Every conversation you have had with your AI experts since 5.1.0 has been teaching them how you work.

• The client who insists on British English.
• The plugin you never recommend.
• The way you want a recommendation costed before anyone touches a site.

They wrote it all down and it has been steering every job since.

From today that memory is yours. Open it, read what they have learned about your agency and each of your clients, and change anything that is wrong. That is what this release is for.

Alongside it, WordPress work is faster and costs fewer credits, because your experts stop asking a site the same question twice.

New features

Your experts now have a memory you can read and edit

The specialists keep one memory for your workspace and one for each project. The workspace memory is read on every job. Each project memory is read alongside it, on that project only. Until now the experts were the only ones who could see either.

Settings → AI Web Agency → Memory shows both.

Open one and you are reading the experts' own notes on how you and that client work. Two ways to change it. Edit opens the file itself. Notes for your experts takes a plain sentence ("stop suggesting Elementor Pro") and folds it in on the next sweep, so you never have to write in the file's own structure.

We tested it the way you would use it, in a task on a client site:

Remember these for this project going forward: the client writes in British English, they do not want to be sold Elementor Pro, and every recommendation should state its credit cost before we act on it.

The three rules landed in the project memory exactly as written, under three headings, attributed to the person who asked. Project memory reaches task chat, so that is where they show up. Owners and admins edit, contributors read, clients and collaborators do not see it.

One thing worth saying plainly, because memory makes people ask: nothing you or your clients put into Atarim is used to train AI models. Not ours, and not our providers'. Every request goes to endpoints that keep neither the prompt nor the response. The full detail is in our Platform privacy notice and DPA.

WordPress work is faster and costs less

Every time a specialist works on a connected WordPress site it first has to find out what the site can do. On a site running JetBackup, Elementor, WPForms and Yoast that is a list of 264 tools, most of them nothing to do with the job in hand, fetched again for every task and every step of every workflow.

Plugin 5.1.4 changes how that conversation runs. The specialist asks for one area at a time, and runs many abilities in a single request instead of one call each. We ran the same listing on the same site before and after updating: 264 tools, then six. Skills carry their own tools, so a specialist activating a skill starts with what it needs rather than searching for each one. Automation steps begin with the tools they used last run.

Fewer round trips to your site means less waiting between a request and a result, and fewer credits spent getting there. Nothing to switch on; update the plugin and it applies to every task and workflow on that site.

Improvements

Restore points tell you when they are ready. On a site running JetBackup, readiness is tracked until the snapshot actually reads as ready, so a point that is still arming is never mistaken for cover. The previous locked snapshot is released and snapshots marked for deletion are cleaned up. We asked Glitch on a test site:

Take a restore point of this site first. Then tell me the WordPress version, the PHP version, and which plugins have updates waiting. Report exactly what happened with the restore point, including whether it is actually ready to restore from.

The answer named the job, said the database was dumped and files were still archiving, and stated plainly that the new snapshot was not yet ready to restore from while the one from 4 September was. That is the sentence you want before risky work.

Or put it on repeat: Weekly backup and verification in the workflow library confirms every site has a recent, restorable backup and warns you when cover slips.

Schedules stay as you wrote them. The picker preserves cron expressions it cannot represent, so "last day of the month" and "second Tuesday" survive an edit. A recurring workflow with no schedule is refused before saving, and a run is priced from its schedule rather than at zero.

Straighter answers from hosting and the site. A firewall that blocks a call says so. A hosting provider that is not connected returns its own message instead of a gateway error. An unreachable WordPress site is reported as unreachable rather than landing in your approval queue as a "destructive" action.

AI review knows when it has finished. Failure is announced once, when it has actually failed; a failed retry keeps your issues; the tasks tab shows loading with a retry instead of "complete" while results stream in.

White-label. The default logo is a PNG that renders in email, and a brand colour without its hash still makes valid CSS.

Smaller things. The Slack channel picker pages and surfaces fetch failures. Help links in settings and inbox point at articles that exist. Tablet scrolling, review stickers in device preview and the graphics small-screen notice are fixed.

Bug fixes

• Onboarding and plugin activation. Terms must be accepted before an account exists. An empty "Other" role is refused. A domain the crawl could not read gets a clear message and a retry, and the plan works from what you typed. An already connected WordPress site says so and offers a way forward. Activating the plugin on a site you already added merges onto that project instead of creating a second one. One-click login from the dashboard establishes a real WordPress session, and a blocked popup tells you. New client accounts no longer crash on first login, and a multi-project client lands on all their projects.

• Clients, folders and archiving. Archiving every project in a client folder keeps the folder visible and the client manageable. Deleting all of a client's projects no longer blocks re-adding them, and a project can be taken back out of a client folder.

• Automations and integrations. Tag Added runs Add Comment. Asana picks up task updates. A ClickUp attachment with an empty path no longer crashes the sync. Make gets its page info. Comment integration references save cleanly. Webhooks carry the original HTML alongside every flattened field.

• Chat and the specialists. A chat thread survives a model that takes over 45 seconds to send its first token. Elementor edits are validated before save, so a bad setting no longer aborts your next Publish. The WordPress connection recovers a lost session. On some hosts WP-CLI reported a successful command as failed; it reports success.

• App surfaces. Pagination in the feedback experience no longer hangs without the Chrome extension. The sidebar keeps its width. Cookie consent sticks. Share no longer crashes from the widget or the project card. Filter by latest activity works. Comment attachments download from the plugin. Old-storage avatars load. Free-plan projects from the Chrome extension keep AI on.

• Billing. Credit pack checkout no longer fails. A cancellation the billing system refused is reported as refused, and cancel-to-free works on both routes.

Coming next

Built and live for our team, opening once they have run against real customer sites:

• Do It without a browser tab. Do It runs from the backend, with undo, whether or not anyone has the page open.
• Browser-side task screenshots with a one-time consent prompt, so logged-in state and dynamic content are captured.

Go and read your memory

Open Settings → AI Web Agency → Memory.

Take a look through what your experts have written down so far. Some of it is probably going to be spot on, and some of it you'll want to tweak or correct, and that's exactly how it should be. The whole idea is for your agency to run the way you run it, and now you finally have a way to check that it does. Have a dig around and let us know what you find.

Vito

One of our tickets this release carried a line that ended up describing the whole thing: silence reads as success.

When a step can't do its job and says nothing about it, the system takes that quiet as confirmation that it worked.

So we went looking for everywhere Atarim could report "done" without having done the work, and we found four:

• A workflow step that couldn't reach your project tool but still showed Success
• A board move that failed and reported that it worked
• A monthly job set for the 31st that skipped every short month without mentioning it
• A backup that passed every check it was given but couldn't actually restore a site

All four now tell you what really happened, which is the theme of this release.

When Atarim tells you a job is finished, it's finished, and there's a working backup sitting behind it. Here is what all we shipped:

New features

A backup before the AI touches anything

Before Glitch changes files or the database on a WordPress site, he can now set a restore point and track it until it is ready. If a change goes wrong, the site goes back to how it was. This runs through JetBackup on the site. On hosts where the server does not run backups on time, Glitch runs the queue himself. So "ready" means ready.

We learned a rule here. A backup only counts if the tool that would restore it can reach it. A backup at the hosting level with a restore tool in the plugin will pass every check and be useless when you need it. The team knows the difference now.

Before you touch anything, set a restore point for this site covering files and the database, and tell me when it is ready and what it covers.

Or put it on repeat. The Weekly backup and verification workflow checks every site has a recent backup that works, and warns you the moment one is missing. Monthly WordPress update round already backs up first, then updates, then rolls back if something breaks.

Glitch can edit more of your sites, and undo it

Ask Glitch in a task to change something on a page. The plugin now handles Beaver Builder and SiteOrigin pages, along with Gutenberg and Elementor. Block edits are more reliable too. And if you don't like a change, it can be undone.

WP-CLI for the big jobs

Some jobs are too big for a normal tool. A search and replace across the whole database. A full database export. Rebuilding every image size. For these, Glitch can now run WP-CLI commands, in the background when needed. Only admins can ask for it. Every run is logged.

Use WP-CLI to list every installed plugin with its version, whether an update is available and whether it is active, then tell me the database size and the three largest tables.

Index can see AI search now

Index used to stop at Google's ten blue links, but now he can see more:

• Who Google's AI Overview cites, and the exact words it quoted.
• Which sites ChatGPT names for a brand or a search.
• What ChatGPT actually answers, with every source.
• And rank history month by month, so a report can show what moved without you saving the old one.

One rule we made sure of: when a site has no AI coverage, Index says zero and stops. He does not guess.

Using live UK search data, tell me whether this site is cited in Google AI Overviews or mentioned by ChatGPT for nutrition counselling searches, and show me how its rankings have moved month by month over the last year. If there is no coverage, say so plainly rather than estimating.

Swap in your own country. The tools need one.

A brand kit that tells you what it could not read

Paste a client's URL. The brief comes back with colours, fonts, tone and audience. We rebuilt how it reads a site. It reads the pages the brief is about, not just the front page. It picks the colours that cover the most of the page. It no longer treats a cookie-banner button as the brand's call to action. It finds the real logo and fonts.

Then a check runs on the result. If the site was measured before it finished loading, the kit is marked as not trusted, and the status bar tells you. No quiet guesses. If you regenerate a brief, it now tells you which fields came back empty instead of leaving them blank. And if a firewall blocks the read, you get a link to the whitelist guide.

Glitch checks for known security holes

Glitch can now scan a connected WordPress site for known weak spots in plugins, themes and core. He tells you the version you have and the version that fixes it.

Scan this site for known vulnerabilities in its plugins, theme and WordPress core. List each with the affected version and the fix version, and tell me which need patching this week. Only report ones you can confirm against the installed versions.

Or put it on repeat. The Monthly security scan finds weak spots before they become a breach, fixes the safe ones, and flags the rest for you.

Claro now works at the hosting level, not just sites

Connect an InstaWP or Rocket.net account and Claro can list your sites, check backups, and run hosting actions for you. Anything that writes or changes something waits for your approval first, and if a run happens unattended, Claro is limited to the read-only tools. You can also link a project to its Rocket.net site manually, and both Rocket.net and WP Activity Log now appear on the project card.

The smaller ones

• Per-project permissions now have their own screen in Project Settings.
• Plugin updates keep plugins on. Updating through the dashboard, or in bulk, no longer turns anything off.
• One task card everywhere. Boards, Inbox, Chat and the sidebar now show the same card, with subtasks rolled up. There is a new full-width card with actions built in.
• Onboarding rebuilt. Sign up, connect, verify, invite and password screens now share one design. Changing your email partway through sign up no longer locks you out of your own workspace.

Security

• Changing your password now asks for your current password. Before this, anyone with a logged-in session could set a new password, sign the real owner out, and stay in. We found it while re-checking a report from an outside security researcher in May. His finding did not hold up. This one did.
• Sign-up analytics no longer carry raw email addresses or invite and activation tokens.
• Brand values pulled from a crawled site are cleaned before they go into the brand mock. A bad site cannot inject code into your project overview.
• Hosting and connected-account tools are kept away from client roles. A run with no one watching cannot publish theme changes.

Improvements

Silence no longer reads as success

• A workflow integration step that cannot run is marked Skipped, with the reason. One the other tool rejects is marked Failed, with that tool's message. Twenty places across eight integrations used to stay quiet.
• A failed board move now says it failed and puts the card back.
• Monthly schedules set to the 29th, 30th or 31st no longer skip short months in silence.
• Automations that did nothing no longer report success.

Long jobs finish

A long run used to die when the chat got too long. Now a long thread is squeezed into a checkpoint and picked back up. A thread can hold one goal and keep working on it when idle, so multi-step work does not stall waiting for you. Approving something wakes the run that was waiting on it. And each tool has its own time budget, with a nudge if the AI starts going in circles.

WordPress work goes to the right specialist

Page builders, forms, theme files, going live, block themes, hosting and client accounts each have an owner now. Claro no longer promises a button that is not on screen, or an approval step that may not happen. The monthly update round can now update core, themes and major versions, and put the site back if the check after fails. A review pass can now say a finding is wrong.

and

• Stop actually stops. The reply ends the moment you hit stop.
• AI actions cost about a third less than they did in 5.1.0.
• Attachment icon on the task list. Status changes that stick. Custom webhooks that fire from every path. Annotate works on touch screens. The Artifacts page collects properly.

Bug fixes

• Password reset screens clear their messages and show the full password rules. A throttled "send it again" shows a real message instead of a status code. Asking for a reset with an unknown email shows the same success screen as a known one, and the docs now say why: it stops people using the form to find out who has an account.
• Workspace invite rows show their errors.
• Opening the feedback experience without a site shows an invalid-link screen instead of spinning.
• A network failure is no longer reported as "project could no longer be found".
• Editing or deleting the first comment updates the task title.
• Two WordPress connections at once no longer knock each other out.
• JetBackup errors say what went wrong instead of printing "%s".
• Partner dashboard numbers corrected.
• Display names stick instead of falling back to the email address.
• Locked projects hide their tasks. Manually added projects get a thumbnail.

Coming next

Shopify, Cloudways and Cloudflare are built, tested, and behind a staff-only flag. Claro will manage store pages and theme files, run Cloudways hosting, and handle DNS, cache and the other things a project owns on Cloudflare.

Do It with no browser tab. The backend can already edit WordPress blocks directly. So a change asked for from the dashboard can run with no browser open. It is running on our own sites first.

Task screenshots from your own browser. The feedback experience will take the shot from your tab after a one-time OK. That way logged-in pages and moving content make it into the task. Same story: ours first, then yours.

Try any of this on a client site this week and tell me what happened. The good and the bad. That is where the next release comes from. 🤘

The AI team got cheaper to run, learned some real new skills, and got a lot stricter about what it's allowed to claim. There's also a WordPress plugin security update in here, so if you skim nothing else, skim that part.

New features

AI actions now cost about 40% less

Same reviews, same workflows, same Do It actions. Your credits just go a lot further. Already live, nothing to change.

You stop paying for AI that didn't deliver

Stop a turn halfway and you used to get charged anyway, then the reply would turn up a minute later. A turn that came back with nothing still cost you, and nothing told you it had failed.

Now stopping a turn costs nothing and posts no late reply. An empty turn shows as failed with a retry button. We also added a credit check partway through long runs and a limit on repeat calls, so nothing loops quietly in the background.

Index can now do SEO the way an agency does

Writing a title tag is the easy part. Any tool can fill that box in. Knowing what to put in it is the actual job, and that takes research.

Index could already edit your SEO plugin. He just had no way to find out what he should be writing. Now he does.

What he can look up:

• Search volume - is anyone actually searching for this
• Keyword difficulty - can you realistically win it
• Keyword overview - the full picture on one term
• Keyword ideas - turn one term into a content plan
• Search intent - what the person searching actually wants
• Keyword trends - what's climbing, and what's dying
• Live search results - exactly who you'd have to beat
• Backlink profile - every link pointing at a site
• Referring domains - which of those links are worth chasing
• Competitor discovery - who you're really up against in search
• Keyword gap - what they rank for that you don't
• Domain overview - the baseline you report back to clients
• Ranked keywords - every term a site ranks for today
• Page audit - prove a fix worked, before and after
• Page content - pull apart a competitor's page in seconds
• Local listings and location targeting - go country, region or city deep

And then he acts on it. Index can change site-wide settings in AIOSEO, Rank Math and Yoast: title and meta templates, separators, brand details, redirects, module toggles. One named change at a time, on a request you typed, read back afterwards to prove it saved.

Research, decision, change. The whole loop, in one place.

Try asking him:

"Using live UK search data, find 20 keywords this page could realistically rank for. Give me the real monthly search volume and difficulty score for each, and mark the ones we could win this quarter."

"Using live UK search data, who are this client's top organic competitors, and what are they ranking for that we're not?"

"Audit this page against the top three UK results for our main term, then fix the title and meta based on what you find."

Here's an example of a competitor report:

Swap UK for whichever market your client sells into. Name it, or you get estimates instead of real numbers.

Or put it on repeat. Content plan and keyword analysis, Monthly keyword ranking report and On-page SEO review are all in the workflow library, and they all run on this new research.

Reports that have to show their working

We ran a monthly security scan 77 times. It reported success every time. On one of those runs it listed two vulnerabilities in a client's report that do not exist.

The scan knew what to say when it found problems, and what to say when it found none. It had nothing to say when the feed didn't answer at all. Asked to fill a table it couldn't fill, it filled it in anyway.

Now every workflow step that writes a document leaves a trail behind it. Each number sits next to the data it came from, each fact next to the tool call that found it. You still read a clean report, and we check it against the trail. Nothing unreadable gets called clear.

Try:

"Run the security scan on this site and show me the source behind every finding."

Or put it on repeat. The Monthly security scan workflow is the one this story is about, and it's the one that now has to prove every finding.

Turning AI off now actually turns it off

Queued runs used to carry on replying after you'd switched AI off. For one agency that meant AI answers landing on live client support tickets.

We now check the setting again right before a run starts, not just when it goes into the queue. Off means off, including for anything already waiting.

New skills

• Backups, handled by Glitch. All-in-One WP Migration, BlogVault, Solid Backups, UpdraftPlus and Migrate Guru. Whichever one a site uses, he can check the backup before anything gets changed.
• Zapier. Ask Claro to fire off a single action in an app the client already uses, without building a whole workflow for it.
• Duda. Work on Duda sites, not just WordPress. Claro runs the project and pulls in whoever the job needs, Pixel on design, Navi on UX, Index on SEO, Lexi on content, Glitch on functionality.

Try:

"Check when this site was last backed up before you change anything."

"Update the hero headline on this Duda site to match the new positioning."

Or put it on repeat. Weekly backup and verification checks every site has a recent, restorable backup and warns you the moment cover slips. It now covers all five of those tools.

Security: update the WordPress plugin to 5.1.2

A researcher told us about a security hole in our WordPress plugin, privately and properly. It affects version 5.1.1 and earlier and is fixed in 5.1.2.

Someone with an author account on your site could have used it to delete files they should never have been able to touch. File actions now check exactly where they're pointing before they do anything, the fields that made the trick possible now reject it, and we've closed a back door that let the AI layer be reached without going through Atarim's login.

Please update to 5.1.2. If you look after a lot of sites, push this one out today.

Thanks to the researcher who came to us first instead of posting it publicly.

Improvements

• Keyword trend numbers now say what they mean. That tool compares keywords against each other, so 80 means popular next to the others in the same question. It does not mean 80 searches. We had it labelled as search volume. If you planned content around those numbers, give them a second look.
• Share links that say what they are. Collaborate links used to unfurl in Slack with the same generic Atarim sentence. They now show the document name and your workspace, for every project type.
• The AI checks its own work. Links in generated posts get opened before the post ships. Proof that a change landed comes from a fresh screenshot, not an old one. Documents carry today's real date instead of a guess. And workflow steps now reliably follow the instruction you wrote for them.
• Elementor edits tell you if something went missing. Reading a page warns you when it's only seeing part of it, and saving reports how many elements actually saved. Before, a save that dropped widgets still said it worked.
• The MCP reports what it really did. Writes used to say success while quietly dropping things they couldn't set. They now tell you what saved and what didn't.
• Guided tours inside the app. Beacons point out the things you haven't opened yet, at the moment they're relevant.
• Your renewal date, when you go to cancel. Pulled live from billing, so you're not guessing what you've already paid for.

Bug fixes

• AI reviews run against the right website. Leave the website field blank in a project brief and Atarim used your email domain instead, then reviewed that. It's now required.
• Tasks don't get stuck on "Thinking". Finished reviews left tasks spinning forever. Comments from another session now show up live, and your own no longer posts twice.
• Prebuilt prompts work. Picking a suggested prompt in chat no longer hangs.
• Copying tasks between projects. This threw a server error. It now copies pages, tasks and comment threads properly.
• Inbox bulk delete. Deleting a page of emails took so long your browser gave up, so they stayed on screen even though they were gone.
• Screenshots of images from other sites. A hero section loading images from another domain gave you a grey box. It captures what's really there now.
• Milestones update live. No more refreshing after you complete or edit one.
• Collaborate tidy-up. The cookie banner no longer throws you back to the homepage, external links stop landing on a blank page, and visiting logged out no longer pops a sign-up form at you.

Coming next

The new Boards are built and waiting behind a flag while we finish testing. Custom boards, your own columns, drag to reorder with undo, a backlog rail, and boards for pages and projects instead of just tasks. Close now.

If you've had a report from us that looked right and turned out not to be, I want to hear about it. That's exactly what this release was built to catch.

🤘

The AI team got cheaper to run, learned some real new skills, and got a lot stricter about what it's allowed to claim. There's also a WordPress plugin security update in here, so if you skim nothing else, skim that part.

New features

AI actions now cost about 40% less

Same reviews, same workflows, same Do It actions. Your credits just go a lot further. Already live, nothing to change.

You stop paying for AI that didn't deliver

Stop a turn halfway and you used to get charged anyway, then the reply would turn up a minute later. A turn that came back with nothing still cost you, and nothing told you it had failed. Now stopping a turn costs nothing and posts no late reply. An empty turn shows as failed with a retry button. We also added a credit check partway through long runs and a limit on repeat calls, so nothing loops quietly in the background.

Index can now do SEO the way an agency does

Writing a title tag is the easy part. Any tool can fill that box in. Knowing what to put in it is the actual job, and that takes research.

Index could already edit your SEO plugin. He just had no way to find out what he should be writing. So now he can do the research first.

What he can look up:

• Search volume - is anyone actually searching for this
• Keyword difficulty - can you realistically win it
• Keyword overview - the full picture on one term
• Keyword ideas - turn one term into a content plan
• Search intent - what the person searching actually wants
• Keyword trends - what's climbing, and what's dying
• Live search results - exactly who you'd have to beat
• Backlink profile - every link pointing at a site
• Referring domains - which of those links are worth chasing
• Competitor discovery - who you're really up against in search
• Keyword gap - what they rank for that you don't
• Domain overview - the baseline you report back to clients
• Ranked keywords - every term a site ranks for today
• Page audit - prove a fix worked, before and after
• Page content - pull apart a competitor's page in seconds
• Local listings and location targeting - go country, region or city deep

And then he acts on it. Index can change site-wide settings in AIOSEO, Rank Math and Yoast: title and meta templates, separators, brand details, redirects, module toggles. One named change at a time, on a request you typed, and he reads the value back afterwards to prove it saved. On an automated run he'll refuse and tell you why, because site-wide settings shouldn't change while you're asleep.

Research, decision, change. The whole loop, in one place. Try asking him:

"Using live UK search data, find 20 keywords this page could realistically rank for. Give me the real monthly search volume and difficulty score for each, and mark the ones we could win this quarter."

"Using live UK search data, who are this client's top organic competitors, and what are they ranking for that we're not?"

"Audit this page against the top three UK results for our main term, then fix the title and meta based on what you find."

Swap UK for whichever market your client sells into. Name it, or you get estimates instead of real numbers.

Or put it on repeat. Content plan and keyword analysis, Monthly keyword ranking report and On-page SEO review are all in the workflow library, and they all run on this new research.

Reports that have to show their working

We ran a monthly security scan 77 times. It reported success every time. On one of those runs it listed two vulnerabilities in a client's report that do not exist.

The scan knew what to say when it found problems, and what to say when it found none. It had nothing to say when the feed didn't answer at all. Asked to fill a table it couldn't fill, it filled it in anyway.

Now every workflow step that writes a document leaves a trail behind it. Each number sits next to the data it came from, each fact next to the tool call that found it. You still read a clean report, and we check it against the trail. Nothing unreadable gets called clear.

For the security scan specifically: a feed it can't read is now reported as unreadable. A version it couldn't check is called unchecked, not clear. And every vulnerability listed has to come from a real result in that same run.

Try:

"Run the security scan on this site and show me the source behind every finding."

Or put it on repeat. The Monthly security scan workflow is the one this story is about, and it's the one that now has to prove every finding.

Turning AI off now actually turns it off

Queued runs used to carry on replying after you'd switched AI off. For one agency that meant AI answers landing on live client support tickets.

We now check the setting again right before a run starts, not just when it goes into the queue. Off means off, including for anything already waiting.

New skills

• Backups, handled by Glitch. All-in-One WP Migration, BlogVault, Solid Backups, UpdraftPlus and Migrate Guru. Whichever one a site uses, he can check the backup before anything gets changed.

• Zapier. Ask Claro to fire off a single action in an app the client already uses, without building a whole workflow for it.

• Duda. Work on Duda sites, not just WordPress. Claro runs the project and pulls in whoever the job needs, Pixel on design, Navi on UX, Index on SEO, Lexi on content, Glitch on functionality.

Try:

"Check when this site was last backed up before you change anything."

"Update the hero headline on this Duda site to match the new positioning."

Or put it on repeat. Weekly backup and verification checks every site has a recent, restorable backup and warns you the moment cover slips. It now covers all five of those tools.

Security: update the WordPress plugin to 5.1.2

A researcher told us about a security hole in our WordPress plugin, privately and properly. It affects version 5.1.1 and earlier and is fixed in 5.1.2.

Someone with an author account on your site could have used it to delete files they should never have been able to touch. File actions now check exactly where they're pointing before they do anything, the fields that made the trick possible now reject it, and we've closed a back door that let the AI layer be reached without going through Atarim's login.

Please update to 5.1.2. If you look after a lot of sites, push this one out today.

Thanks to the researcher who came to us first instead of posting it publicly.

We tightened up crash reports while we were in there. Emails, web addresses and tokens now get stripped out in your browser before a report is sent, so nothing identifiable leaves your session.

Improvements

• Keyword trend numbers now say what they mean. That tool compares keywords against each other. A score of 80 means that keyword is popular compared to the others in the same question. It does not mean 80 searches. We were labelling it as search volume, which made it look like a count. If you planned content around those numbers, they were comparisons, not totals. Worth a second look.

• Share links that say what they are. Collaborate links used to unfurl in Slack with the same generic Atarim sentence. They now show the document name and your workspace, for every project type.

• The AI checks its own work. Links in generated posts get opened before the post ships. Proof that a change landed comes from a fresh screenshot, not an old one. Documents carry today's real date instead of a guess. And workflow steps now reliably follow the instruction you wrote for them.

• Elementor edits tell you if something went missing. Reading a page warns you when it's only seeing part of it, and saving reports how many elements actually saved. Before, a save that dropped widgets still said it worked.

• The MCP reports what it really did. Writes used to say success while quietly dropping things they couldn't set. They now tell you what saved and what didn't.

• Guided tours inside the app. Beacons point out the things you haven't opened yet, at the moment they're relevant.

• Your renewal date, when you go to cancel. Pulled live from billing, so you're not guessing what you've already paid for.

Bug fixes

• AI reviews run against the right website. Leave the website field blank in a project brief and Atarim used your email domain instead, then reviewed that. It's now required.

• Tasks don't get stuck on "Thinking". Finished reviews left tasks spinning forever.

• Comments from another session now show up live, and your own no longer posts twice.

• Prebuilt prompts work. Picking a suggested prompt in chat no longer hangs.

• Copying tasks between projects. This threw a server error. It now copies pages, tasks and comment threads properly.

• Inbox bulk delete. Deleting a page of emails took so long your browser gave up, so they stayed on screen even though they were gone.

• Screenshots of images from other sites. A hero section loading images from another domain gave you a grey box. It captures what's really there now.

• Milestones update live. No more refreshing after you complete or edit one.

• Collaborate tidy-up. The cookie banner no longer throws you back to the homepage, external links stop landing on a blank page, and visiting logged out no longer pops a sign-up form at you.

Coming next

The new Boards are built and waiting behind a flag while we finish testing. Custom boards, your own columns, drag to reorder with undo, a backlog rail, and boards for pages and projects instead of just tasks. Close now.

Sixty-one skills, nineteen prompts and forty-three tools went through this pass.

If you've had a report from us that looked right and turned out not to be, I want to hear about it.

That's exactly what this release was built to catch.🤘

Two releases, one clear direction: fewer places where Atarim gets in your way. Collaborate now loads sites that used to sit there frozen, nine languages finally show real sentences instead of raw code, and clearing a backlog takes one click per task instead of five.

New Features

Triage a whole backlog with right-click

Right-click any task card in Inbox, Boards or Collaborate to set its status and priority, pin it, mark it read or unread, copy its link, open it in another view, or delete it. It works the same way in every list, and the change shows straight away rather than after a pause.

If you start your day clearing a backlog, this is the change you will feel most.

Enter your own coupon code

You can now enter a coupon code when you upgrade, and the credit lands on your account immediately. No emailing us to have it applied by hand.

Pick monthly or annual on our website and the upgrade screen opens on the choice you already made, instead of resetting to the default.

A comment box with one clear send button

The comment thread has a single tidier layout and one send button that holds the rest of the options: reply as a comment, send it to Claro, or leave an internal note your client cannot see. Attaching a file, annotating a screenshot and dropping in a canned response now live together in one menu instead of scattered around the edges.

Nine languages that read as actual sentences

Some screens were showing raw internal labels instead of real text. A Spanish customer who hit the upload limit saw a code string rather than a message telling them what had gone wrong. Fifty-eight of those placeholders were reaching customers across Spanish, French, Arabic, Hebrew, Swedish, Norwegian, Dutch, Finnish and Bulgarian. There are now none.

If you have ever given up on Atarim in your own language, it is worth another look.

A rebuilt Settings area

Settings has new Billing, Security, Canned Responses and MCP pages. Security and notification preferences now save themselves as you change them, while AI Settings and your branding keep an explicit save button so nothing changes under you by accident. Every row in the user permissions table has a tooltip explaining what it actually controls.

Improvements

• Empty Inbox and Boards explain themselves. A new account no longer opens onto blank panels. Each empty view now says what it is for and what to do first.

• More room on every screen. The top navigation bar is gone, which hands every screen more vertical space, and spacing is consistent across sidebars and headers. Dark mode now covers screens that used to stay stubbornly light, including the error page.

• Tags can be more than one word. Typing a space no longer cuts a tag short, so "needs design review" stays a single tag.

• Status and priority work on a phone. The pickers no longer run off the edge of a small screen.

• A locked template offers you the upgrade. Previewing a template your plan does not include now takes you into the upgrade flow instead of simply refusing.

• Cancelling to the free plan is one step. No checkout window opens just to downgrade.

• Enter submits. Inputs across the app now accept the Enter key.

• The project brief overview has been rebuilt.

• Enterprise: your IT team can allow Atarim through the firewall. Our traffic now identifies itself as Atarim. Previously it was indistinguishable from anonymous traffic, and it arrived from whichever data centre was nearest the person opening the link, so one customer watched us appear from Ireland and Sweden inside an hour and had no safe way to write a rule for it.

Bug Fixes

• Pages that loaded but ignored every click. On some sites the cookie consent banner hid itself as intended and left the rest of the page frozen behind it. Everything looked completely normal and nothing responded. Two separate versions of this were found and both are fixed, including one affecting large sites that build their banner in a way our previous approach could not reach.

• Images missing when you viewed your own site. Two unrelated causes, both ours. One broke image addresses containing certain characters, which took out hundreds of images on a single news homepage. The other had image services rejecting us as an unauthorised embedder.

• Tasks on sites without a certificate. Creating a task did not work at all on a site served over plain HTTP. Staging sites and older properties are back in play.

• Pages your own server called an error. Some sites serve a perfectly good page while technically reporting a failure. We showed our error screen instead of your page. Now the page loads.

• Tasks that vanished in the Chrome extension. On certain sites, particularly ones built with older page builders, a task's marker never appeared on the page, so the task looked lost. It always existed, and now it shows. The task popover also stays open after you create a task instead of closing on you.

• Comments overwriting each other. With two people active on the same task, an incoming comment could save against a stale copy of the thread. Pasted screenshots now attach as proper files rather than bloating the comment.

• Collaborators can use the AI features again. Task AI chat and AI review were always meant to include collaborators, and were being refused. Glitch is back within reach for them. Guests stay excluded, as designed.

• Crashes on the project screens. The Inbox project dropdown could take the page down, and the project brief could crash when a brand kit came back in an unexpected shape or when you opened its last tab. All resolved.

• Invites and client details. Inviting an existing team member works, renaming a client no longer throws an error, and removing an invited user actually removes them.

Smaller things.

• Archived project search pages correctly.
• Empty technical detail rows are hidden rather than shown blank.
• Subtask thumbnails fall back to the page screenshot instead of a broken image.
• The AI thinking indicator is clickable again.
• The client screen has the right help link and working single-email entry.

These four are are available in beta, ask us and we will switch them on for you.

• Collaborate Preview opens the real, live page inside Atarim next to the chat panel, so feedback, chat and the actual website sit in one view. Switch between browsing and commenting, check the page at different device sizes, set its design status, and share it. Desktop only for now.

• A redesigned Projects screen built around Web, Artifacts and Design, so a project's state reads at a glance instead of needing a click.

• The Collaborate sidebar, where the side panel becomes Chat, Feedback and Assets. Feedback lists that project's tasks and updates live as your team works, with search and sort. Assets lists pages and artifacts, and clicking a page opens it in the preview.

• Workflow automation templates, a browsable library for setting up repeatable automations.

The first release since V5 launched, and it has one job: closing the gap between you and the work. Clients land where the work is, AI answers show you the actual thing instead of describing it, and reviews come to your inbox instead of making you go find them.

New Features

Your clients now land in their project, not a lobby

When a client finishes onboarding, the final step now says "See My Project" and means it. They arrive on their project dashboard with the onboarding conversation already open, so their first impression is the work itself, not a welcome modal they have to click past. The wizard also stays properly full-screen the whole way through.

AI answers now show you the thing, not just talk about it

Ask Claro about a task, a page, a project or an artifact, and the answer now includes the actual card, right where it belongs in the sentence. No more reading a description and then hunting for the thing being described. This works everywhere you chat, so the answer looks the same whether you are in a project or the workspace.

Watch tasks move while you chat

Task cards in workspace chat are now live. When a specialist is working on a task, the card reflects the run as it happens, and you can expand it inline to read the full comment thread without leaving the conversation. You stay in one place; the status comes to you.

Preview artifacts without leaving the Projects screen

Artifacts now open read-only, in place, right on the Projects screen. Skim a deliverable, confirm it is the right one, and use the breadcrumb to hop back to the grid or into the source project. On smaller screens you get the full Collaborate view instead, so nothing gets cramped.

Reviews now live in your inbox

Review threads render inline in your inbox list, in their natural position, so you can open a review from the same place you handle everything else. Glitch's review subtasks show up in the list and in the review session too, so the QA trail is all in one view.

Improvements

• Deliverables are now Artifacts. Same work, clearer name, updated everywhere you see it. If you go looking for Deliverables, look for Artifacts.
• Subtasks are visible on the board. Kanban cards now show a subtask badge and counts, with a one-click jump to the parent task.
• AI comments are read-only. Specialist comments can no longer be edited or deleted, so the record of what was done stays intact.
• Single-word names work at signup. Registration now accepts any name, no more inventing a surname to get through the form.
• Chat feels smoother. Fixed scroll jitter while answers stream in, and the reasoning panel now hands over cleanly instead of flickering.
• Dark mode and overlay polish. The add-website modal now looks right in dark mode, and previews and confirmation dialogs layer correctly instead of hiding behind other UI.

Bug Fixes

• Your brand no longer follows you out the door. Logging out of a whitelabeled workspace now restores the default login page, colours and logo included.
• OTP verification survives a refresh. Reloading mid-verification now resumes exactly where you were instead of dropping you into an old flow.
• Role checkboxes toggle properly during onboarding. One click, one toggle, as it should be.
• Sidebar spacing is back to normal after client activation. The first nav item no longer crowds the workspace logo.
• Locked buttons keep their own styling. The Pro badge no longer restyles buttons it sits on.
• Pages are classified correctly as artifacts. A detection fix means pages no longer show up as the wrong type.

It's been months in development, and we're very pleased that it's finally here!

As of today, Atarim subscribers can activate the Atarim V5 beta!

This is the best thing we have ever built, and you get it before anyone else.

V5 introduces the AI agency that plugs into yours.

It does real agency work, proactively identifies things it should work on, and makes sure nothing reaches a client without your thumbs up.

🤘 Introducing The AI Agency, Plugged Into Yours

You know Atarim as the place your website feedback, projects and workflows live.

V5 keeps all of that. Except now, Atarim does the work.

And not only that. Atarim works the way agencies work, too.

• It thinks in tasks, approvals, pitches and workflows, because that’s how the best agencies in the world run.
• When one of the AI picks up a piece of work, they raise it as a task you can see, question and iterate on, the way real colleagues do.
• Your clients get an AI of their own to ask, explore and suggest ideas with, inside your dashboard, while what actually ships stays your call.

And you don't have to ask! With Atarim V5, the AI goes looking and finds opportunities and things you didn’t have time to search for, or didn’t know to.

What Can Atarim V5 do?

Point Atarim at any site on your books and it finds the problems and the opportunities for your clients, suggests the changes to you, and can do them for you.

It designs. It writes copy and creates imagery.

It builds new pages, optimises performance, writes and repairs code, and takes it live.

It can do almost whatever you ask for, and plenty you haven’t thought to. Just like a real team.

The Inner Circle Now Collaborate and Deliver Work Like a Real Team

The work is done by six expert AI specialists, one for each discipline a web agency needs: design, copy, SEO, UX, engineering and project management.

You know those AI experts as the Inner Circle, and they can collaborate with each other as well as you. Talk to them like real colleagues. They now work like them.

Claro, your AI project manager, holds every detail of every client. Ask him anything.

And when the AI spots an opportunity on a client’s site, it lands on your desk scoped and priced at your rate, ready to pitch.

As for everything else, the plugin updates, the backups, the broken links, the uptime checks, the SEO sweeps?

Let the AI handle it for you and treat every client like they're your best.

Workflows have had a MAJOR overhaul

Behind Atarim sits a library of hundreds of ready-made workflows, based on how thousands of agencies actually serve clients: the 30-day check-in, the backup before the update, the report before the invoice.

And now, Atarim can suggest the right ones for every one of your clients.

Even the monthly blog post, written based on keywords the AI found on the website, in the same tone of voice with a brand-new featured image matching their brand guidelines.

Need a workflow that doesn’t exist? Just describe it and the AI it will build it for you.

Go and Try it Out!

No waitlist or early access to join this time! Just click the V5 toggle inside your account (you'll see it at the top of the app).

🔴 And if you haven't already, install or update the Atarim WordPress plugin.

We'll be opening up wider access as we reveal Atarim V5 to the world very soon.

This is a huge release as we move from just managing feedback to taking creative action!

Click Do It on any comment and your idea is live on the page in seconds with our WordPress plugin. No drafts, no queues, no waiting on anyone. Not ready to commit yet? Hit Show Me to preview the change in place, keep what works, roll back what doesn't, all without leaving the page. The gap between AI suggestions or client feedback and seeing it shipped just disappeared. We also squashed some more pesky bugs.

From Feedback to Creative Action

This is the one. Click Do It on any comment and the change is live. Not drafted. Not queued. Not "added to the sprint." Live on the page, while you're still looking at it.

The headline that's been bugging you for a week. The button nobody can find. The paragraph the client keeps redlining. Done. Live. Moving on. In the time it takes to read this sentence. Not great? Click undo and it's like nothing happened.

Works inside your WordPress sites today, with more platforms rolling out soon. The next time someone asks how long a "quick change" will take, the answer is: it's done.

New Features

“Show me” + “Do it” - See it before you commit to it (in Beta)

Not ready to pull the trigger? Hit Show Me first and watch the change appear on the page like a preview. Keep it or put it back. No mockups, no Loom recordings, no "can you send a screenshot of what you mean."

The guesswork between "I have an idea" and "that's exactly it" just disappeared.

Pin the tasks that matter

You know that one task you keep scrolling to find? The blocker, the urgent client request, the thing you're tracking all week.

You can now pin it straight from the sidebar and it stays at the top. No more scanning your backlog every time you want to check on it.

A real "what's next" moment after Figma imports

Importing from Figma (using our Figma plugin) used to drop you back into the dashboard with a vague sense of "okay, now what." Now you land on a clean next-steps screen with your project thumbnails and clear actions: copy the share link, jump straight into the project.

The handoff feels smooth.

Smoother WordPress plugin activation behind the scenes

WordPress Plugin activation now passes a new mcp_token along to your redirect URL, which sets us up for some bigger AI-meets-WordPress stuff coming soon. You won't notice it day-to-day, but it's the plumbing that makes the next wave possible.

Improvements

Comment cards that feel right

We weighted the comment card animations and locked the task panel in place so things don't jump around as you click between comments. Mirrored the same polish over to Graphics so the two experiences feel like the same product. Small thing but a big difference in how the app feels in your hands.

Color picker that reads the way your brain expects

Within comments, the color swatch now sits to the left of the hex value, the way every other design tool shows it. We also added color highlights and inline emojis to make the comments feel a bit more alive.

Project Settings is now admin-only

Project Settings has moved behind admin permissions. This closes a gap where team members could get into settings they shouldn't have been touching. Cleaner permission boundary, fewer accidental changes.

Disabled projects actually stay disabled

If a project is set to collaboration-disabled, clients can no longer create new tasks on it. Existing tasks stay commentable so nothing in flight gets stuck, but the project respects its state the way you'd expect.

Brand new Analytics You Can Read at a Glance

Your Activity Trend is back in the picture. That graph under Customize Graphs that's been sitting empty? It's finally showing you what your team's been up to. Open it up and you'll see the rhythm of the week, when things picked up, when they slowed down, the patterns worth paying attention to.

Dark mode that's actually easy on the eyes. If you work in dark mode, the Set Stage modal in analytics has been rough, washed out text, contrast that made you lean in. It now matches the rest of the app, so you can stay in dark mode without switching back to light just to read a screen.

A ratio that finally explains itself. Client vs Team contribution used to be a number with no context. Hover over it now and you'll see what it actually means, who's driving the work, where the balance sits, and whether things are trending the way you want. Useful before a client call, useful in one.

Bug Fixes

• Share links no longer open to a white screen. Fixed the extension handoff race that sometimes left collaborators staring at nothing when they opened a share link in a new tab.
• Graphics delete modal now speaks your language. Translations now apply to the Graphics version delete flow instead of falling back to English.
• Project Settings permission boundary closed. Team members can no longer slip into Project Settings (see above under Improvements).
• Collaboration-disabled projects respect their state. Clients can't create new tasks on disabled projects anymore.

Two releases. One clear direction: less friction between you and your work.

Whether you're running an AI review, collaborating with a client, or managing a packed project backlog - This update makes Atarim feel faster, smarter, and more in tune with how you actually work.

New Features

A visual project brief that captures everything about your brand

Click the Project Brief button on any project and Atarim now builds you a rich, visual overview of everything the AI, and your team, needs to know: brand colours, typography, project goals, unique value proposition, what success looks like, inspiration sites, services and products, target audience, and brand voice. All of it, surfaced and organised in one place with a single click.

It's not just a brief, it's the full context layer that makes every AI interaction smarter from the start.

You decide who works with AI - and how

Not every team member, and not every client, should have the same relationship with AI. A new permission setting lets you control whether AI mode is on or off by role. Admins and team members get it on by default. Clients and collaborators don't, unless you choose otherwise. Clean, intentional, yours to control.

No more logging into WordPress twice

When you open a WordPress project from the Atarim project screen, you're now authenticated automatically. One less prompt, one less interruption - you're just in.

No more sending empty messages by mistake

If someone writes "see attached", "find the file below", or anything that implies a file, but doesn't actually attach one, Atarim flags it on the spot before the message goes out. It's a small thing that prevents a surprisingly common and frustrating back-and-forth.

See task status without opening anything

Task cards in the sidebar now show status and priority directly. You can scan your whole backlog at a glance and know exactly where things stand - no clicking in and out of individual tasks.

AI that remembers the conversation

When you open AI chat inside a task, Claro now has the full comment history as context - not just the task title. The responses you get back are grounded in the actual conversation that's happened, making them far more useful.

Improvements

Everyone's cursor is visible by default

Everyone's cursor is visible by default - Live collaboration cursors now appear automatically. If you're working together, you'll see it.

A sidebar that finally feels right

Spacing, proportions, drag handles, task counts, layout ratios after AI handoff - a wave of polish that adds up to an experience that feels significantly tighter and more intentional.

AI review progress that matches reality

The progress counter during AI reviews now moves in sync with when results actually appear. No more hitting 100% and then staring at a spinner.

The same filters everywhere

The filter panel in the sidebar now works identically whether you're on a web project or a graphics project. One less thing to re-learn.

Smarter sharing with large teams

The Share modal now paginates long user lists. Finding the right person is no longer a scroll marathon.

Custom email content in workflows

You can now write custom HTML body copy for your Send Email workflow steps. More control over what goes out when automations fire.

Dark mode, done

The last remaining dark mode inconsistencies have been resolved. It looks the way it should, everywhere.

Fully translated Collaborate experience

All strings in the Collaborate view are now properly translated across supported languages.

Bug Fixes

• Figma Plugin: Fixed authentication failures, broken share buttons, and wrong-workspace redirects when using the Figma plugin
• Inbox: Fixed the inbox not switching when you change workspaces
• Chrome extension: Fixed the extension dropping out when you opened the Atarim dashboard in another tab mid-session
• Collaborate authentication: Fixed users showing as unauthenticated when navigating between pages, even with an active session
• Email settings: Fixed a critical block preventing the email settings page from opening at all

• Thumbnail refresh: Fixed the refresh action failing silently
• Graphics task counts: Fixed badge numbers not matching the actual tasks visible in graphics projects
• AI chat auto-scroll: Fixed the chat not scrolling to the latest AI reply automatically
• Onboarding modal: Fixed it reappearing after you'd already completed setup
• Graphics task creation: Fixed being unable to create a second task without manually closing the first

This release brings several workflow improvements to help you collect feedback more clearly, recover tasks more easily, and collaborate faster across websites, graphics projects, and the WordPress plugin.

New Features

Reviewer greeting for feedback sessions

When someone opens a Graphics project to review it, they now automatically see the latest open General task as a greeting message. This lets you guide reviewers with instructions on what feedback you need, helping you receive clearer and more useful comments from clients or teammates.

Remap tasks to a different element on the page

If a page changes or an element moves, you can now remap a task to a new element directly from the tech info panel. This helps recover “lost” tasks after redesigns and keeps feedback connected to the correct place on the page.

Guest mode in the WordPress plugin

Guest mode is now integrated into the WordPress plugin, allowing visitors to leave feedback without logging in. This makes collecting feedback on staging sites or review environments much faster and easier.

Improvements

• Completed tasks hidden by default in Graphics projects – Keeps the workspace cleaner and helps reviewers focus on what still needs feedback.
• Workspace-aware project activation via plugin – Projects added through the WordPress plugin now activate in the correct workspace, preventing confusion when teams work across multiple workspaces.
• Slack links now switch to the correct workspace automatically – Clicking project links from Slack now takes you to the right workspace instead of loading the last one you used.
• AI chat now opens in AI mode by default – The sidebar remembers your AI mode preference so you can jump back into AI assistance without switching modes each time.
• Clearer sidebar navigation and labels – Updated icons and labels help support additional content formats and make navigation easier to understand.
• Cleaner device preview experience – Removed the unnecessary “device preview out of bounds” warning for a smoother preview workflow.
• Improved guest flow in Graphics projects – Updated labels and messaging make it clearer when users are collaborating as guests.
• Improved team member visibility in People screen – Pending invitations and active users are now clearly distinguished so you can easily manage your team.

Bug Fixes

• Fixed session drops when navigating pages in Browse mode – You’ll no longer be switched to Guest User while navigating a site, preventing failed comments or lost work.
• Fixed issue preventing multiple tasks from being created without refreshing – You can now add tasks consecutively without overloading the page.
• Improved task and comment publishing speed – Tasks and comments now appear almost instantly instead of taking several seconds to publish.
• Fixed task status and priority colors not updating immediately – Status changes now reflect instantly without requiring a refresh.
• Restored missing design tasks after integration updates – Previously hidden design tasks are now visible again.
• Fixed canned responses not loading correctly in the new Graphic Feedback Tool – Saved responses now load and insert their content as expected.
• Fixed translation settings reverting to English – Your selected language now persists correctly across sessions.
• Fixed People list sometimes showing no users – The People screen now reliably displays all users.
• Fixed Chrome extension annotation issues – Annotations now work correctly when using the Chrome extension.
• Fixed Chrome extension visibility of internal tasks – Admins and team members can now see internal tasks correctly in the extension.
• Fixed cookie consent banner blocking AI review – AI review now resumes correctly after a cookie banner is dismissed.
• Fixed internal tasks opening AI chat instead of task details – Internal tasks now open the correct task details view.
• Fixed annotation screenshots capturing UI elements – Screenshots now capture only the page content without the comment input overlay.
• Fixed plugin activation issue that could add projects to the wrong workspace – Projects added via the plugin now attach to the intended workspace.
• Fixed comment editor not clearing after publishing – The comment field now resets properly after posting.
• Fixed Slack form submission links loading the wrong workspace – Form submission links from Slack now open the correct submission.