reach with your thumb, and the bar at the bottom names the page you are in and tells you whether it is saved.
How to use it
A shared database used to arrive as a table wider than the screen, so whoever you sent it to saw a title column and had to drag sideways for every other field. On a phone each row now arrives as its own card, with its status and figures laid out to be read. The page also says when it was last updated — and that now counts edits to the rows, not only changes to the columns.
How to use it
Page icons used to be a grid of twenty emoji, so most pages ended up wearing the same few. Now there is a searchable set grouped by what you are marking — documents, work, status, people, time — plus the ActVox mark, and you can upload your own image for a page that deserves its own.
How to use it
One agent key now opens the whole workspace to an outside assistant — pages, reports, content drafts, the knowledge base, the file tree and a change feed — through a Python SDK or an MCP server you run on your own machine. The assistant sees exactly what the key allows and every write it makes is checked against the latest version and recorded under the agent's name.
How to use it
Agent keys now come in finer cuts: settings and strategy, reports, content, the knowledge base — read or write each on its own — plus three powers you grant deliberately: execute, share and publish. A key can never combine publish with a writing scope, so a person always approves what an agent posts, and every key is rate-limited per minute and per day.
How to use it
Until now an agent key only reached the pages, databases and whiteboards that agent had created or been granted one by one. A key with the new Workspace Operator scopes can read and update everything in the workspace — your pages included — while every edit still checks that nobody changed the page underneath it. It still cannot share, touch billing, manage members or issue keys, and anything it does is recorded under the agent's name.
How to use it
An agent is now a lasting identity and its keys are just credentials. Issue a fresh key for the same agent, switch it over, then revoke the old one — everything it created, was granted, shared or recorded stays exactly where it was. Before, replacing a key meant creating a new agent that started with nothing.
How to use it
An agent you already use — Claude Code, Codex, OpenClaw or any tool that speaks an API key — can now read and maintain your pages, databases and whiteboards directly. It makes precise, conflict-checked changes rather than rewriting whole documents, so a correction from an agent and an edit from a teammate can land on the same page without destroying each other. You decide exactly what a key can touch, and you can hand an agent maintenance of one specific page, database or board it did not create — and take it back instantly.
How to use it
Sharing no longer means hunting for the right menu. Tell the assistant to share the page or database you are looking at, choose public or organisation-only, and approve its suggestion — it mints the read-only link (or finds the one that already exists) and hands it straight back in chat. Nothing is ever shared without your explicit approval.
How to use it
A connected tool can now put a working chart, map or small interactive on a page instead of a screenshot of one. In a browser that supports credentialless frames it starts automatically; otherwise you see its title and plain-text description. The running artifact is sealed off from your account — it cannot read your session, your cookies, or anything else on the page — and printing the page still gives you the description rather than a dead frame.
How to use it
After certain sync errors a page would quietly stop accepting edits: resize handles disappeared, column dividers vanished, and images or embeds inside columns looked deleted — with nothing on screen saying why. The page was whole on the server the entire time. It now rebuilds itself from the saved copy when you open it, and a browser that cannot read one update is no longer allowed to overwrite that content for everyone else.
How to use it
Organisations lived in the top menu and workspaces in a separate switcher above your files, so moving between two clients meant two controls in two places. Your workspaces now nest under the organisation they belong to, in that one menu. The sidebar starts at your files again, and switching organisation keeps the menu open so you can go straight to a workspace inside it.
How to use it
A page could only ever be one column deep, so anything meant to be compared had to be read one after the other. Now you drag a block onto the left or right edge of another and they sit side by side — up to five columns, with dividers you drag to decide how much room each one gets. On a narrow pane the columns stack themselves back up, so the same page still reads on a phone.
How to use it
Asking the assistant to improve, rewrite or add to the page in front of you produced a second, near-identical page in the sidebar instead — for a whole month, every single time. The words that mean editing were missing from the part that decides what you asked for, so every request read as “make something new”. Your open page is now what the assistant works on, and the duplicates stop.
How to use it
Images went into a page one at a time, each taking a full block, so a set of ten was a page you scrolled rather than a set you looked at. A gallery holds up to ten in a single block and switches between a grid and a swipeable carousel with one press. Give an image a title and its cell becomes a card, which is what makes a gallery work for products or people rather than only for photographs.
How to use it
A link used to stay a link, and a bookmark card was built from the address alone — same hostname title on every one, which is why they read as having no preview at all. Now a pasted link keeps its place in the sentence and offers, beside it, to become a mention carrying the site's favicon and real title, a card with its description and image, or a live embed. The offer disappears the moment you keep typing, so it is available without ever being in the way.
How to use it
Splitting the workspace was buried in a menu, so most people never found it. Now you drag a tab to the right or bottom edge of the pane and it opens there — the same gesture works for anything dragged out of the sidebar. A workspace also now holds at most seven tabs, with “Close other tabs” in the tab menu, because eight or nine left every one of them too narrow to read.
How to use it
A whiteboard created from a page landed at the top of the sidebar, next to boards from every other page, so the connection to the page it belonged to was lost the moment you made it. Boards now nest under their page, can be dragged into and out of one, and a page you delete asks what to do with anything nested inside it rather than quietly leaving it behind.
How to use it
A draft could only carry what the AI made for it. Now you attach your own files — several at once, in the order you want them, each removable on its own — and they publish through to Instagram, LinkedIn and Facebook. Each file is checked as you add it, so a post that a platform would reject tells you now rather than after you press publish.
How to use it
You could already explain a rating, but the button was easy to miss and always asked what went wrong — odd, under a compliment. Now a love or a dislike opens the note by itself, asking what worked or what did not. A like stays quiet, and dismissing the note costs nothing: your rating is already saved.
How to use it
A thumb told us an answer missed, never why — the wrong numbers, the ignored brand guide, the page it edited instead. Now a rating can carry a note in your own words, and it is yours to reread, change or take back. Rating still happens on the tap: the note is offered afterwards, and skipping it costs nothing.
How to use it
The assignee cell held one name, so work shared by two people had to pick a favourite — and only that person was told about it. Now you can tick everyone responsible, each name shows on the row, and everyone newly added is notified.
How to use it
Generating drafts with AI had stopped producing anything — the run started, took your time, and returned nothing. It works again, and the credits spent on the failed attempts were refunded automatically.
How to use it
Images used to fill the whole column, so a small logo took as much room as a full screenshot. Now you can size each one — by dragging, by preset, or from the keyboard — and the size sticks when you reload or share the page.
How to use it
Whiteboards were only reachable by converting a diagram. Now you can add one straight from the slash menu: the page keeps a preview card, and the board opens in its own tab under that page.
How to use it
Connect Google Search Console and your workspace fills with real search data: 16 months of history import on connect, and daily clicks, impressions, and positions keep arriving on their own. No more switching to Google to check how pages rank.
How to use it
New workspaces now start from one small dialog instead of a separate setup page. Start blank for docs and chat, or paste your brand link and land in a workspace with your analysis already running — and a setup chat waiting either way.
How to use it
Highlight a passage on any page and ask the agent to improve or change it. The proposal now touches only your selection — the headings and paragraphs around it stay untouched — and it works even when the selection spans several paragraphs. You review every change before it lands.
How to use it
A row you open as a page used to show only its title and text — every property stayed locked in the table. Now the full set of properties travels with the row: edit them in the card or on the row's page, and add a new property right there instead of going back to the table.
How to use it
Databases in the sidebar used to be a single closed item — reaching a row meant opening the whole database first. Now a database expands right in the sidebar to show its rows, and one click takes you straight to a row's page.
How to use it
Since the move to the new address, dialogs — row cards, delete confirmations, rename prompts — opened below the visible screen, and menus inside cards would not respond to clicks. If the workspace seemed to ignore you these past days, it was not you. Everything opens centered and responds again.
How to use it
The website now lives at actvox.com and your workspace at os.actvox.com. Splitting them makes the public pages far lighter and lets the workspace ship on its own schedule. Two things change for you: links to the old research.actvox.com address no longer work, and everyone is signed out once as part of the move. Your workspaces, reports and content are untouched.
How to use it
Hermes, Codex, and Claude Code can now work with your workspace through a scoped key instead of a shared account. The hosted installer brings in reviewed skills, including Document Studio for source-backed decision memos, PRDs, research reports, and operating plans. Public sharing stays off unless you explicitly enable it.
How to use it
The graph now groups related knowledge into named clusters instead of leaving you with a wall of nodes. Ask which ideas move together, what is most connected, what links two nodes, or what sits nearby. Every answer comes from the stored graph and focuses the canvas on the evidence behind it.
How to use it
Your workspace now holds whiteboards alongside pages and databases: an open canvas you and your teammates can draw on at the same time, each cursor visible as it moves. Inside a page you can write a diagram as plain text and watch it render, and the assistant will draw one for you when you ask for a flow or a chart. Any diagram can be forked onto a whiteboard when you want to move the boxes yourself.
How to use it
Until now the assistant only kept what it decided to keep on its own. Now you can hand it a fact directly and it holds on to it - a launch date, a client's preference, a rule your team works by. You can ask what it remembers, and you can tell it to forget something you no longer want kept.
How to use it
When the assistant makes a page or a database, it now opens by itself instead of waiting to be found. While it works, the tab and the page both show you where it is. And you no longer have to describe the passage you mean - select the text and hand it straight over.
How to use it
AI image models cannot draw a real logo - they produce a lookalike. Now, when you ask for an image with your logo, the assistant generates the picture and then stamps your actual logo file onto it, exactly where your placement settings put it. If a draft is open, the stamped image becomes its selected visual.
How to use it
Your workspace logo now lives in one place. Upload it in the Brand hub or in Settings - both write to the same Brand DNA, and the sidebar, the image editor and the assistant all use the same file. A new Placement panel decides where the mark lands on each format, from square posts to stories, with a live preview.
How to use it
Until now your brand reached only the pictures - the assistant drafted text without it. Now the rules you write yourself travel into every draft, whether you start one in the compose wizard or press Write with AI in the editor. Give a channel its own voice once and drafts sound like you by default, without picking it every time.
How to use it
Two very different things were both called a report - the full analysis of your website, and a written document - and asking for one always produced the other. Now the assistant can start the real analysis for you, and when your request could mean either, it asks once and offers both instead of guessing. Asking in Russian, Ukrainian or Arabic works too.
How to use it
Instagram drafts accepted generated images only, so footage you had already shot had nowhere to go. Now you can attach your own MP4 or MOV - up to 100 MB and 90 seconds - and the draft previews it exactly as a Reel. Publishing straight to Instagram is not open yet, so a video draft stays a preview until it is.
How to use it
Start an audit from the home page and your workspace chat now opens mid-conversation: your request is there, and the assistant answers with a live card for the report it is already building. You can open the report while it generates and watch sections fill in - and when it finishes, the chat says so without a refresh.
How to use it
A file attached in chat used to vanish once the conversation moved on. Now it becomes a real workspace file: Artifacts gains an Uploads section with search and filters, and each card says whether the file is in your Knowledge Base. Filing is always your call - attach for free, then index it yourself or just tell the agent to.
How to use it
Not every video should be generated. A TikTok draft now accepts your own MP4 or MOV directly - it is validated, stored with the draft, and survives reloads, without spending anything on an AI video job.
How to use it
Connect your GitHub account and the agent can read the code behind your site - how the pages are put together, what framework runs them, where the content lives. It reads only: nothing is ever pushed, changed or published. You choose up to three repositories, and you can disconnect at any time.
How to use it
A report used to be a photograph of the month you ran it. Now you can refresh it - on demand, or automatically each month - and every run is saved as a dated snapshot beside the original, with a note on what changed. A refresh re-runs the audit only, so your drafts and images are never touched.
How to use it
After entering your email, sign-up asked for a six-digit code but showed nowhere to type it - just a blinking cursor in blank space. If that stopped you from signing up, or made you think your browser was broken, it was us, not you. The boxes are visible now, on sign-up and sign-in alike.
How to use it
Your balance now sits in the sidebar, so you can see what is left without going looking for it. Opening it keeps you inside the workspace instead of sending you off to a settings page or the public pricing page. Whatever you had open is still open when you come back.
How to use it
Adding a cover used to mean accepting whichever slice of the picture happened to land in the band. Now you place it yourself before it uploads - move it, zoom in, and set how tall the band should be. Nothing is cut away, so you can change your mind later.
How to use it
Generating social drafts had been reporting a timeout on work that had actually finished, so you retried something that had already run. Drafts now arrive as they should. Picking one network also generates just that one, rather than quietly writing for all of them.
How to use it
Editing a report section used to mean typing into a block of raw text where one stray character could throw the whole thing away. Now each section opens as a proper form - a rating, a review count, a list of issues, each in its own field. A bar at the top counts anything you have not saved yet, so nothing gets lost when you move on.
How to use it
Every generated file used to look the same on the page, whichever kind it was. Now each card shows what it actually is - presentation, document, spreadsheet, image or video - and pictures and clips show a real thumbnail. Sort by name or date, and swap the whole page to a compact list when the grid gets long.
How to use it
Two skills with overlapping trigger phrases used to fight quietly, and you only found out when the wrong one answered. The editor now tests any message against your whole catalogue and names the skill that would win. You also get a live preview of the skill's card as you write it.
How to use it
The Recent section used to list only chats. It now merges chats, pages, databases, artifacts and running agent tasks into one feed sorted by last activity, so getting back to yesterday's work is one glance instead of a hunt across tabs. Filters and grouping are remembered per workspace, and chats you archive stay out of the way until you ask for them.
How to use it
Chats, pages, databases and generated artifacts can now be pinned to a Pinned section at the top of the sidebar. Your daily set stays one click away in the order you choose, instead of being re-found through search or scrolling every morning. Pins are yours alone - they do not rearrange anything for your teammates.
How to use it
A new Skills and Connectors page lists every skill your assistant has, with a plain description of what each one does and ready-made prompts you can try in one click. The same page connects LinkedIn, Instagram and Telegram, shows which connections are live, and lets you request the integrations you want next.
How to use it
Leaving a tab used to throw away whatever you had not saved yet - a sentence being typed, a cell being edited, a title being renamed. Tabs now stay alive when you switch away, so your text, cursor and scroll position are still there when you come back. If a save ever does fail, you get told, and it retries.
How to use it
New databases now start from a dialog instead of one fixed set of columns - pick a marketing template, copy the structure of a database you already have, or start empty. Several templates arrive pre-filled from your own reports, calendar and strategy. Once it exists, every cell type opens its editor on a click, and you can drag columns and rows to the size you want.
How to use it
Anything tabular in your workspace can leave as a real spreadsheet. Numbers arrive as numbers and dates as dates, so you can sum and sort them straight away instead of cleaning up text first. Reports export one sheet per section, and any section that is prose rather than data is named on a summary sheet so you can see nothing was quietly dropped.
How to use it
Any content draft can now become a short vertical video - the kind you post to TikTok or Reels. You pick a concept, we generate it, and it saves onto the draft and into your gallery. Videos can run up to two minutes.
How to use it
We can read your website or Instagram and store your real colours, typefaces and logo. Every image, carousel and video generated after that starts from your brand instead of a generic look. Anything you edit by hand is protected - later scans arrive as suggestions.
How to use it
A generated image opens on a real canvas where you can move things, add text, stamp your logo and change colours. Small corrections no longer cost you another generation - and another wait.
How to use it
When part of a research run fails, the report used to look finished - with a zero score where the missing analysis should have been. It now carries a notice naming exactly which sections came back empty, so you can regenerate those instead of trusting a wrong number.
How to use it
Every article in the library is now a complete piece you can actually work from - audit-to-workspace workflows, agency SEO and AEO audits, AI visibility scoring, and how to run content operations with a small team. Several of them were previously a paragraph and a promise.
How to use it
Team costs $25 per seat, and the seat count follows your member list on its own - add a teammate and the plan adjusts at renewal. The token allowance grows with the seats, and the whole team spends from one shared balance.
How to use it
Your balance now sits beside a running total of what you have spent today. Instead of noticing a drop and guessing, you can see the day building up as it happens.
How to use it
A report is a diagnosis; a workspace is where you act on it. One button now carries the whole thing across - your business name, your profile, the sources it found and the content drafts it suggested. The report keeps an “Added to workspace” mark so you never do it twice.
How to use it
Every report now opens as tabs with score rings and key-insight tiles, so the shape of the problem is visible before you read a word. Audit findings carry a severity grade, can be filtered, and show the evidence behind them.
How to use it
Billing runs on tokens now, so a short question costs far less than a full research run - instead of both costing one credit. Long conversations are billed at cached rates, which makes an extended chat cheaper than it used to be.
How to use it
Pictures now keep their real shape instead of being cropped to squares, and each one tells you whether it is already in a post or still unused. An unused picture can be dropped straight into a draft that has no image yet.
How to use it
One top bar now runs across every workspace page and no longer shifts when a sidebar closes. A collapsed sidebar disappears completely instead of leaving a grey strip, and both reopen at the width you set them to.
How to use it
Looking for every single change?
Read the full changelog →
Run a free audit and experience the latest improvements firsthand.
Start my free audit