What we shipped

Early searches are handled without error.

Draft guidance appears only when a draft is returned. After saying "Yes" to a recurring-report offer, Active Intelligence previously gave activation instructions even when no draft had been created. Activation guidance appears only alongside an actual saved draft.

An ambiguous "yes" prompts a clarifying question. When a report with no data asked both a data question and a recurring-offer question, a plain "yes" could mean either. Active Intelligence asks which one you meant rather than acting on an assumption.

Campaign chips reach the Scheduled Tasks agent. When you attached a campaign chip and said "it's this one," the agent previously lost the chip context and asked "which task?" repeatedly without creating anything. The chip's context is passed through correctly, so the draft is created for the right campaign without follow-up questions.

A replies widget appears on the Automation report details page for any SMS block, listing the Contacts who replied to that message. It sits above the failures widget and pages through recipients five rows at a time. A link to view the full list of recipients will be available when the Automation recipients page ships.

The "SMS performance" breadcrumb on the SMS broadcast recipients page links to the SMS campaigns reports list.

Previously, if the accounts grid was on page 2 or a sorted view, the impersonated session would land on that same page and sort state, often showing an arbitrary or empty slice of a different account's data. Table page, sort field, and sort direction are now cleared before the page reloads when switching accounts. Filter parameters such as tags and status are preserved.

Any cookie previously used to access the Billing tab has no effect going forward — the tab is simply available to all partners in production.

MCP gateway: Automation Canvas links and HTML email import

Two additions for external integrations using the MCP gateway.

Automation Canvas links. After generating or editing an Automation, the run result includes a direct URL to open that Automation in the Automation Canvas. For Automations with email blocks, the link focuses the specific email block.

HTML email import. The Campaign generation endpoint accepts an optional field containing a complete email document, which is imported as the Campaign exactly as provided. The document must be a single complete HTML email, up to 500,000 characters.

The Campaign generation endpoint accepts an optional field containing a complete email document (up to 500,000 characters), which is imported as the Campaign exactly as provided. When an inline HTML design is included on an import request, it takes precedence over every other attached format.

Fidelity for hand-authored HTML is unmeasured — whether arbitrary markup decomposes into editable blocks or lands as a single block is untested.

Several fixes to the links management experience:

• Storefronts are read in full, so a link used in a later storefront is correctly counted in "Used in," the "Not used" filter, and the delete warning.
• Changing a search or filter clears the current selection; a link deleted from its row is also removed from the selection. Bulk Copy and Delete act only on links currently in view.
• A second Edit on the same link opens its form again after a Save or Cancel.
• Where-used filter menu rows are disabled while the data cannot be read.
• Focus returns to the element that opened a panel when the panel closes.
• Escape closes the panel only from within it, and not over an unsaved link or an open edit.
• Click counts, share-of-clicks calculations, and chart labeling are corrected.
• A failed save displays the service's error reason below the toast.
• Bulk delete refreshes the list once after all deletions complete, not once per link.

When setting up a Form with a Language field mapped to an existing field source, reopening that field step previously lost the mapping and blocked the Next button. The mapping is now retained.

The landing pages editor fills the entire viewport when open. The navigation bar and header no longer appear around the editor, giving you an unobstructed workspace.

Inserting a personalization tag or emoji into a Campaign subject line saves correctly and propagates to translated variants on translation-enabled Campaigns.

Previously, all three surfaces — the takeover, the API key modal, and the onboarding fallback — showed a generic "invalid key" error, causing accounts with IP security enabled to keep trying other keys. The error message now tells you specifically that the connection was blocked by your source account's IP security settings, and instructs you to check your email for a blocked-IP notification from that provider, authorize the IP, then try again.

The table shows Post, Listening, Comments, DMs sent, Reply rate, Likes, and Posted date. You can search, filter by Flows, Type, Channel, or Hidden posts, choose which columns to display, and set rows per page. Each row has a menu to view the post on Instagram, create a new flow for that post, or hide it from Engage. Bulk hide and show are supported with an Undo option.

Above the table, a summary strip shows Posts by type, Listening, Comments, and DMs sent, along with a sync timestamp and a manual sync button.

Selecting a row opens a post detail panel alongside the table, showing the account, picture, full caption, four engagement figures, and the flows answering that post with match details. "New flow for this post" opens the Automation Canvas on a comment trigger with that post pre-selected.

Facebook Page comment counts remain at 0 pending a backend update, and recent comments are not yet shown in the panel.

At the top of the page, a metric strip shows links in use, links not in use, clicks, conversions, and revenue. The table below is searchable and filterable, with bulk copy and delete, and a one-click copy button on every row.

Clicking any link opens a panel that slides in from the right. Inside: the short link with a copy button and QR code download (PNG, JPEG, WebP, or SVG); a 12-week clicks chart; editable destination and tracking fields; a "Used in" section listing the flow steps and storefront cards that use the link; and a recent activity feed showing matched Contacts. Creating a new link uses the same panel.

A few things worth knowing: the activity feed is labeled "Recent activity" rather than "Recent clicks" because conversions appear as their own rows. Revenue is shown as a plain number with no currency symbol, consistent with how Insights and Storefronts display it. Referral links are not shown on this page. The column for each link shows when it was created, not when it was last clicked.

The subject field in the email composer drawer on deal and contact record pages is inset to match the alignment and styling of the message body field, and the focus ring draws on the field's own border rather than spanning edge-to-edge across the drawer.

Contacts card moves to the top. The Contacts card appears at the top of the right column on the deal detail page. Previously it rendered below all custom fields, so accounts with many fields required scrolling to reach it.

Required fields are marked. Deal Title, Deal Value, Deal Owner, Primary Contact, and any required custom fields display a red asterisk after their label. Pipeline, Stage, and Status remain unmarked, matching the previous behavior.

Custom field groups collapse and expand. Custom field groups on the deal detail page can be collapsed and expanded. Previously all groups were always shown expanded with no way to collapse them.

Time in stage. The deal detail page shows how long a deal has spent in each stage. The stage picker displays the current stage with its duration — for example, "To Contact ( 2 days )". The stage menu shows time for each stage the deal has visited; unvisited stages show the name only. The Overview stage field also shows current stage time.

Open task rows show more detail. Open task rows on the Tasks tab show the task type before the title (for example, "Call: Gap call"), the exact due date and time in your account's date and time format in your browser's time zone, and the task note truncated at 100 characters. Completed task rows are unchanged.

Previously, URLs appeared as plain text and only a relative age was shown on note timestamps. Note timestamps show the full date and time in your account's configured format and time zone, followed by the relative age — for example, "10/01/2026 02:45 (a few seconds ago)".

Contact rows in the Contacts card on a deal details page display each contact's phone number alongside their email address, with the phone number appearing first. Rows with neither phone nor email omit the contact line entirely.

Content Manager: custom crop, full-folder carousel, and upload progress

Three significant additions to the Content Manager image editor.

Custom crop. A crop tool supports drag, resize, and ratio locks — Custom, Square, 4:3, 3:2, 16:9, and 4:5. Pressing Apply saves the crop as a single undo step and closes the crop panel. Crop corners are visible against any background. Flip, Undo, Redo, Reset, and Save are disabled until a crop is applied or exited.

Full-folder carousel. The image carousel walks your entire folder rather than stopping at the first 20 images. Next, Previous, arrow keys, and the thumbnail strip all load more images as you go, with pages prefetched near either end. The counter shows your position within the whole folder. Opening an image from a link or on reload locates it within up to 200 images. Every folder has its own bookmarkable URL.

Upload progress panel. A panel in the bottom-right corner shows per-image and overall upload progress. Progress reaches 100% when bytes are sent and stays visible until the save is confirmed. You can cancel individual uploads or all at once, collapse or close the panel, and see failure reasons — files over 10 MB can't be retried; network or server failures can. The panel hides itself 5 seconds after a successful finish unless there are failures. Library and folder lists refresh once per batch rather than once per file.

Additional improvements: a new folder scrolls into view and clears any active search; AI-generated folders use a sparkle icon; Delete is available directly from the image header in folder view alongside Remove from folder; and the "View full image" button moves to the right end of the editor toolbar.

Previously, opening the archive modal outside the Contacts table context could result in an archive request being sent with no contacts selected, while still showing a success message. The Archive button now stays disabled until contacts are selected, and the action is blocked regardless of how the button is triggered.

Contact detail page: collapsible Recent Activity and accessibility improvements

A chevron next to the Recent Activity heading lets you collapse and expand the activity feed on the contact detail page. Scroll position and poll timers are preserved while the feed is collapsed, so nothing is lost or reset. Collapse state resets when you reload the page.

Additional accessibility improvements ship alongside this change:

• The rail resize handle is keyboard-focusable, with arrow keys moving it in 2% steps and Home/End jumping to the extremes.
• Collapse toggles on the Tasks and Deals cards have accessible labels.
• When a header channel button opens a tab, focus moves to that tab's panel.

The tab strip on the contact detail page now shares the same tinted fill, 1px border, and rounded top corners as the Recent Activity header, connecting flush to the panel below. Open Tasks and Deals panel headers use the same fill color, giving the page a consistent look.

Task and deal modals opened from the contact detail rail render above all other page elements. Previously, parts of the secondary nav and the custom objects panel could appear on top of an open modal.

Previously, each row's field selection replaced the shared options data, causing options in other rows to disappear — for example, "Steward" options would vanish after adding a "Last Contact" field in a second row. Options are merged rather than replaced, so each row keeps its selections intact.

Dropdown options for custom fields also appear in a consistent, configured order across all selection types (dropdown, checkbox, property).

Columns held hidden by a policy no longer appear in the toolbar's Columns menu. Previously these columns showed as unticked options; ticking one saved a preference that the policy immediately overrode, and the column then appeared briefly when the menu closed.

Campaigns list: The Campaign list follows the same pattern as Flows: a totals strip, search, filter, status indicators, and row and bulk actions. Tables paginate when there are more than 10 rows. The Report opens in a drawer instead of inline, and the skipped-reason breakdown reads as a table.

Campaign creation: One editor handles creating, editing, and reporting. A phone preview shows your message for a sample contact; the rail holds Message, Audience, Delivery, and Reach sections. Fields use a centered, full-width layout. Step actions are pinned to a sticky footer. Audience and exclude-segment lists use toggles instead of checkboxes, with reach breakdown as a table. The review step reads as a table with top-aligned labels. Chips can be picked from a dropdown of added ones and created in a modal with a label and color.

Scheduled Campaigns are locked after scheduling. To make changes, cancel the scheduled send and duplicate the Campaign. Creating a list from the contact rail requires the list-add permission; without that permission and no existing lists, the menu links to Lists instead.

Toasts across Engage display correctly — with visible backgrounds and icons.

On screens 768px wide and below: the blueprint card no longer touches screen edges, tap targets are 44px, and long blueprint titles and trigger/action names wrap instead of overflowing. The "Activate now" button goes full width when the card stacks.

When the Active Intelligence sidebar is open, the blueprint card stacks based on its own available width rather than the window width, so the layout stays usable in narrower windows.

The accounts grid in ActiveCampaign HQ shows only the account you're logged into and its direct child accounts, so you see a clean, relevant list instead of every descendant across all levels. When you're in a sub-account, the grid is scoped to that account. The parent dropdown is searchable and translated.

Previously, "Black Friday / Cyber Monday" and "Holiday Shopping Season" appeared as the top two ideas for virtually every account, and 63 accounts received technical data signals as topic suggestions. Suggestions draw from your top-performing campaigns ranked by click rate, your best-performing campaign type, and country and industry holidays relevant to your business. A model rewrites up to 15 ranked candidates into polished, on-brand topic suggestions with rationale, and results are cached for 24 hours so repeat requests are instant.

When you're already building from a past campaign using the duplicate or "based on campaign" flow, suggestions offer fresh angles rather than follow-ups to that campaign. Suggestions are returned in your language when it can be detected.

In a 79-account test, technical topics dropped from 63 accounts to 1, and past campaigns appeared for 59 of 79 accounts.

When Active Intelligence presents a Scheduled Task as a clickable suggestion card, clicking the card passes your original request directly to Scheduled Tasks — the task is created from the card's pre-filled details without asking you to restate what you wanted.

Suggestion cards created before this change have no linked ID and won't behave this way when clicked.

This applies everywhere topic suggestions appear: the source preflight in Duplicate with AI, the chat's "from a past campaign" preflight, and the topic ideas above the composer.

Numbered lists in Active Intelligence responses are readable in Light theme, using the same dark text color as bulleted lists, headings, and paragraphs.

Previously, the second creation could reuse the first card or fail without explanation. Active Intelligence handles both requests in sequence rather than conflating them, so you always know exactly what was saved and what still needs to be created. Same-message retries still return the cached result.

Active Intelligence reads attached PDF and CSV files

You can attach a PDF or CSV to an Active Intelligence chat message and ask questions about it — and get a direct answer, not a redirect to build something.

For PDFs (and DOCX files), Active Intelligence returns the document's stored text in its reply. For CSVs, it previews the first 20 rows (up to 4,000 characters), sniffs the delimiter automatically, and tells you whether all rows are shown or some were left out.

If you attach a CSV and ask to import contacts, that request still routes to the contact import flow as before. XLSX and XLS files are not readable in this release.

If a file can't be found after upload, Active Intelligence reports it as not found rather than ending the conversation with an error.

When you attach a file — such as a PDF or DOCX — and ask Active Intelligence a question about it, Active Intelligence answers from the file's contents.

Two additional routing improvements are included: attachment rule descriptions correctly name the file types that can be read, and when a PDF and another file type are attached together, the routing rules for all file types are applied rather than only the first match. Contact import via CSV continues to route to the import tool as before.

For segments with a stored run from the last 90 days, the Contacts tab reads that result directly. For previews, segments with no stored counts, incomplete data, or runs older than 90 days, a fresh run is requested as before.

Industry Intelligence trend insights filter out guides, event pages, forum posts, and earnings reports from adjacent professions that only borrow your sector's vocabulary. When search results contain no real market signal for your sector, the insight says so rather than building a generic trend story on weak sources.

Before this change, the relevance filter kept results based on year-in-title or trend-topic keywords, not on whether they reported actual market activity in your sector. In before/after testing on production input, the previous filter kept irrelevant results in 9 of 10 runs; the updated filter kept nothing in 10 of 10 runs on the same input.

After Active Intelligence delivers a report answer, it offers to turn that report into a recurring Scheduled Task. The same report is offered only once per conversation; a different metric, Campaign, list, segment, or time window gets a fresh suggestion. Accepting the offer routes through Scheduled Tasks to create a Draft task. If your follow-up is ambiguous, Active Intelligence asks a clarifying question rather than creating a task.

The offer appears as text in the chat. A suggestion card in the UI is coming in a future update.

After Active Intelligence answers a question about rates, trends, or comparisons across a time window — for example, "What are my open and click rates this month?" — it offers to turn that report into a recurring Scheduled Task. The offer appears as the reply's only question, so saying "yes" unambiguously accepts it.

Active Intelligence makes the offer only when it applies: single-count lookups, questions about one named Campaign, written content requests, and actions that change your account data don't trigger it. One-time reminders (for example, "Dec 1, 9am remind me to review the holiday campaign") are created as non-repeating tasks. Any task that would change account data on every run asks for your explicit confirmation before it's created.

The four situations covered:

• Registration reads: If your user group can't read SMS registration status, Active Intelligence skips the registration prompt and asks an admin to confirm the send location rather than claiming the number is ready to send.
• Credit and rate lookups: A permission-denied response on SMS credit balances or country rates tells you to contact an account admin directly, with no generic fallback.
• SMS data queries: When Automation or broadcast data is blocked by missing permissions, Active Intelligence names the specific permission needed — for example, "Missing SMS permission: one of management.reporting, campaigns.view, campaigns.manage, campaigns.send" — and directs you to Settings › Users and Groups. An authentication issue is distinguished from a permission issue and tells you to refresh and retry.
• Business Goals SMS linking: If linking an SMS Campaign to a Business Goal is blocked, Active Intelligence reports that only the SMS link was refused and that everything else done for the goal is unaffected, then provides admin guidance.

When you ask Active Intelligence about an Automation that includes a Wait Until step, it describes what actually happens to contacts at that step. Each Wait Until time-limit setting is explained correctly: no limit waits for the condition; "for up to" and "continue" move to the next action; "exit" ends the Automation.

Active Intelligence reports the contact figures that are available — entries, completions, currently active contacts, week-over-week change, and conversions — and states plainly that step outcomes, sends, failures, skips, and daily trends are temporarily unavailable. It doesn't suggest expanding the date range or infer the Automation's structure from missing data.

Previously, on accounts where your profile stored a "same design, all-new content" preference, Active Intelligence read that preference and skipped the form, rewriting all content without asking.

The form appears whenever you haven't explicitly confirmed what to keep in the conversation. If you type "same design, new copy" in your message, you'll see the form once before the content is rewritten.

Active Intelligence conversations were stopping with a "Conversation limit reached" message after reading several Campaign email bodies. Campaign reports were returning up to roughly 92,000 characters of raw HTML per report. Campaign bodies are returned as plain text, and each report is around 8,000 characters instead of 92,000. You can ask for the full HTML explicitly if you need it.

When the Campaign editor can't handle a request, Active Intelligence answers in the chat itself rather than falling back silently. Scenarios now handled directly include: editing or copying a Campaign when none is resolved (asks which Campaign), copying a template (explains a template is not a Campaign), split-testing with a new Campaign (builds the Campaign with its subject test), split-testing with no Campaign resolved (asks which Campaign), and several error states such as an untargetable email or a design the service cannot adopt.

Selecting "Change look" on an existing Campaign in Active Intelligence opens the brand picker. Previously, under certain conditions, the action returned a "no other layout fits" reply instead of the picker.

The −, zoom, and + buttons in the Engage canvas zoom group render as a single joined control with a shared outline. Previously, each button drew its own top and bottom edge, causing hover and open-menu fills to bleed into the seams and the control to read as overlapping borders in two shades of grey.

When you edit an automation email in the mini editor and navigate away without saving, the editor asks what you want to do with your changes so nothing is lost silently.

The refresh label under the total contacts card reads "Refresh contact count (last updated [date])" in your account's date and time format. When a count is still being computed, the label reads "Refresh contact count" without a date. The agent segment drawer shows the same information.

The refresh button is disabled while the refresh is running, and is available both on the Segments page and in the Active Intelligence sidebar.

Clicking, pressing Enter, or pressing Space on the clear button clears the input and returns focus to the search field. The button's accessible name is "Clear search".

Active Intelligence reads the full conversation and applies a clear order of priority: your acceptance reply wins first (for example, "yes, make it weekly"); if absent, the recurrence in your original request ("every quarter" → 3 months, "last month weekly" → 1 week); then the reporting window you named ("last month" → 1 month, "last 7 days" → 1 week, "last quarter" → 3 months); and finally a default of weekly when nothing else is specified. The offer wording itself never sets the cadence.

When Active Intelligence creates a task with the default one-week cadence, the task starts tomorrow rather than waiting until the following Sunday. The cadence description shown after a task is created is taken directly from the stored schedule, so what you read matches what was saved.

A Remix option appears in each row's menu on the Automations overview list, so you can start a variation of any existing Automation without opening it first.

The Engage Audience page has a new layout. Segments move into a navigation panel beside the contact table, replacing the previous flat layout. You can create and edit segments from a side panel that opens with a "Describe it with AI" prompt. Opening a contact's details keeps you on the page — the contact opens beside the table without navigating away.

Possible duplicate contacts surface as a dedicated "Review" entry in the navigation, visible only while duplicates exist. The figure strip shows live data: total size, top channel, and dormant count for everyone; and people, reachable, and opted-out counts per segment. Contact status is set from one consistent list everywhere: Active, Inactive, Spam, and Opted out.

Paging: views with 100 or fewer contacts, or any search result, are sortable and client-paged. "Select all" selects only the current page.

When migrating a contact database, all migration-related cards — the import progress card, the "Your data is here" card, and the import success card — appear on the Inspire me tab alongside the export-phase card. Previously, once an import started, the import progress card only appeared on the Your dashboard tab, leaving users on the Inspire me tab with no migration card visible. The Your dashboard tab and the legacy layout are unchanged. The "Your data is here" card applies only to certain import types.

The selection is checked against the channel's type, size, and count limits before insertion. Drag-and-drop and paste continue to upload from your device as before.

In the "Create New Form" modal, the "Add a tag" field works correctly with keyboard navigation after typing a search term — pressing ArrowDown or Enter selects a matching tag, and the dropdown no longer flashes empty between keystrokes.

Rapid repeat clicks on the remove control are ignored, and the control is disabled while a folder change is saving. Menu options that aren't available — such as "Replace original" — are now visually styled as disabled rather than hidden, so you can see what's possible at a glance. Small image previews are no longer upscaled, preserving their native resolution.

Adding an emoji to a Campaign subject line or preheader and immediately deleting it with backspace keeps it deleted. This applies to subject line, preheader, from name, from email, and reply-to email.

A "Delete Task" button appears in the Edit task modal footer whenever you're editing an existing task on the deal details page. It's not shown when creating a new task, and it's greyed out while a save or delete is in progress. Deletion happens immediately and a success message confirms it.

The Recent Activity widget shows "Deal Owner changed to [new owner name]." The Activity tab timeline shows the transition as "Deal owner [old] → [new]", falling back to a generic "Deal owner changed" message when the previous owner can't be resolved.

The DNS status box in the domain setup modal stays within the modal body.

When the account's Conversation assignment toggle is off, assignment-history status messages no longer appear in the conversation thread. Non-status messages and other status events continue to show normally.

The underlying history is not deleted — turning assignment back on restores the full history. A day separator is also suppressed when the only messages for that day were assignment status messages.

Searching from the global header and then searching again on the Contacts list returns results for the new term, and the search box shows it correctly.

Accounts migrating from certain platforms were previously placed into an onboarding experience that replaced the homepage. When the migration card moved to in-progress, that experience advanced and migration progress became inaccessible. You now land on the homepage with migration progress cards visible from the start.

After a Brevo or HubSpot import finishes, a completion card appears on the agent page with a "Review my data" button that takes you straight to the import review screen. The card appears as soon as the export summary loads — and even if the summary can't load, the card still appears so you don't lose access to the review entry point. Closing the review screen returns you to the agent page.

• Brevo: covers unsubscribed and blocklisted contacts.
• HubSpot: covers contacts opted out via the email opt-out field on a list.

Selecting a connected integration from the list adds context that Active Intelligence uses when responding. The same list appears in the Add-context button menu. Only integrations in a connected state are shown; the group is hidden when nothing is connected. This applies to all surfaces that use the chat composer, including Scheduled Tasks.

Previously, the Segment field showed "--" for a large number of Campaigns. When the segment name isn't available directly from the Campaign details, it's looked up from the segments service. The Segment field shows a loading indicator while the name resolves.

On the Brand Kit page, the Set as default, Edit, and Delete buttons on More Brands cards stay within the card boundary, so the Delete button is no longer clipped at the card edge.

Split by property adds a branch-on-condition step with form fields and operators; arms read False and True.

Run stats shows a "Runs by status" breakdown and a 30-day bar chart of completed, running, and failed runs.

Delivery steps include link delivery and storefront delivery, with pickers drawn from your existing links and storefronts.

Reaction trigger supports an optional emoji list. It is listed but disabled on Messenger.

Retries on sending steps let you set retries count, wait between retries, and stop-trying-after.

Additional steps available: agent handoff, add to Automation, create deal, sequence enrollment, and end. The webhook step picks a webhook subscription instead of a typed URL; the add-to-segment step lists manual segments only.

Current limits: the Automation picker shows only the first 10; the deal stage list is not filtered by pipeline and options read "Pipeline · Stage"; an existing webhook step with a URL shows as needing setup until a subscription is picked.

The automated report details page includes a failures widget below the KPI row, showing the breakdown of SMS delivery failures for the specific automation block. Rate calculations use the block's own send count. An error code the widget doesn't recognize is shown as an unknown error.

When you ask Active Intelligence to create a Campaign based on a past Campaign via chat, the source form shows topic idea suggestions — the same suggestions already available in the "Duplicate with AI" flow. When your prompt already specifies what the new Campaign is about, no topic ideas are shown.

When you open the Automation Canvas from a chat in the Active Intelligence sidebar, the sidebar stays visible alongside the canvas so you can act on an AI suggestion without losing your conversation.

Previously, the prompt was misrouted and you ended up in a questionnaire flow with no form to show for it.

Active Intelligence treats any request to create a new or first form as a create request, even when the phrasing asks for ideas first. It builds an inline newsletter signup with full name and email fields from your brand kit and business context when no specific goal, layout, or fields are named — without asking for those details upfront.

When building a Campaign from a past campaign using Active Intelligence, a second preflight form no longer appears after the first is answered.

Several reliability issues in the Active Intelligence chat experience have been addressed:

• Stream resumption reconnects at the exact event, preventing duplicate content when a connection drops and reconnects.
• Reloaded runs apply their content once. Previously, after a page refresh mid-run, every chunk was applied twice, toggling the drawer and refreshing caches twice.
• Requests without a valid session no longer reach Active Intelligence. Previously, load-time requests reached the backend with no usable credential and failed silently.
• Thumbs-up and thumbs-down feedback is recorded in analytics again. The tracking event had stopped firing when an earlier change replaced it with a different call.

Specific improvements:

• When a run's first step fails, a fallback reply and a clear error message reach the chat immediately rather than leaving the conversation stalled.
• Typing a slash command with no further text produces a clear reply.
• When a request gathers more content than the model can process, the chat replies with a clear explanation rather than timing out.
• Every way a run can end — completion, agent error, timeout, cancellation — produces a terminal state you can act on.
• A retried turn stores the conversation once, not twice.
• When you stop a chat turn, the loader clears, the stop reply appears as its own paragraph, and the conversation looks the same after a page reload.
• After a stall, sending or retrying reattaches to the live run. A stalled run that is still live on the server continues streaming rather than ending with "Request timed out."

Free-form replies sent within the 24-hour customer-service window are charged under the SERVICE category rather than being free. Two exceptions remain free: service messages within the free monthly allowance per business phone number, and messages sent within the free entry point window from a Click-to-WhatsApp ad (covering all categories for up to 7 days).

Active Intelligence correctly describes both exceptions and points Click-to-WhatsApp questions at the free entry point window. Both free cases are handled correctly.

Reconnect notices reach the right people. The "Action required: reconnect your connectors" banner and the red dot on the Tools button appear only for users who have connectors that need reconnecting. Both retire once you've opened and closed the Connectors modal, or connected a connector. The Tools menu also renders correctly when the connector list is empty, keeping the connections manager and the custom connector entry reachable.

Previously, on accounts where no connector had ever been set up, every user saw the banner and red dot even though there was nothing to reconnect.

Visual polish. Connector tiles render correctly in both light and dark themes. Connector logos sit on fixed light plates so dark logos stay readable in dark mode; missing or failed logos show a placeholder icon. The modal uses a consistent size across all views — list, details, custom connector form, and consent — and adapts correctly on narrower screens. The Tools menu closes when the Connectors modal opens. Status chips stay on one line.

The Activity and Knowledge tabs in the Active Intelligence sidebar render once per interaction and stay idle afterward. Search, filter, range menu, paging, and opening a turn all behave as before.

Active Intelligence checks SMS permissions before attempting to create or edit an SMS Campaign, giving accounts with granular SMS permissions clear, actionable feedback instead of a failed write.

Five permission states are handled:

• No Campaign manage permission: no write is attempted; a preview-only result is returned pointing to SMS Campaigns › Manage and Settings › Users and Groups.
• Manage permission but no send permission: the Campaign is saved as a draft; the response tells you that someone with SMS Campaigns › Send permission must review and send it.
• No send permission on a schedule edit: the scheduled date is not applied; a mixed edit (schedule plus other fields) saves the other fields and reports that the schedule was not set.
• No SMS registration: guidance points to the account administrator.
• Permission denied by the platform: the existing permission-denied message is returned.

This applies to accounts with granular SMS permissions enabled.

You can have much longer back-and-forth conversations with Active Intelligence before hitting a length limit. A single shared limit previously cut off roughly 412 conversations per month, often well before the model ran out of capacity. Limits now match the capacity of the model handling your conversation.

For conversations running on the current default model (which has a 1-million-token context window), the warning threshold rises from 280,000 to 400,000 tokens and the stop threshold from 300,000 to 500,000 tokens. Based on production data from the past 30 days, this reduces conversations reaching the stop point from around 412 per month to around 31. Conversations on models with smaller context windows keep the existing limits.

Your existing conversations are automatically covered by the new limits.

Active Intelligence can read and act on your Engage inbox and channel flows, so you can manage conversations and flows without leaving the assistant.

In the Engage inbox, you can ask Active Intelligence to find, count, or search conversations, read messages, summaries, notes, and labels, and take action: set a conversation's state, assign it, manage labels, or add a note. You get a clear view of what needs attention and can act on it in the same conversation with the assistant.

With Engage flows, you can ask Active Intelligence to list and inspect your flows, check run history and step performance, draft a new flow (created in a paused state so nothing goes live until you're ready), or pause a live flow.

Replies and flow activations require your approval. When you ask Active Intelligence to send a reply or activate a flow, it shows you a preview first. The action only goes through after you confirm it. This applies every time — Active Intelligence will not send replies or activate flows on its own.

Active Intelligence respects your role and channel permissions in the Engage app. It can only see and act on conversations and flows you already have access to. Editing, deleting, duplicating, or making bulk changes to conversations or flows are not supported.

This capability is available when your account has the Engage app enabled.

You can ask Active Intelligence to remove all recipients from a Campaign, leaving it as a draft until you choose a new list. Previously, asking Active Intelligence to remove a Campaign's list had no effect — the list stayed in place while the reply implied it had been cleared.

What Active Intelligence does when you ask to remove a list: - Clears both the list and any segment, writing empty sets to the Campaign. - If the Campaign still shows a list or segment after the write (for example, because a read returns stale data), the response tells you and directs you to remove them in the review panel. - If your request asks for no list while also naming a specific list or segment, Active Intelligence makes no change and asks which you meant. - The Campaign stays a draft until a list is chosen.

Previously, hygiene questions could result in a pivot to open and click metrics or a focus on a single automation rather than your full portfolio. Active Intelligence now compares each automation against your account baseline and flags outliers only when they deviate meaningfully: a bounce rate more than twice your median, or an unsubscribe rate above 1%.

When nothing stands out, Active Intelligence says so rather than manufacturing a finding. Every cited metric is tied to its data window and send count. Missing metrics are flagged as missing rather than reported as zero.

This improvement applies to automation hygiene questions. Engagement-performance and Campaign-content insights are unchanged.

The WhatsApp tab on the redesigned contact detail page links directly to the contact's active WhatsApp conversation and includes an "Open inbox" button. The tab label shows the unseen message count, so you can see at a glance whether there are messages waiting without opening the conversation.

The tab is hidden automatically when no WhatsApp inbox is connected to your account or when the contact's WhatsApp visibility is turned off — so it only appears when it's relevant.

This tab is part of the redesigned contact detail page, which is not yet broadly available.

The toll-free number card reads "Best for brands that text nationwide." The meaning is unchanged.

The registration card displays five states: pending, registered, needs update, rejected, and loading. When a registration is rejected, the card identifies the specific information to update in plain language — for example, "Information to update: Business website" — rather than showing a raw system field name.

The full registration flow is not yet end-to-end usable and is not available by default.

Hovering or keyboard-focusing a tag in the Tags tab on the contact page shows a Tool Tip with the date and time the tag was added. The date follows your account's configured time zone, locale, and datetime format — the same format used in the activity stream.

If a tag's name is too long to fit in its chip, the Tool Tip shows the full name above the date. Tags with no recorded date show no Tool Tip.

Surveys and Storefronts have been removed from the Manage section of the Engage navigation. The Manage section shows Links, Posts, and Settings.

Both pages remain accessible by direct URL if you need them.

On the Campaign summary screen, the Subject line and Preheader fields and their emoji buttons are the same height and align correctly.

The notice appears only when whole-design testing is off and you're on a variant other than the first. It doesn't appear for non-split Campaigns, when whole-design testing is already on, or when you're viewing the first variant.

The dialog contains the same controls as before — winner determination, winner criteria and wait period, email share allocation per variant, and the whole-design checkbox — but works in draft form: Cancel discards all changes, and Save writes only what changed. Share fields let you clear and retype a value freely while editing. Even-split shares are shown as an exact distribution and are read-only.

The Campaign editor canvas also reads in the order a split test is understood: settings row first (Recipients, Send, More settings, Split test settings), then the variant strip, then the selected variant's subject and preheader above its design.

The dialog appears in every Campaign editor surface, including the Active Intelligence full page and sidebar. Non-split Campaigns, templates, and Automation emails are unaffected.

A version picker lets you choose which variant to resend; it defaults to the winner, or shows no selection if no winner has been decided. The resend becomes a single Campaign with 100% of the send going to the chosen variant, rather than re-running the split.

Also fixed: Resend requests that contain only the resend type — without a name or subject — complete successfully.

Previously, the prompt text was dropped and an empty message was sent instead, meaning Active Intelligence received no instruction. The Active Intelligence sidebar was not affected by this issue.

When the segment preview contacts tab fails to load, the error state shows a warning illustration, the error message, and a retry button. Clicking retry re-runs the contacts query without a page reload.

The Segment integration settings panel — including Save and Disconnect options — is available again in the app marketplace. It was removed when the previous apps page was retired; this restores full settings access in its replacement.

When you open the Scheduled Tasks composer, the input starts as one compact line. Clicking it expands the field, activates the AI gradient border, and reveals the Add, Tools, and Add context buttons. Click away with nothing typed and it collapses back to its compact state.

Scheduled Tasks moves from the AI nav drawer to a top-level primary navigation item at its own URL. Accounts on the previous URL are redirected automatically.

The toll-free registration card is visible to accounts that have never started a registration. Previously, the card hid itself for those accounts — exactly the ones that need the Register button. Accounts already pending or approved are unaffected.

When a group lacks the SMS campaigns Manage permission, Active Intelligence shows a clear permission-denied state in the SMS card. The card displays the agent's message as a warning, shows a title, and omits the primary action button. Accounts without the SMS add-on still see the upsell state, because a missing add-on takes precedence over a missing permission.

The phone preview in the SMS builder reflects what contacts will actually receive.

• If your message doesn't include opt-out language, the preview shows the footer and explains that it's added automatically — once every 30 days per contact. If you want it in every message, add it to the message body yourself.
• If your message already includes opt-out language, the preview confirms the automatic footer won't be added.

The detection logic in the preview matches the logic used at send time, so there's no gap between what you see and what your contacts get.

Actor-initiated assignments show who made the change and to whom. Automatic assignments by first-responder rules appear without an actor name. A fallback label appears when a user has no display name, and any status event the app doesn't yet recognize surfaces as "This conversation was updated" rather than being hidden.

The refreshed Contact record adds two new tabs: SMS and Notes.

The SMS tab shows a contact's full SMS history as a flat list, newest first. Each row displays the message text (with personalization tags filled in), a sender avatar, a sent or received label, and the date in your account's format and time zone with a relative time indicator. Replies are labeled separately. Automations that sent a message include a link to that Automation. Older messages load as you scroll, and Inbox rows link through to the Inbox conversation.

The tab is only visible when your account has SMS access. When no messages exist, the tab shows an empty state. Unsubscribes and messages to internal users are not shown.

The Notes tab brings notes access into the updated contact detail page.

Both tabs are part of the Contact Detail Page Refresh, which is expanding the contact record from a 2-tab layout to a 9-tab layout. Neither tab is visible to customers yet.

SMS consent text in embedded Forms — floating box, modal, and floating bar types — uses the form's configured body font color. Forms without a body color set continue to display the consent text in the default grey. Inline forms are unchanged.

Each assignment event appears in the thread at the point it happened. Actor-initiated assignments show who assigned the conversation and to whom. Automatic assignments use appropriate copy. Events with missing metadata show a generic fallback rather than disappearing.

Previously, "View all time" could produce inconsistent figures across the KPI snapshot and the row list. Both surfaces reflect all-time data when that option is selected.

Previously, the KPI tiles and block list were scoped to your chosen range, but the per-block rows showed all-time figures. Selecting "View all time" or clearing the range sends the correct parameters throughout.

When resending a split Campaign, a version picker lets you choose which variant to carry forward. The resend becomes a single Campaign with 100% of the send going to the variant you select. The picker defaults to the winner, or shows no selection if no winner has been decided.

Also fixed: Resend requests that contain only the resend type — without a name or subject — complete successfully.

Clicking "Your Dashboard" opens the agent homepage directly at the My View tab. Clicking the Reports heading itself still lands on All Campaigns Performance, unchanged.

The button appears below the total contact count. While a refresh is running, the button is disabled so you won't trigger duplicate recalculations. The refresh label reads "Refresh contact count (last updated [date])" in your account's date and time format; when a count is still being computed, the date is omitted. The same control and last-updated information appear in the Active Intelligence sidebar.

When your session token expired, the platform used to send several simultaneous refresh requests at once — and because each one invalidated the last, most of them failed. The result was a full redirect to the login screen, often in the middle of what you were doing.

All concurrent refresh requests within a tab share a single request. The expected drop in failed refreshes is from roughly 55% down to around 9% — meaning far fewer unexpected login interruptions during normal use.

The Engage template gallery has a new visual design. Each template card shows an Instagram DM on a phone in a frosted frame over a light brand gradient — one shade per template category: engagement, onboarding, sales, and retention. On hover or keyboard focus, the gradient deepens and the phone rises. Featured templates get a sparkle badge; coming-soon templates get a clock. Grid cards no longer show a description or category label.

Search and filter controls are hidden while the catalog holds fewer than 12 templates and appear automatically once it reaches that threshold.

The "Use template" flow is updated to match the Saved replies and New deal modals: the × in the heading closes the modal, and focus starts on the first field the template requires.

The Engage Audience page has a new layout. Segments move into a navigation panel beside the contact table, replacing the previous flat layout. You can create and edit segments from a side panel that opens with a "Describe it with AI" prompt. Opening a contact's details keeps you on the page — the contact opens beside the table without navigating away.

Possible duplicate contacts surface as a dedicated "Review" entry in the navigation, visible only while duplicates exist. The figure strip shows live data: total size, top channel, and dormant count for everyone; and people, reachable, and opted-out counts per segment. Contact status is set from one consistent list everywhere: Active, Inactive, Spam, and Opted out.

Paging: views with 100 or fewer contacts, or any search result, are sortable and client-paged. "Select all" selects only the current page.

After re-authenticating on a URL that contained a query string, your session is restored correctly and subsequent actions complete as expected.

Previously, re-authenticating on a URL that contained a query string could result in errors and failed API calls after the session was restored. Re-authentication on any URL, including those with query strings, restores your session correctly.

The CVC field on payment forms displays its placeholder text as expected. Customers filling out their card details see the correct prompt for both three-digit and four-digit security codes.

Typing a page number outside the valid range in any paginated table — such as All Campaigns, Recipients, Clicked Links, or Segments — and pressing Enter or leaving the field clamps to a valid page. Values above the last page clamp to the last page; values below 1 clamp to page 1. Pressing Escape reverts the draft without navigating. Next and previous buttons remain clickable while a page request is in flight.

When two Contacts are merged, the merged Contact keeps the stronger privacy choice: if either Contact had opted out of email open tracking, the merged Contact is written as opted out, with the opt-out timestamp set to the merge time.

Contacts where neither had expressed a preference are left untouched — no opt-out is written.

The merge log and activity stream record an "Open Tracking Opt-Out" entry when the two Contacts differ on this field.

The wait-until date/time field in the Automation Canvas offers minute offsets as a timing option, giving you finer control over when a wait step ends.

The "Saved at" timestamp, revision history control, and refresh button are aligned consistently in the Manage Contacts header status bar.

When the app runs inside a frame from another origin, a session expiry previously caused the page to stop responding rather than redirect to login. The redirect targets the top window correctly so login always happens at the right place, even in embedded contexts.

Industry Intelligence scheduled insight cards display a meaningful, descriptive title. Previously, some cards showed the literal word "Summary" as their title instead of a real one.

When you open a Form via its preview link, a banner across the top of the page indicates that the view is a preview. If the Form is paused, the banner additionally states that it is not accepting submissions.

Three visual improvements to the Flows list:

• Triggered, Completed, and DMs counts are right-aligned with tabular figures, including their column headers, so numbers stack cleanly by digit.
• The channel disc sits 8px from the flow name, tightened from 12px.
• The Active flows tile no longer shows a separator dot between its two counts.

The Files tab on the contact detail page lets you view, upload, download, rename, and delete file attachments for any contact. The tab is available for all contacts.

A previous issue where uploading a file from the page header caused the Files tab to load indefinitely is also resolved — the file list refreshes correctly after a header upload.

Several improvements to platform startup and resilience shipped together:

• Faster startup. Two startup files download in parallel instead of one after the other, reducing the time before the app is ready.
• Boot timeout protection. If the app hasn't finished loading after 30 seconds, an error screen with a retry button appears so you're not left with a blank page.
• Slow-page notice. If a page's loading state has been visible for 15 seconds, it switches to a "This page is taking longer than usual" message with a reload button. The page still renders normally if data arrives.
• Session recovery. When a session check fails, the app re-checks your session automatically (at most once per 30 seconds) and only redirects to login if your session has actually expired.
• Chunk-load recovery. If part of the app fails to load after retries, a "We couldn't load this section" prompt with a reload button appears instead of a broken page.

The token is renewed once and every caller receives the updated token without a redirect.

Each card in the Engage templates gallery draws its example exchange from real inbox thread parts and lists only the steps its flow actually contains — no guessing what a template does once you're inside it. "Comment on a post" appears as the featured card at full width.

Clicking a card opens "Use template." The "Comment on a post" template prompts you for the reply text, prefilled, before posting a public reply ahead of the direct message.

Seven template names were updated: Comment on a post, Welcome new follower, Came from an ad or Reel, Story reply, Qualify commenters with AI, Flash sale, Win back quiet contacts. Each card includes a sample inbound message and a sample DM reply, a Featured label, short step names, and a public reply field in the Use template dialog.

The renamed templates take effect on the current Templates page immediately. The new gallery cards appear when the accompanying front-end update ships.

Embedded app sections on Contact and Account detail pages stay visible after navigating away and back, switching tabs, or when two views of the same app appear on one page. Each view manages its own lifecycle independently, so closing one does not affect others.

Embedded app sections on Contact and Account detail pages were disappearing after navigating away and back, switching tabs, or when two views of the same app appeared on one page. Each view manages its own lifecycle independently, so closing one does not affect others.

Four ecommerce summary endpoints are now published in the public API spec, giving developers documented access to revenue, order status, product rankings, and customer behavior aggregates via the v3 API.

Templates display in your language when you set up a new pipeline. Previously, the catalogue was available in English only.

The PUT /dealTasks/{id} spec previously omitted two fields the server enforces on every request, causing bulk task updates through the API to fail. The edate field is now documented in the request body, and outcomeId now states when it is required and what it must match. No server behaviour changed.

The custom objects panel defaults to an empty state when it can't retrieve its configuration — the same as it does for accounts with no custom objects configured. Previously, the same failure caused the entire page to stop loading, affecting approximately 257 crashes and 160 accounts per day on the Contacts page.

Account date formats that mix literal text with date specifiers now render correctly everywhere in the platform. Previously, formats like these caused an error on the Campaigns index and showed "preview unavailable" on the advanced settings page.

The date format preview in Settings uses the same converter as the rest of the platform, so the preview matches what you see elsewhere.

The consent step shown before the OAuth redirect for custom connectors is translated into your account's language. Previously it appeared only in English.

Orders that included a product the account already owned were previously rejected. The order builder keeps only one entry per product, so those orders complete as expected.

Conversation list and toolbar are redrawn with consistent measurements. Filter, Newest, and Select controls are sized consistently; Status options fold into a select beside Owner. Filters by channel, status, owner, assignee, and AI agent are available, with applied-filter chips, sort controls, a "start with this view" option, hover-checkbox selection, a selection bar, and bulk actions. Each conversation card shows an assignee and status line.

Thread pane has new bubble and reaction placement, inline notes placed in the transcript by timestamp, a typing indicator, an Active Intelligence replies toggle, and a messaging-window strip showing time remaining. Reactions appear on the message they react to rather than as a separate bubble. Loading an older page of the transcript prepends above your current position without jumping the thread backwards.

Composer has mode tabs for Reply and Internal Note with their own drafts, a flattened emoji picker, and an attachment row. The reply composer collapses to a single line and expands on focus, with image and audio attachment options and a character counter. Accepts saved replies via a menu, a / menu and library with ActiveCampaign field insertion, a reply-to strip, and image attachment by upload, drop, or paste.

Active Intelligence draft action in the composer fills the input with a generated reply you can edit before sending. The generated reply is never sent automatically. The action is unavailable when the send window is closed. A generation failure leaves any existing draft intact.

Contact panel includes Activity and Notes tabs, a Contact Profile entry, and Nurture, Automations, Deals, and Lists cards. Notes can be edited and deleted directly from the panel. An AI brief sits above collapsible sections — Details, Automations, Deals, Lists, Notes, and Past conversations — with a headline, a "N new since" indicator, a refresh control, and next steps that open an add menu with a target pre-staged. Teammates can choose which Details and header facts to surface.

Labels can be added and removed from the thread and survive a reload.

Saved replies are create-and-read only — a saved reply cannot be edited or removed after creation.

Several surfaces return empty because the relevant read endpoints are not yet available: the account's label list in Add labels, edit and delete in Manage saved replies, Assign to… on a selection, the rail Summary card and its Refresh, contact details and past conversations, and leaving a list or an Automation.

When a Contact is deleted or merged, its associated Instagram and Messenger channel identities are cleaned up at the same time. Hard-deletes cover all Contacts including soft-deleted duplicates.

In the app: The bulk Edit and "Edit all" buttons carry the full target — specific contact IDs, the current filter or search set, or "all" — at the moment of the click. An empty selection refuses to open the modal rather than falling back to "all."

Confirmation step for "edit all": When editing all Contacts in the account, a warning appears and you must type the exact contact count to enable the Apply button.

API enforcement: The bulk edit API requires web app requests to declare their scope — by IDs, by filter, or all — as a query parameter. Requests missing or with an invalid scope receive an error. Requests scoped to IDs require a non-empty contact ID list; requests scoped to filter require at least one active filter constraint. API key requests are unaffected.

The contact record includes an SMS tab displaying the contact's SMS message history. The tab shows an empty state when no messages exist and a loading indicator while messages are being fetched.

Two new fields let you manage open-tracking opt-out status per contact through the Contacts API, meeting requirements in markets such as France and Italy where this opt-out is required separately from consent to receive marketing email.

v3 API (POST and PUT /api/3/contacts): open_tracking_optout (writable, 0 or 1) sets the opt-out status. open_tracking_optout_updated_at (read-only) is a server-stamped ISO-8601 timestamp that updates whenever the value changes. Sending an explicit null returns a 422 error. The fields are covered by contact sync and bulk sync.

v1 API (contact_add, contact_edit, contact_sync): an optional open_tracking_optout parameter (0 or 1) sets the status. Omitting it leaves the stored value unchanged. Out-of-range values are rejected with a clear error message.

The contact record displays "Opted out · {date}" or "Consented · {date}", with localization for 16 additional locales.

Contact CSV exports also include a new column for open-tracking opt-out status, recording 1 for opted out and 0 for opted in.

This applies to emails sent before the opt-out was set as well: if an open is reported for an opted-out contact, no activity is recorded, no deal activity is created, and no Automation open trigger fires. Click and reply tracking are not affected.

The contact notes panel keeps its height when the embedded content reloads or navigates away, rather than collapsing and requiring a page refresh to restore.

The contact limit notification banner appears at the top of the Automations overview, the Automations empty state, the Campaigns index, and the Active Intelligence landing page.

Accounts on purchase-order, net-terms, or contract billing see an assisted "Contact us" option in the banner. The banner is hidden for accounts with a pending cancellation.

After a contact import completes, the review screen offers an optional tidy-up step. If any of the contacts you imported arrived already unsubscribed, you can archive them in one action — without touching contacts who are active, unconfirmed, or bounced on any list.

The step is opt-in: nothing is pre-selected, and a confirmation screen gates the action before anything is committed. Only the main account administrator can run it.

A few things to know before you use it:

• The archive is scoped to contacts whose every list membership is an unsubscribe. A contact who is active on even one list — including lists outside the import — is left alone.
• A contact who unsubscribed inside ActiveCampaign after the import is treated the same as one who arrived unsubscribed from the source. Both will be archived if all their memberships are unsubscribes.
• If your account's delete allowance runs out mid-run, the screen reports how many contacts were archived and keeps your selection so you can retry the remainder when your allowance refills at the next billing period.
• A minimum wait applies before a restore is permitted. Your delete allowance is not refunded when contacts are archived.

The tidy-up option only appears once the system has confirmed that at least one unsubscribed contact exists in the imported lists. If that check hasn't completed or couldn't run, the option stays hidden rather than offering a destructive action on uncertain data.

The confirmation modal, section labels, and descriptive copy throughout the cleanup screen have been updated to reflect this change. A new partial-error state is also handled: if a cleanup run archives some but not all contacts, the screen shows how many were archived and prompts you to try again for the rest.

The contact created Automation trigger is available to all accounts without any additional configuration.

Connected email settings revert to their last confirmed value when a change is cancelled or the update fails, so the settings panel always reflects what's actually saved.

If your account has two or more Brand Kits, you can tell Active Intelligence which one to use for a chat session. The Brand Kit selector appears in the @ typeahead and the Add context menu. Kits are listed with your primary kit first, marked Default. Typing filters by kit name. Picking a kit applies that kit's brand context, tone, and competitors for the session.

The selected kit stays active for the whole chat — you don't need to re-select it on every message. Once a chat has messages, its kit is locked; to use a different kit, start a new chat. Before the first message, you can pick and replace a kit freely.

Context replacement is also improved: the Add context button stays available when a chip is already attached. Picking context of the same kind — Campaign, Automation, Brand Kit, or File — replaces the existing chip rather than blocking the pick. The @ typeahead works the same way, and a footer in the menu names what will be replaced.

Accounts with zero or one Brand Kit see no change.

The checkout payment form gains placeholder text for address, state, ZIP code, name on card, and expiration date fields. Four billing labels are recased for consistency: ZIP / Postal Code, Name on card, Card number, Expiration date. The expiry and CVC helper text can be hidden while field-level error messages continue to display.

Campaign preview mini editor saves reliably across all contexts

Edits you make in the Campaign preview mini editor save correctly every time, across the Active Intelligence full page, the Active Intelligence sidebar, Automation emails, and the legacy campaign preview.

Saves happen when you finish typing, not on every keystroke — fewer unnecessary background operations and a cleaner editing experience. If a save fails, a notification appears with a Retry option, and your edit stays pending until it succeeds.

Several specific behaviors are also corrected:

• Split Campaigns save body edits to the variant you're looking at. Previously, editing with variant B selected saved to variant A instead.
• Switching variants or reopening the editor restores what you last wrote, not the original version.
• Restoring a version reopens the editor with the restored content, so your next edit doesn't overwrite the restore.
• Outlook button copy updates correctly when the button link includes a query string or the label contains special characters.
• Editor-only styling is kept out of the saved document, so what gets saved matches what your contacts receive.

The Campaign performance report has a new metric showing opens that aren't tied to an identified Contact.

The campaign editor loads fonts faster. Previously, the editor made approximately 81 separate font requests every time it opened and each time the font picker was used. Font loading uses a single combined stylesheet served from a content delivery network, so fonts appear more quickly and the repeated requests are gone.

Web-safe fonts are unaffected.

Campaign builder textarea fields display at their correct size. Fields that had been migrated to the updated design system previously rendered at nearly double their intended height. The correct minimum height is restored, and clicking the preheader field label now focuses the field.

Affected buttons included Save buttons in share modals, the generate-link and refresh buttons in the share link control, the CSV download button in the export modal, a navigation chevron in report widgets, and a popup icon in subscribe-to-list fields.

The bulk Edit and "Edit all" buttons carry the full target — specific contact IDs, the current filter or search set, or "all" — at the moment of the click. An empty selection refuses to open the modal rather than falling back to "all."

On the API side, bulk edit requests must declare their scope explicitly. Requests missing or with an invalid scope are rejected. scope=ids requires a non-empty contact ID list; scope=filter requires at least one active filter constraint.

When a bulk status change across multiple Flows only partly applies — because one batch failed — the confirmation bar reports how many flows were updated and how many could not be updated, and offers a Retry button. Retry resends only the flows the write missed, read fresh after the list has refetched. The selection is kept so you can see which flows were affected.

Brand Kit website image scanning shows accurate status

The "Images from your website" section in Brand Kit reflects the real state of a website scan at every stage.

• Clicking Refresh shows "Scanning" from the moment you click until the backend reports a result.
• When a scan fails, an error message appears above any images already collected.
• If a website crawl stops partway through, the image section shows the failure right away instead of staying stuck on "Scanning" for up to 15 minutes.
• Images collected in earlier scans are no longer removed when they don't appear in a new crawl.
• A page left open polls automatically while a scan is pending, so it switches to "Scanning" when a scheduled crawl starts without requiring a page reload.

Brand Kit displays your website domain as entered, without adding www. to subdomains. A domain like news.ycombinator.com previously appeared as www.news.ycombinator.com in both the form chip and the brand tile header. Your stored data was correct throughout; this was a display issue only.

When the image validation service is temporarily unavailable during a Campaign preview, the error is reported as a service outage rather than as an invalid image on your account.

Typing a Brand Kit name after @ — for example, @espn.com — previously returned "No results" even when the kit existed under Your Brands. The Add context search box had the same gap.

Both searches match kit names locally and case-insensitively, so your kits appear as you type. The default kit is marked with a Default badge. In a conversation that already has a kit locked, searching for a different kit shows a note explaining how to switch.

Brand Kit gains a per-kit toggle — "Auto-add new website images" — under the "Images from your website" section. When you turn it off, scheduled brand monitoring skips adding new images for that kit. Your initial image scrape and the manual Refresh button are unaffected.

New kits default to on. Kits created before this option was added default to off.

What you get in the list: A totals strip at the top shows your portfolio at a glance. Each row displays the trigger and up to five steps as visual tiles, with a count of any additional steps. Rows show the channel, an Active or Inactive state indicator (green for active, grey for inactive, red for failures), and Updated and Last run columns. The list sorts by Last run by default, with a pinned name column so it stays visible as you scroll.

Bulk actions: Select any Automations and a bar appears with a count, a Select all across pages option, and Turn on, Turn off, and Delete buttons. Bulk changes process up to 200 Automations at a time. If only some of the selected Automations could be updated, you see a clear count of what changed and what didn't.

In the Automation Canvas: The header carries an Active/Inactive switch, Test run, Save, and a menu for less-common actions. Steps appear on visual tiles with a unified color system.

Terminology changes: "Recipes" are now "Templates." States are Active and Inactive; the row actions are Activate and Deactivate. Re-engage sequences follow the same language: a live sequence is Active, a switched-off one is Inactive, and the row action is Deactivate. Sequences held unstarted remain Draft.

Run history, Re-engage sequences, and Templates all use the updated toolbar, breadcrumbs, and drawer.

The Automations performance report at Reports → Automations performance is rebuilt as a native report, replacing the previous embedded analytics view. It loads in about 1.5 seconds; on large accounts with a 14-day window, it previously took 33–38 seconds.

The report is paginated and sortable by automation name, total sends, open rate, click rate, bounce rate, and unsubscribe rate. You can filter by folder, automation, status, and date range, include or exclude Apple Mail Privacy Protection opens, and export results to CSV. The default quick date range is last 14 days (previously last week), and an "all time" option has been added.

Exporting to CSV opens a modal that polls until the file is ready and downloads it in place — no redirect required. The export respects the active table filters, including Automation ID filtering that was previously dropped. If the export takes longer than 10 minutes, the modal points to the Import/Export status page.

Group permissions for non-admin users are respected.

The rebuilt Automations Performance report is paginated and sortable by automation name, total sends, open rate, click rate, bounce rate, and unsubscribe rate. You can filter by folder, automation, status, and date, choose whether to include or exclude MPP opens, and export results to CSV. Group permissions for non-admin users are respected.

The contact lookup in the Automation path view's Review Runs selector paginates, so contacts are no longer missing from the list on accounts with large contact databases.

When an Automation run ends because of a manual action replay, the Automation history path view shows a clear, translated label explaining why. Previously, this end cause displayed as blank.

Previously, a stale flag carried forward from a prior attempt could cause a contact to pass through a later goal without the condition being checked again. Goal conditions are evaluated fresh on every retry. Normal goal processing for contacts not on retry paths is unchanged.

Previously, the wait-until field did not offer minute-level precision. You can now set a wait step to end at a specific minute offset, not just at the hour.

In the Automation Canvas path view, a run's status appears as a colored chip in the header, and completed runs show the date they finished alongside the date the contact entered. The status colors match the Review Runs dropdown, so both surfaces stay consistent.

The "Drafting content…" loading indicator in the Automation Canvas works correctly again for async blocks, so newly updated blocks display their loading state as expected.

Automation Canvas: Connector lines between steps draw straight when the step ports are aligned. Every connector — including Add step stubs and fork arms — ends in a consistent chevron. The "Add step here" gap closes when the pointer moves outside it.

Inbox read receipt: The read receipt indicator leads with a double-check icon. "Seen" on Instagram reads "✓✓ Seen". The double-check shows on every channel that supports read status; SMS shows no tick because the channel doesn't provide one.

When you ask the Automation Canvas assistant what a Wait, If, Jump To, or start-trigger condition means, it describes the condition using the same human-readable sentence the builder displays — not just a numeric ID. This means the assistant can accurately explain what a condition does without you having to look it up yourself.

Two existing API v3 endpoints — GET /automations/{id} and GET /automations/{id}/blocks — are now documented in the public spec, so tools that answer "what is in this automation?" can return accurate metadata and block-level detail rather than treating a list row as the full answer.

When the app runs inside a cross-origin frame — for example, embedded in a Google browser surface — a session expiry previously caused the page to stop responding rather than redirect to login. The redirect targets the top window correctly, so login always happens at the right place, even in embedded contexts.

Anonymous email opens are recorded by count and trigger webhooks per open. Previously, these batched anonymous opens were not counted or acted on individually.

Each anonymous open batch updates your campaign open counters accurately. Webhooks fire once per batch, after a successful write, so you won't receive duplicates if a retry occurs.

Anonymous opens do not trigger Automations.

When duplicating a Campaign with the "refresh" or "design" option, Active Intelligence shows topic suggestions directly in the preflight form, under an "Ideas for you" header. Click any suggestion to pre-fill the topic field, or start typing and the suggestions step aside.

Suggestions are drawn from your account's previous Campaigns, website content, and calendar events, then ranked for relevance and freshness. They load in the background while you review the form, so the form opens at full speed regardless.

This feature is rolling out gradually.

When you ask Active Intelligence about connecting a tool, it checks both MCP connectors and classic app connections in the same lookup and presents whichever options apply — already-connected ones appear first. Active Intelligence won't offer setup steps for something you've already connected.

• Connectors are enabled via the Tools button under the message input.
• App connections are managed on the Apps and Integrations page.

If the connector catalog is temporarily unavailable, Active Intelligence reports that connector status is unknown rather than incorrectly showing a connector as disconnected.

After Active Intelligence generates a first branded Campaign for an account that hasn't committed to a Brand Kit, it presents a set of other brand looks to consider. Each suggested card includes a recommended flag, supporting evidence, and a kit preview.

Selecting a look records your choice and clears the suggestion block. The suggestions appear only once per account and only when two or more brand candidates exist with none yet committed.

Active Intelligence's site summary reads content from your main page and up to five important pages, with each page capped at 20,000 characters independently. Recent site changes are also included, with the newest changes kept first.

Previously, the entire prompt was truncated at 20,000 characters total. Because important pages come after the main page in the prompt, they were silently dropped before the model ever saw them — meaning your site summary was effectively built from your homepage alone.

The label change is visible in the Active Intelligence preflight panel. No configuration is required.

When you choose "Duplicate with AI" from the Campaigns list or "Reuse a design" from the Start-from drawer, a short preflight form appears in the Active Intelligence sidebar before building begins. You answer a few questions about the source Campaign, then submit — Active Intelligence uses your answers alongside the Campaign context to shape the result, so the output reflects your intent rather than making assumptions about it.

• Duplicate with AI: the Active Intelligence sidebar opens with the preflight form in place of the landing screen. Submitting sends your answers with the Campaign context. Cancelling closes the drawer without sending anything.
• Reuse a design: picking a past Campaign opens a second step in the Start-from drawer, with a Back button if you change your mind. Submitting adds your answers to the existing turn. Templates keep their one-click behaviour and are unaffected.

If the preflight questions are unavailable for any reason, both flows continue exactly as before — no answers are sent and nothing is blocked.

When Active Intelligence is drafting email content for send steps in an Automation Canvas, the preview panel displays a writing spinner on those steps while generation is in progress. This gives you a clear signal that content is being written rather than leaving the step appearing empty or static.

When you ask Active Intelligence to build a segment using specific lists or tags, it builds from what it can find and tells you plainly which ones it couldn't match. The response names each unresolved item directly, so you know what the segment was built without and can decide whether to refine your request.

Previously, unmatched lists or tags were silently dropped — the segment was built and saved without any indication that part of your request wasn't fulfilled.

When Active Intelligence builds or edits a Campaign, its replies use the actual names of your lists, segments, and Campaigns. All four paths that could surface IDs in replies have been closed. When a name can't be read, the reply stays worded without it rather than falling back to an ID.

When your Active Intelligence memory storage is within 250 characters of its 1,000-character limit, Active Intelligence warns you and asks you to delete or shorten a memory before saving a new one.

Previously, Active Intelligence could confirm that a new preference was saved even when it was actually dropped because the storage limit was exceeded. You'd only discover the preference was lost later, when it wasn't applied.

A small blue info icon appears beside the widget's "Updated" timestamp; hovering or focusing it shows the label and an explanation of what the widget could and couldn't match. In edit mode, the note appears as a chip in the action row and clears as soon as you start editing the prompt.

The only case that still returns no widget is when there is genuinely no contributing data source to draw from.

Previously, some of these dimensions were absent or sparse in content analysis. Each recommendation is required to name the specific figure or account pattern behind it — so insights are grounded in your actual data, not general observations.

Engagement-performance and automation-health insights are unchanged.

When a chat run was cancelled while its request was still in flight, it could continue streaming alongside the replacement run, mixing content from two separate runs in the same thread.

A cancelled run stops before it starts streaming. It no longer resets the chat phase under the run that replaced it, and a superseded run's retry no longer interferes with the active run. The conversation thread stays clean.

When you attach files in the Active Intelligence chat, a warning appears as soon as you reach the attachment limit.

You can ask Active Intelligence to combine two existing segments into a single combined segment. Specify two segments by name or ID, choose how to combine them (OR by default, or specify another operator), and pick what to do with the result: save it as a new segment, update one of the two in place, or preview the combined audience without touching either original.

Active Intelligence reads both segments' full condition sets, joins them, and never deletes anything. If combining under a given operator would change the meaning of either segment, Active Intelligence will tell you which operator to use instead rather than writing something unexpected.

The Active Intelligence chat's Add context dropdown and @ menu display the correct "Your Brands" group heading, a "Default" tag next to the account's primary Brand Kit, and a replacement notice when swapping a context chip. These labels are localized for all supported languages.

The advanced filters in the Accounts Overview match the columns shown in the accounts table. The "contains" and "does not contain" filter operators are also available.

If best-send data isn't available for your account, the behavior is unchanged.

Active Intelligence in the Automation Canvas and Automations overview leans toward taking action when your request is vague, rather than stopping to ask clarifying questions.

When you ask a WhatsApp question from the Deals CRM context, Active Intelligence routes it to WhatsApp campaign generation and presents the native WhatsApp integration first. Common misspellings — "whatsap", "watsapp", "wasap", "wpp" — are recognized and routed correctly. Previously, the Deals CRM context caused Active Intelligence to answer directly and recommend third-party tools without mentioning the native integration.

When you ask Active Intelligence what happened to a named contact in an Automation — why a step skipped them, or what they did at each stage — it pulls that person's per-step event history and skip reasons. Previously, those questions returned account-wide aggregates that couldn't tell you what happened to one specific person.

Questions about a specific contact (by email or phone number) route to the part of Active Intelligence that can see individual event records. Questions about Automations with no contact specified — like listing all your Automations, or asking why contacts generally get skipped — continue to return account-wide answers as before.

One gap remains: if a contact never entered an Automation at all (as opposed to being skipped inside one), Active Intelligence doesn't yet cover that case.

When Active Intelligence picks an audience for a campaign, saved segments are ranked by how well their rules match the campaign brief, how often they've been used in recent campaigns, their reach, and recency — before any cap is applied. Search matches rule text as well as segment names. Lists rank by name fit, then active subscriber count.

Every value requested for a composed audience segment is checked against what was actually saved. If a value was dropped during composition, the error names each missing value so the issue is visible rather than silent.

Composed segments created during a turn that never attached to a campaign — because the turn failed, asked a clarifying question, or was retried — are deleted when the turn ends, so orphaned saved segments no longer accumulate on your account.

Run status is also more accurate: a run that ended on an error is recorded as failed rather than complete. Cancelled runs stop generating content promptly, so a cancelled run stops billing. A run that has already committed a campaign is left to finish.

Widgets are only held as "coming soon" when Active Intelligence has genuinely looked for data and found none yet. A prompt it can't fulfill — for example, a nonsense string — is declined on the spot, so your Reports page isn't cluttered with widgets that keep re-running without producing anything.

Active Intelligence no longer stops a Campaign turn when it hits an unrecognised response format. You get a readable response rather than a failed turn, and you can check the Campaign preview to see where it stands.

Clicking "Your Dashboard" opens the agent homepage directly at the My View tab. Clicking the Reports heading itself still lands on All Campaigns Performance, unchanged.

Loading the conversation timeline for a Contact with a large WhatsApp history completes reliably, without very long load times. Previously, a Contact with an unusually large history could cause the timeline to take over a minute to load. The timeline returns the conversations collected up to a 50-page limit; if that limit is reached, you see the conversations gathered so far.

Previously, the webhook fired once per list the contact was on, plus an additional legacy call — producing multiple events per complaint.

One thing to be aware of: a webhook scoped to a list other than the sending list will no longer receive the event, where it did before. If you have webhooks scoped to specific lists for spam complaint tracking, check that they're scoped to the list the Campaign was sent from.

Previously, a two-card row with one unread card produced a mismatched layout. The side card now matches the wide card's height using the same stacked layout as a three-card row. Rows with three cards, two or more unread cards, or all-read cards are unchanged.

Sending an update request with a new template name saves that name.

Behavior details: an explicit name in the request body takes priority over any name copied from a source Campaign or template. An empty name is ignored, keeping the existing name. A non-string name returns a 400 error before any changes are made.

This update covers the name field. Other documented fields — subject, content, category, editor version, additional data, and user — are unchanged by this release.

Sending an update request with a new template name saves that name.

Behavior details: an explicit name in the request body takes priority over any name copied from a source Campaign or template. An empty name is ignored, keeping the existing name. A non-string name returns a 400 error before any changes are made.

This update covers the name field. Other documented fields — subject, content, category, editor version, additional data, and user — are unchanged by this release.

Automation runs that became stuck during the incident window of October 4, 2026 at 9:00 PM through October 5, 2026 at 5:31 PM (Central) have been recovered and replayed. Contacts who were stuck in an Automation during that window will resume advancing through their journey.

The Site Tracking domain list previously surfaced an internal system record alongside your customer-added domains. That entry is filtered out of the list and cannot be deleted through the API.

Shopify is retiring the script tag method that Forms used to appear on Liquid storefronts. Forms on Shopify stores move to a theme app embed that you enable in the Shopify theme editor.

The Forms ecommerce tab shows whether the embed is active, inactive, or unknown, with a "Turn on in Shopify" button and a "Check again" button so you can confirm the status after enabling it. Status shows as unknown when it can't be verified — for example, on a password-protected storefront.

During the transition, if both the old script tag and the new embed are running at the same time, a guard prevents your Form from appearing twice on the page. This only applies to Shopify storefronts; all other sites are unaffected.

Action required: To keep your Forms working on Shopify, enable the ac-forms theme app embed in your Shopify theme editor, then check the status in the Forms ecommerce tab.

The Segment menu in the Active Intelligence Campaign send panel shows each saved segment's size from its last run, displayed as active contacts out of total along with the date it was measured. If no measurement date is available, the row prompts you to refresh the segment on the Segments page.

List segments show their name only, as they carry no count data. If your account has no saved segments, the picker shows an empty list.

The segment card uses compact view whenever the side panel drawer is open, keeping the layout usable regardless of which path you're on.

Scheduled Tasks start and repeat on the schedule you actually set.

Task start date and cadence description. When Active Intelligence creates a task with the default one-week cadence, the task starts tomorrow rather than waiting until the following Sunday. The cadence description shown after a task is created is taken directly from the stored schedule, so what you read matches what was saved.

Recurring report scheduling. When Active Intelligence offers to schedule a recurring report, it derives the cadence from your original request using a clear order of priority: your acceptance reply wins first (for example, "yes, make it weekly"); if absent, the recurrence in your original request ("every quarter" → 3 months, "last month weekly" → 1 week); then the reporting window you named ("last month" → 1 month, "last 7 days" → 1 week, "last quarter" → 3 months); and finally a default of weekly when nothing else is specified. The offer wording itself never sets the cadence.

Previously, accepting any recurring-report offer always created a weekly task, because the scheduling step ran without conversation context and defaulted to weekly regardless of what you had asked for.

Two improvements make Scheduled Tasks more discoverable and actionable.

Clearer status at a glance: Each Scheduled Task card shows a distinct chip color and icon for its status — Draft, Active, Paused, or Done. Non-done cards gain an actions menu with Edit, Activate, Pause, and Delete options gated by the card's current status. Done cards retain a plain View button.

Better onboarding card order: The Scheduled Task idea card in the onboarding panel is positioned as the third visible card, after Brand Kit and Google Ads. The weekly performance report card is fifth. This ordering is stable across different account states.

Active Intelligence chat in split view: In the narrow split-view chat, the four chat action buttons are icon buttons with hover Tool Tips so the Schedule task option fits on the row without truncation. The home page and full chat keep labeled buttons.

Scheduled Tasks appears in the main navigation at its own URL for accounts with the updated navigation. Accounts still on the previous navigation continue to find it in the drawer as before, and any saved links to the old URL redirect automatically.

The Salesforce integration preserves currency, number, and text field values of zero during hourly account syncs, so a $0.00 currency value no longer causes a sync error.

When you fix and resubmit an SMS campaign registration that was rejected, compliance can approve it again so it reaches the carrier. Previously, resubmissions were blocked, leaving the registration stuck showing the old rejected data.

SMS broadcast approvals are saved reliably

Three changes make SMS broadcast approvals reliable. First, the save retries automatically before reporting a failure, so a brief connection hiccup doesn't stop the broadcast. Second, if the outcome is genuinely uncertain — for example, after a connection timeout — you're told to refresh before retrying, rather than receiving a generic error that gives you nothing to act on. Third, if the save succeeded but a later step failed, the broadcast is correctly recorded as approved.

If you have broadcasts currently stuck in "Pending Review," refresh the page and check their status.

The Automations tab in SMS reporting displays six KPI tiles: Credits consumed, Delivery rate, Click rate, Reply rate, Fail rate, and Opt out rate. Credits consumed counts successful sends only. The Campaigns tab retains its existing seven tiles unchanged.

Previously, the KPI tiles and block list were scoped to your chosen range, but the per-block rows showed all-time figures. Selecting "View all time" or clearing the range sends the correct parameters throughout.

Omitting both start and end date parameters returns all-time data, which was the previous behavior. Malformed dates, impossible dates, or an inverted range return a descriptive error.

The Send Now button on the resend-to-new-contacts flow sends your Campaign as expected.

As part of this fix, the form's validation runs when you click Send Now. If the source Campaign has a blank subject line, you'll see a "subject cannot be empty" alert before anything is sent — previously, a Campaign with a blank subject could be sent without warning.

Trend tabs write a metric parameter to the URL, making individual views linkable. The channel filter is multi-select and scopes the chart. The existing By Ad / Reel table is retained at the bottom.

Dynamic registration Forms support a new radio button field type. When a field in the registration form has options defined, it renders as a vertically stacked radio group with proper accessibility attributes. The review step shows the translated option label for radio selections. Helper text and error text follow the active color theme.

Email Product Catalog blocks that select products by filter display product images. When no base image is available, the block uses the first available variant image.

Base product images still take priority: if a base image exists and has a valid URL, it's used. If the base image URL is empty, the block falls back to the first available variant image. If neither is available, the placeholder appears as before. Products that already rendered an image continue to render the same image. Manually selected products are unchanged.

Password reset links work for admin accounts whose password record holds a legacy placeholder value from older installations. Previously, these accounts received a rejected reset link. All admins can complete a password reset regardless of how old their account is.

The PATCH /conversions/{id} endpoint is now documented in the API contract, matching the parameters, request body, and responses already in place for PUT /conversions/{id}. The bulkCreate operation's documented success status is also corrected from 200 to 201, with the actual response shape now specified.

The notice updates if you change the recipient. This is rolling out gradually and may not be visible in your account yet.

For contacts who have opted out of open tracking, 1:1 emails sent through a connected inbox go out without an open-tracking pixel. If an open is reported for an opted-out contact — including on emails sent before the opt-out was set — it is discarded: no activity is recorded, no deal activity is created, and no Automation open trigger fires. Click and reply tracking are not affected.

The Contact detail page gains a native Open Tasks panel in the right rail, appearing after Recent Activities and before Deals. The panel is headed "Open Tasks (n)" with an "Add a task" action.

Each task row shows type and title, due date (in red when overdue), and assignee. Tasks linked to a deal show which deal they're from. A complete checkbox triggers an outcome form when needed, then drops the count and refreshes Recent Activity. A completed-tasks toggle appears at the foot of the panel. Tasks can be edited or deleted through a shared modal; delete requires confirmation.

The panel handles the "Add a task" deep link and the header menu action. Clicking a task in the Recent Activity stream opens the panel's edit modal. The panel is capped at 100 tasks and shows "Showing x of y" when the cap applies.

When a non-admin user tries to delete a list via the API, the response status code correctly indicates a permissions issue. Callers checking the response body see no change.

Selecting the last remaining conversion in the goal settings drawer works correctly. Previously, picking the final option in the conversions dropdown could cause a rendering loop.

The one-time password email sent during an unusual login challenge reliably triggers Gmail's "Copy code" verification-code card. The block order of the email is fixed to maintain this behavior across future changes. A labeled line above the six-digit code lets Gmail surface a one-tap copy button on Android and iOS. Desktop email clients show one additional bold line above the code box — everything else is unchanged.

The global navigation renders correctly even when account data is missing expected fields — such as group membership, language settings, or user information — rather than collapsing to a placeholder state.

The GET /deals/{id} reference previously listed only three fields and used a limited example schema. The documentation now covers:

• The full deal record structure
• The 403 response returned when a user doesn't have permission to view deals they don't own
• The restricted-pipeline response shape
• Both 404 cases: a deal that doesn't exist, and a deal with neither a primary contact nor an account
• The contact field, documented as nullable — a deal requires either an account or a contact, but not both

No API behavior changed; only the reference documentation was updated.

All 31 query parameters accepted by the GET /automations API endpoint are now declared in the API specification. Previously none were declared, which caused API clients built from the spec to reject valid parameters before requests were even sent.

Parameters now documented include limit, offset, ids, search, active, has_message, label, filters, orders, and more. If you use an API client generated from the spec — including the ActiveCampaign MCP server catalog — valid parameters that were previously refused will now go through correctly.

Form submissions that arrive with a malformed or unsupported character set are processed rather than silently dropped.

Form submissions that triggered a slow or unresponsive avatar lookup could stall the contact redirect for 20–80 seconds. The lookup timeout is capped at 1 second.

A timeout leaves the contact's avatar status unchanged for retry on the next submission or preference-center visit, rather than permanently marking the contact as having no avatar after a single slow response.

The Form builder opens without error for Forms whose custom field data arrives in a keyed format rather than a standard array.

The Engage Insights page has been restyled to match the current design, with every panel present in a fixed order: KPI cards, By channel, ActiveCampaign, Delivery, Trends, Top links, Automations, and four breakdown cards.

A range switcher, All channels filter, and Export action appear in the header. The Daily trend tabs are linked to a URL parameter, so a specific tab is restored when you navigate back or share a direct link. Each panel has its own loading skeleton, and zero values display as 0, $0, or 0% rather than appearing blank. The layout reflows at three breakpoints to fit different screen widths.

Report counts are preserved when contact or asset names can't be read, CSV headers are aligned, and trend tooltips include locale-aware day labels.

A dedicated Emails tab is available on the Contact record page. It lists all emails associated with a contact — one-to-one sends, deal emails, and Campaign sends — newest first, 25 per page with a Load more option. Each row shows the subject, sender and recipient, and a relative date. Opening a row displays the full email in a side drawer with from, to, cc, date, and body. You can reply directly from the drawer, and a "Send an email" button opens the composer. The list refreshes after a send.

The ecommerce overview dashboard is available again, with all navigation entries and revenue reporting functionality restored.

Two improvements shipped together: short result sets no longer trigger a redundant database query to calculate totals, and the underlying query runs a lighter sort on the default view. The result is faster load times across search and pagination with no change to how results appear.

When creating a new pipeline, you can choose from 13 industry-specific templates — filterable by industry, such as Software or Real Estate — or define your own custom stages. Stages can have custom titles, colors, and ordering. Up to 50 stages are supported. If you don't specify stages, the three default stages are created as before.

The Contacts widget on a Deals CRM record loads correctly for users who don't have full deal list access. The widget displays the contact as expected.

The GET /api/3/deals endpoint accepts the same parameters it did before a recent change that introduced regressions. Large or complex Deals queries complete as expected.

Separately, GET /api/3/deals requests that combine a large limit with multiple include relations return a clear HTTP 400 response stating the maximum allowed limit for the number of relations requested, rather than failing with a memory error. Requests with fewer than two relations in include are unaffected.

GET /api/3/deals requests that combine a large limit with multiple include relations return a clear HTTP 400 response stating the maximum allowed limit for the number of relations requested, rather than failing with a memory error. Requests with fewer than two relations in include are unaffected.

When editing a deal's status on the Deals CRM side panel detail page, the search field that previously appeared is gone. Status editing works without it.

Datetime custom field values in Contact profiles display the correct time regardless of which time zone you're in. Previously, the value shown could be off by an hour for viewers whose local time differed from UTC.

A small blue info icon appears beside the widget's "Updated" timestamp; hovering or focusing it shows the label and an explanation of what the widget could and couldn't match. In edit mode, the note appears as a chip in the action row and clears as soon as you start editing the prompt.

The only case that still returns no widget is when there is genuinely no contributing data source to draw from.

Four improvements ship alongside the fix: failed AI image description calls now wait before retrying rather than retrying immediately. Picking a Content Manager image in the Campaign editor preserves its alt text. Opening the Content Manager picker multiple times no longer causes each pick to fire more than once. Templates saved by Active Intelligence are visible in the template library.

Previously, those contacts could slip through the checks and be subscribed even when they should have been blocked. The check now runs correctly for all email addresses regardless of special characters.

Contacts entering two Automations with "Add lead or contact to Salesforce Campaign" blocks at the same time could get stuck permanently and never advance. Those Contacts complete as expected. This works with the current released Salesforce package versions without any change on the Salesforce side.

Contact create, update, and sync operations accept a channel identities block, allowing a Contact to be identified by a platform-specific identity — such as Meta's per-business user ID — in place of an email address or phone number. This supports handling of Instagram and Messenger DMs where many people have no email or phone on record.

Identities are handled like tags and lists: all three write operations accept the block, an ownership check runs before any write, and identities are attached immediately after the Contact is saved. Every write response includes the Contact's identities so the correct row IDs are available. A conflict during a concurrent request returns an error; the next attempt resolves to the winning record through the identity lookup.

Searching for Contacts from the admin site search bar and following the link to the Contacts page returns results that match your search term. Previously, the search term was silently dropped, so the Contacts page showed your full contact list instead of the people you searched for — meaning any bulk edit you applied from those results affected all your Contacts, not just the ones you intended.

If you have bookmarks to Contact search results from the old search bar, those links also resolve correctly.

Contact scores no longer dip and recover when the segment engine is briefly unavailable. Previously, a throttled or failing engine call caused the affected scoring rule's points to be dropped from every contact in the batch, then restored on the next pass — causing scores to cross thresholds in both directions within the same minute.

The scoring engine now distinguishes between two situations: a missing segment (the rule is skipped, but the batch is still written) and a transient failure (the batch is left unwritten and retried on the next pass). Your scores stay stable either way.

A contact who previously unsubscribed and then resubscribes through a Form — including third-party integrations — is reactivated on the lists included in that Form.

For Forms with double opt-in, the contact receives the opt-in email, and confirming the link activates the list and triggers the expected subscribe Automations and webhooks. For Forms without opt-in, the list is activated immediately. Only the lists associated with the Form are affected; other lists the contact belongs to are unchanged.

Two issues on the contact page are resolved. The Custom Objects panel threw an error when the schemas request failed, affecting approximately 257 crashes and 160 accounts per day. The notes panel threw an error when the document body was null, affecting approximately 40 crashes and 25 accounts per day. Both are resolved.

After completing a contact import, clicking Continue in the onboarding flow finishes onboarding directly — the same as clicking Close. The Brought over review screen remains accessible, and the splash sample report is unchanged.

The Details panel in the inbox conversation rail shows the full set of fields from the linked ActiveCampaign contact record: standard fields (first name, last name, email, phone, account, job title), contact Scores, and Custom fields you have permission to view. Permissions are applied before the data is returned, so you only see fields your access allows.

If the contact record isn't linked or can't be read, the rail draws from the conversation identity as before. The contact record fills in a missing email or phone, but the identity's own values take precedence.

The Details panel in the inbox conversation rail shows the full set of fields from the linked ActiveCampaign contact record: standard fields (first name, last name, email, phone, account, job title), contact Scores, and custom fields you have permission to view. Permissions are applied before the data is returned, so you only see fields your access allows.

If the contact record isn't linked or can't be read, the rail draws from the conversation identity as before. The contact record fills in a missing email or phone, but the identity's own values take precedence. Date fields read as the named day rather than UTC midnight; choice values arrive as their labels.

A replay action card is available in the Automation Canvas. You can add it to any Automation the same way you add any other action card.

Both search parameters are now accepted, matching the behavior of the Contacts list.

Dynamic registration Forms render conditional fields correctly. A field appears or disappears based on the value of its controlling field — which can be on any step of the form, not just the current one.

• Variant labels, placeholder text, and helper text update to match the controlling field's value.
• Half-width fields are never paired with a hidden neighbor, so you won't see stranded half-empty rows.
• The required indicator reflects conditional required rules.
• The review step and saved drafts omit hidden fields, matching exactly what gets submitted.

Australian and Danish registration flows are unchanged.

Previously, this content was missing from the preview. The designer "Preview Email" path and the live-send path are unchanged.

One limit applies: if a contact received the same campaign from multiple automation runs, the preview uses the latest run, because the activity log does not store which specific run sent each email.

Emails built in the campaign designer keep their chosen fonts when recipients open them in Outlook for Windows. Previously, a font stylesheet link in the email HTML was blocked for unauthenticated recipients, causing Outlook to fall back to Times New Roman.

This fix applies to Campaigns saved going forward. Mail already sent is unchanged. Scheduled and Automation sends will use the corrected font links the next time the Campaign is saved.

Campaigns waiting on approval no longer lose their approval status when saved after this change, as long as no one edited the content.

When a mail server permanently rejects the sender or recipient address on a Campaign email, that Contact is recorded as a failed send along with the server's rejection message. Your send records reflect what actually happened at delivery — including the server's reason for the rejection.

Accepted envelopes continue to be reported as sent.

Campaigns whose tracked links had been removed were failing validation on every send attempt and falling back to a slower sending path, blocking roughly 11,700 send attempts per week in the US.

Those Campaigns send correctly. Link state is saved before validation runs, so removed links are cleared first and the Campaign passes cleanly.

Previews and test sends no longer track links that were deleted from a campaign. Previously, deleted links remained tracked during the link refresh step for previews and test sends, causing approximately 67,000 preview and test send requests per week in the US to be rejected with an error. Removed links are now marked stale the same way they are for real sends. System links (web copy, forward, update, unsubscribe) and social links are preserved.

The arrows previously appeared as bare chevrons with no background, making them hard to see and click against an email preview. They render with a bordered surface.

Three Campaign API behaviors are corrected:

• Scheduling a Campaign with a send date writes to the column the sending system reads, so the Campaign sends at the time you specified.
• Enabling a Campaign is refused when the send date is missing, unreadable, or more than one day in the past — returning a clear error instead of creating a Campaign that silently never sends.
• Sending an empty list of recipients removes all sending lists as expected.

A date-time value with an explicit offset is used as-is; a value without one is interpreted in the account's time zone and refused if that time zone is not set. Impossible dates such as February 30 are refused. Recurring and RSS Campaigns are unaffected.

API clients that previously sent past or unreadable send-date values will receive an error going forward.

Bulk-editing segments of 500,000 or more Contacts — adding tags, lists, or automation blocks — completes without timing out.

Removal operations (removing from tags, lists, or automations) are unchanged.

When both a segment filter and a text search are active on the Contacts index, the "Edit all" button is disabled and shows a tooltip explaining the restriction. The combination is also blocked at the API level, regardless of how the request arrives.

This restriction is permanent: counting affected Contacts at scale with both a segment scope and a text search is prohibitively expensive, and the operation was already producing unpredictable results. Selecting specific Contact IDs and then clicking "Edit" is unaffected — an explicit selection overrides the segment and search scope.

The "Learn more" link in the contact import connect screen and the first-login key capture step has been removed. The link previously pointed to a documentation page that returns a 404 error.

The Branding header in the form editor stays anchored above the form while you scroll, so it remains visible and accessible throughout your editing session.

Brand monitoring previously failed to scrape important pages whose URLs used plain HTTP, logging approximately 830 errors per week. Those pages are upgraded to HTTPS before scraping, so they're included in brand history runs. Both newly discovered pages and already-stored important-page records are affected.

Active Intelligence brand monitoring retries website images that previously could not be queued for processing due to temporary infrastructure errors such as DNS timeouts or expired tokens. Previously, these images were marked as permanently failed and never re-dispatched, meaning brand logo and image data could go missing after a transient error. Retried dispatches share the existing dispatch budget. Images that failed during analysis, upload, or timeout, and skipped images, are still excluded from retry.

The Brand Kit tone evaluation accepts well-formed prose descriptions that use em-dashes, semicolons, colons, and quoted brand terms. Previously, these were read as markdown formatting rather than standard prose punctuation, and a valid description could be rejected incorrectly.

Sentences are counted by terminal punctuation only — a sentence with multiple clauses joined by an em-dash, semicolon, or colon counts as one sentence. The acceptable sentence range is 4–8, and length is only flagged when the count is unambiguously outside that range.

Brand Kit images whose URLs redirect — for example, a CDN moving a file to a new path — are fetched successfully instead of being marked as failed. Previously, all redirected image URLs failed during scraping.

Between 21–28 September 2026, approximately 1,360 of 17,658 scraped image URL validation warnings were caused by redirects. Brand Kit follows up to three redirect hops, checking each hop for security risks before making the request. Redirects to non-HTTPS URLs or internal addresses still fail. All existing image size, format, and dimension validation is unchanged.

On accounts with more than one Brand Kit, each logo file is named after its brand rather than using a generic name like logo-light.png. Previously, every kit's images shared the same generic names, so looking up a logo by file name returned the most recently uploaded kit's image instead of the correct one. This affected 3,487 accounts with more than one Brand Kit. Existing files keep their current names, and nothing that reads a stored kit changes.

When building a Brand Kit from a website, ActiveCampaign retains your company's logo even when the only available artwork is a vector (SVG) file. Previously, SVG logos were skipped and no logo was stored. Raster formats (PNG, WebP, JPEG) are still preferred; SVG is used only when no raster version is available. This applies to the brand-logo import path only.

Brand Kit scraped images whose URLs contain special characters — including template fragments, spaces, soft hyphens, zero-width spaces, and non-ASCII path segments — receive metadata and display correctly. Previously, these images were marked as failed or dropped during verification. Images already saved are not re-uploaded when the Brand Kit is refreshed.

Brand Kit image scraping covers your main page and all important pages discovered during the crawl, not just your home page. The "Refresh images" button runs an images-only crawl of those same pages, up to 500 URLs per run.

Images are reconciled on each run: new images are added, unchanged images are kept without re-uploading, and images that have disappeared from your site are removed from the Website Images folder. Removed files remain in Content Manager but are unfiled from that folder.

When you change your website URL in Brand Kit, previously scraped image records are cleared before the new scrape begins, so old and new content don't mix.

Images from Shopify stores and URLs with special characters — including international domain names — are imported correctly.

Images the AI model rejects are marked invalid immediately rather than being retried over and over.

A timing issue with the 60-minute grace period is also resolved. Previously, a race condition could cause a Brand Kit image record to skip its full 60-minute waiting period and jump straight to a failed state. Records wait the full 60 minutes before being marked failed, giving processing a fair chance to complete.

When a Brand Kit competitor or tone enrichment job was refreshed, it could appear to fail almost immediately — even though the underlying work was still running normally. The job's age was being measured from when it was first created, sometimes 60–120 days earlier, so the system treated it as overdue the moment it restarted.

Jobs now measure their age from when they were last set to run. A refresh gives the job a full window to complete before any timeout is applied.

If a job was reset before this change, it may show as failed one more time. The next refresh will correct it.

If a Brand Kit's email templates fail to generate, opening the Brand Kit page triggers automatic regeneration rather than leaving placeholder text in place indefinitely. Regeneration is limited to once per 10 minutes per kit to prevent repeated attempts.

Previously, sites hosted on shared website builders caused Brand Kit to store the builder's domain instead of yours. Tone extraction and competitor research then ran against the builder platform, returning results about the hosting service rather than your business. Evidence from production showed 6 of 18 tone extraction failures in a recent week were caused by this.

Three fixes shipped together:

• Brand Kit keeps your requested domain even when a lookup resolves to a different platform domain.
• Tone extraction scrapes your website URL directly, so already-stored kits benefit too.
• Competitor research uses the host of your website URL, with a fallback to the stored domain.

Importing a URL also stores the domain correctly, so a second import of the same site is rejected as a duplicate rather than creating another kit.

A separate reliability improvement means tone extraction handles pages that return an error or where a tone can't be determined — these are treated as skips rather than failures.

Bounced contacts show a correct date in their list membership history. Existing contacts with a missing bounce date are also repaired automatically.

Scheduled insight runs that previously saved late — or required retries — due to large tool responses save on the first attempt. Between August 25 and September 25, 330 failed save attempts across 6 tasks on 5 accounts caused insight emails to arrive 10–40 minutes late.

The fix measures the size of the stored transcript before saving and trims only the largest tool responses — one at a time — until the item fits. Trimmed content is replaced with a marker that follow-up chat can read, so the conversation stays coherent. Insights whose transcripts are already within the size limit are stored exactly as before, with no trimming.

The path view in the Automation Canvas returns the complete set of blocks, including ghost blocks, so the view accurately reflects the full structure of your Automation.

Both entry points trigger the same monitoring prompt, so you can reach automation monitoring without leaving the canvas. This option is not yet enabled in production.

The entered and exited contact counts on your Automations reflect the true running totals. Previously, saving an Automation could silently overwrite those counters with stale values, causing the displayed counts to fall behind reality. The counters are updated correctly regardless of when or how often an Automation is saved.

New data points cover: Contacts who entered and completed the Automation during the period, current Contacts in the Automation, and a week-over-week comparison. Without an explicit date window, the report snaps to the last complete Monday–Sunday week. Automations with historical activity but a quiet recent week still produce a report.

Authentication redirects containing path separators are handled correctly, and the encoding used matches what the platform expects. A mismatch introduced in early October had caused redirect failures and prevented the authentication cookie from being set.

Each Campaign is counted once in the All Campaigns performance report, regardless of how many times the same label appears on it. Sends, opens, clicks, bounces, unsubscribes, and revenue figures reflect each Campaign exactly once.

Clicking the CSV export button on the All Automations report starts a background export using whatever folder and status filters you have active. When the file is ready, a success notification appears with a link to the import/export status page where you can download it.

CSV columns included: Name, Status (Active/Inactive/Paused), Contacts, Campaigns.

The export button is part of the native All Automations report, which is rolling out gradually.

The Reports view for all Automations shows each Automation's name, status (active, inactive, or paused), unique subscriber count, and Campaign count. You can filter by folder, Automation ID, or status, and sort by name or subscriber count. Hidden Automations are not included, and each user sees only the Automations their group has access to.

Active Intelligence discloses when it applied your brand to a generated Campaign, naming the source of the brand choice once the engine confirms the brand survived into the final output. A Rebrand action is available whenever a brand was applied automatically or from an existing Brand Kit, or when no brand exists — letting you pick, keep, or adjust the brand on the same turn without starting over.

If a brand was found while a Campaign was still being built, the next successful generation leads with a note that your brand has been found since your last email, shown once per newly confirmed brand.

Active Intelligence's site summary reads content from your main page and up to five important pages, with each page capped at 20,000 characters independently. Recent site changes are included, with the newest changes kept first.

Previously, the entire prompt was truncated at 20,000 characters total. Because important pages come after the main page in the prompt, they were silently dropped before the model ever saw them — meaning your site summary was effectively built from your homepage alone.

Summaries also display at their full length. They were previously cut short at incorrect lengths.

Attached email files (.html, .xml, .eml, design export) that Active Intelligence previously couldn't open now load correctly. Files are located by their unique name within your account's uploads, regardless of how they were stored. If a file is genuinely missing, Active Intelligence asks for a fresh attachment rather than retrying indefinitely.

When Active Intelligence shows a preflight form during Campaign generation and you reload the page, the form reappears with the same context it had before the reload. Any answers you'd already entered are restored, and continuing from there follows the same path as answering inline. A form you had already dismissed stays hidden.

Switching tabs on the Active Intelligence homepage updates the address bar. Any tab can be bookmarked or shared as a direct link. If you open the homepage without a tab parameter, your last-used tab opens as before.

Reopening a system alert in the Active Intelligence notifications center returns to the existing chat for that alert, across any device. Previously, each time an alert was reopened, a new chat started.

Conversations started from a system alert are saved per user and restored when you reopen the same alert. If a saved conversation no longer exists, a new one starts automatically.

The Active Intelligence widget charts display correctly for users whose account language is set to Brazilian Portuguese. Previously, the charts failed and took down the agent page for those users.

When you ask Active Intelligence to create a Campaign without specifying a URL, it draws on your brand's own monitored site pages as source material. The generated email reflects your actual products, offers, and voice rather than generic content.

Pages are passed to the email composition service automatically, separate from any URL you explicitly attach. If brand monitoring is off or no pages are stored, behavior is unchanged.

Product images from your pages, not nav chrome. When harvesting images from a linked page, the engine skips navigation menus, headers, and footers before selecting pictures. Multiple renditions of the same photo collapse to the widest version.

Images from a named page outrank Brand Kit images. Pictures from a page you explicitly name are ranked above generic Brand Kit images, so off-subject kit photos are refused rather than placed.

The engine includes a guard against inventing content from site pages that wasn't in your brief.

When you import an Automation recipe and have access to Active Intelligence, the setup prompt appears in the Active Intelligence sidebar. If the sidebar is closed or a new chat is started while setup is in progress, the import drawer opens as a fallback so setup is never lost. For accounts without Active Intelligence access, the import drawer works exactly as before.

The setup card only appears on the builder page for the specific Automation being imported; opening a new tab or navigating elsewhere clears it. Per-step import cards after clicking "Get Started" still open in the legacy drawer.

Active Intelligence reads your effective SMS permissions and explains a limitation before attempting to generate SMS content, so you get a clear answer before any generation begins.

When Active Intelligence creates a Campaign, it sets the audience and the page respects that — no second write, no conflict. Previously, the page applied the audience again under different rules, which could produce a mismatch and show a false "send plan invalid" error. That false error no longer appears.

The schedule is still written by the page. A send whose only component is the agent-set sending address is accepted correctly.

Snapshot counts show the date of the last run; a snapshot with no recent count is reported as "no recent count" with a pointer to refresh it on the Segments page. The one segment attached to a Campaign in a given turn still gets a single live estimate, matching what the Campaign page itself does.

Before saving a composed segment, its condition tree is compared against the account's most recently updated saved segments; if an equal segment already exists, it is reused and nothing new is saved. A composed segment that leaves out a requested field value or tag is never saved, and the reply says what was left out.

Lists, segments, tags, fields, and the sending address are read concurrently, once per turn, reducing latency.

Brand monitoring previously checked only a brand's homepage for changes. When the homepage was unchanged but other key pages had changed, the result was "no updates recorded."

Brand monitoring now reads important-page history alongside homepage history, so changes across your brand's key pages are captured. It also includes important pages on your site's www host and subdomains — for example, news.example.com — when those pages appear in your site map. Previously, these pages were discarded even when the site map listed them, because the check required an exact host match.

Pages still excluded: subdomain or www URLs the site map did not return, parent domains, and look-alike hosts. The brand's own homepage is also excluded from the important pages list, since it is already scraped separately as the main page.

On a split-test Campaign in the Active Intelligence canvas, the "Split test" pill sits on the variant row, to the right of the variant tabs and "Add variant" button, rather than dropping onto its own line below the settings row. Non-split Campaigns are unaffected.

The weekly Automation monitoring report surfaced by Active Intelligence includes period-scoped Contact counts alongside lifetime totals.

New data points cover: Contacts who entered and completed the Automation during the period, current Contacts in the Automation, and a week-over-week comparison.

Without an explicit date window, the report snaps to the last complete Monday–Sunday week. A single request can compute both the current and prior period windows for comparison. Automations with historical activity but a quiet recent week still produce a report.

POST and PUT requests to API v3 endpoints that sent a double-encoded JSON body previously caused an HTTP 500 error. The API returns HTTP 400 with a descriptive message — for example, "Request body must be a JSON object, string given" — for double-encoded strings, bare arrays, numbers, and booleans. Every other request is unaffected.

If your integration sends requests to POST /api/3/contactTags and has been receiving errors, check that your request body is a JSON object rather than a JSON-encoded string.

API v3 calls that use a Bearer token or API token complete normally even when the session store is temporarily unavailable. These calls don't rely on session persistence, so they fall back gracefully. Browser and admin sessions are unaffected.

The expanded documentation covers read endpoints for contact lists, contact tags, field values, notes, bounce logs, tracking logs, contact automations, contact goals, and geo IPs. Notable behavioral details are now documented: contact automations pages only when a limit is sent; geo IPs is capped at 100 rows with no guaranteed order. Null-value handling is documented for a set of fields. No behavior in the API itself changed.

The pink/blue AI gradient header is consistent across all pages that use the shared AI prompt header — SMS, Forms, Segments, Business Goals, and Automations. The Automations list skeleton no longer flashes a pale-blue style before the header loads.

Previously, a blank date field caused the wait block to stall permanently — the step would never advance and the Automation would be stuck. The wait block skips the comparison cleanly when no date is present.

Sorting is handled server-side, so every Campaign in your account stays reachable across pages regardless of how many you have. The sort arrow on the column header shows the active direction.

In the Re-engage sequence detail drawer, each step's label and message copy appear on separate lines with clear spacing between them.

When you edit a segment by typing a new prompt in Active Intelligence, the segment card reflects the updated conditions immediately. The segment detail and audience queries refresh as soon as the segmentation response streams in, so what you see matches what you asked for without a manual reload.

On the automations tab of the SMS performance report, a new Automations dropdown lets you narrow the report to one or more specific Automations. The block list and KPI snapshot both update to reflect your selection. Your filter is saved in the page URL, so you can share or bookmark a filtered view. Switching to the campaigns tab clears the filter.

In the SMS builder, the Schedule/Send button is disabled for users who don't have send permission, with a Tool Tip on hover so they know who to hand off to. The builder itself is only accessible to users with campaign management permission; others are redirected.

On the SMS index, users without management permission see a read-only view: the Create button is hidden, row checkboxes and bulk actions are removed, and Delete/Duplicate are gone from the row menu. Draft rows show a "View SMS" button; sent and sending rows keep "View report."

This is part of the SMS Permissions rollout and is off by default. When it's enabled for your account, the behavior above takes effect automatically.

The SMS Automations report date filter previously scoped the KPI snapshot but not the row list. Selecting a date range with no activity showed every automation with blank metrics instead of an empty state.

The filter applies consistently to both the KPI tiles and the row list. When the selected date range returns no results, a clear empty state appears with recovery options. A malformed or inverted date range returns an error rather than silently falling back to a 30-day window.

Row-level metrics remain all-time figures and do not yet reflect the selected date range.

If your message variables contain HTML that the Visual editor couldn't represent — tables, image layouts, style tags, full document markup — opening them in Visual used to flatten that markup to plain paragraphs, and saving would overwrite your original code.

The Visual editor renders those blocks as editable sections in place. Your original markup is preserved exactly. Toolbar actions — bold, italic, underline, strikethrough, alignment, lists, indent, text color, block type, link, and clear formatting — all work inside those blocks. Blocks you don't touch round-trip without any change. Personalization tags, conditional content, and the rest of the toolbar behave the same regardless of content type.

Two things to know: Clear Formatting does not remove links — use the link popover's Remove button for that. Font size changes don't apply inside raw blocks.

The landing pages editor no longer shows a second scrollbar alongside the page editor.

In the form builder's field editor, clicking the "Required" checkbox label reliably toggles that field's required state for all field types.

When you submit a Denmark SMS number registration, the review step waits for the number to be fully provisioned before navigating to the hub. The hub shows the correct status immediately on arrival — ready when the number is provisioned, or a specific address or purchase failure reason if something went wrong — rather than a long pending state that didn't reflect what was actually happening.

On a successful registration, you land in the hub. On a failure, the review page stays open and surfaces the specific reason without implying a document rejection. Status chips and send-readiness indicators treat a provisioned number as approved and a failed one as an error.

The Deals panel sits in the right-hand rail of the contact detail page. It defaults to showing open deals and lets you filter by open, won, lost, or all. Pagination is built in, and an "Add deal" button creates a deal and immediately updates the panel and count.

Restricted deals show only their title and a no-permission note, and are sorted last in the list.

When potential duplicate deals are detected, a banner appears inside the deals list and scrolls with your deals rather than crowding the activity stream.

Accounts on a starter plan see an upgrade prompt with a link to the Deals upgrade page instead of an empty panel.

In the Deals CRM tasks view, the four task buckets — Overdue, No Task Assigned, Upcoming, and Scheduled — share fixed column widths. Contact Name, Account, Deal Title, Next Action, Deal Age, and Deal Value line up consistently across all buckets regardless of what's in each cell.

When a bucket has no tasks, it shows a clean header with a count of zero — no expand/collapse control and no empty table. Buckets with tasks continue to work as before.

When you open the Add a task modal on a deal detail page, the task type defaults to Call, the due date defaults to today, and the due time defaults to the current time. Save is enabled the moment the modal opens.

The Call default matches by name from your account's task type list. If your account has no task type named Call, the dropdown shows its usual placeholder and you pick from your list. Editing an existing task is unchanged: it shows the task's saved type, date, and time.

The "Welcome to Deals CRM!" modal no longer appears when a pipeline search returns no matches. It's reserved for accounts that have no pipelines at all.

When Active Intelligence sets up your reporting dashboard for the first time — a process that can take up to about four minutes — the header and controls stay visible throughout. A step-by-step progress display walks through what's being built: pulling numbers, creating visualizations, and assembling each widget, with a midpoint marker and a closing message when the widgets are ready.

Skeleton cards below the header show real chart shapes dimmed to signal they aren't live data yet. Reduced-motion settings are respected: charts appear dimmed and static instead of animated.

A custom connector setup form is available in the integrations area, so you can configure a custom connector directly from the product.

The contact record displays a contact's email open-tracking consent state in two places: a header row reading "Email open tracking: Opted out · [date]" (shown only when the contact has opted out) and a read-only field in General details showing the state and date.

The field distinguishes three states — opted out, consented, and never asked. Contacts that have never been asked show an empty field value, hidden by the panel's Show Empty Fields toggle.

When no task is assigned, a placeholder is shown. Restricted deals show neither the stage color nor the next-task line. The account page deals view is unchanged.

The contact import card appears in the right rail of the Inspire me tab, beside the Build engagement ideas cards, rather than on the Your dashboard tab. When the import card would be the only content on Inspire me and account data hasn't loaded yet, the tab waits rather than opening a blank rail. The legacy homepage placement is unchanged.

The Store column on a Contact's ecommerce tab shows the correct store name for every order, including those from connections beyond the first 20.

The Active Intelligence prompt header on the Campaigns page displays a violet, blue, and pink gradient band across the top, matching the visual treatment used in the Automations first-time experience.

Also: clicking a suggestion chip refills the prompt every time, including when you've edited the text it placed and click the same chip again.

If your campaign's audience is a single tag that isn't among the first 100 tags loaded in the dropdown, the field fetches that tag by ID and displays it correctly. The fetched tag is cached immediately, so there's no delay or blank flash.

Search in the tags dropdown also covers every tag in your account — your search text is sent to the server, so no tag is silently out of reach. Tags already shown stay visible while a new search loads.

The dropdown still shows up to 100 tags when you open it without searching.

When a campaign's audience uses a single "Tag exists" custom condition, the Send To field correctly identifies the segment, shows "Edit conditions," and opens the builder pre-filled with your saved conditions. Existing campaigns are fixed without any changes to your data.

Accounts with one or fewer Automations see a guided first-time experience on the Automations index page instead of an empty list. The page shows a prompt box that opens Active Intelligence chat, three personalized suggestion chips, a "Help me build a prompt" button that opens a Prompt Builder, and a tabbed section of three personalized blueprint cards — each showing a Trigger → Action flow diagram with an "Activate now" button.

Chips and blueprints are cached for 24 hours. If either call fails, static fallback suggestions render. The experience disappears once an account's Automation count rises above one.

When you import an Automation that requires setup, the Active Intelligence sidebar opens with a setup card offering "Get Started" and "Skip setup" buttons. The card matches the previous import drawer's behavior: an entitlements warning banner appears for upgrade-entitlement imports, and buttons show a loading indicator while the builder is processing.

If you dismiss the sidebar — by starting a new chat, navigating away, or opening another call-to-action — the builder falls back to the original import drawer, including the "Back to recipes" option. The setup card only appears on the builder page for the specific Automation being imported; opening a new tab or navigating elsewhere clears it.

This prompt appears only for accounts with Active Intelligence access. Accounts without it continue to use the existing import drawer unchanged.

Per-step import cards after clicking "Get Started" still open in the legacy drawer; moving those into the sidebar is planned for a future update.

Previously, steps placed after a follow gate were hidden or never ran because the builder didn't know how to display the gate's paths. Steps saved after a gate on the same path load onto the follower arm, which is where contacts actually travel.

A gate with empty arms is not flagged as incomplete. Like any split, a gate ends its path and can't be inserted mid-flow.

A Recipes page lets you browse a catalog of pre-built Automations, filter it, and install one in a single click. Installing creates the Automation in a paused state, so nothing goes live until you switch it on.

The buttons appear on the latest settled reply, only when a confirmation with both options is present. Clicking either sends your answer through the same path as the existing Automation continue button. The buttons are hidden if a later message arrives, if the reply is still streaming, or if a newer Automation result supersedes the current one.

The Automation Canvas builder loads correctly after you preview an Active Intelligence-generated automation and click "Review and Activate." Some new accounts previously saw a blank builder page after that step.

The Approvals tab stays mounted while a previously-failed suggestions read is retrying. The initial load still hides the tab, and a successful empty read still redirects away from the approvals URL as before.

The button appears in the tab header on the right side, consistent with the layout of other deal tabs. Upload behavior, authentication, and notifications are unchanged.

Active Intelligence can act on IDs that appeared earlier in the same conversation, whether you typed them, a tool returned them, or they appeared in a table, without requiring you to restate them.

When a chat request times out, Active Intelligence displays a warning banner with a Retry option. Two cases are handled: a request that never started, and a reply that began then went quiet. Retry sends the same message again. The banner appears as soon as the stall timer runs out and follows your light or dark mode setting.

The warning banner on the Active Intelligence instructions page — shown when no instructions have been saved — displays its full description text below the title.

Email addresses used as contact IDs: when you pasted a list of email addresses, Active Intelligence was sending each one as a numeric contact ID, which the API refused. It recognizes that an email address is not an ID, looks up the correct numeric ID first, and continues.

Unrecognized parameter names: query parameters the API doesn't accept were refused 526 times across 334 conversations in the seven days to October 2, halting the task each time. Active Intelligence checks parameter names against the API before sending, so invalid names are caught and corrected before the request goes out.

One limit remains: segment UUID values on contact lookups will be refused until the API catalog is updated to reflect that the API accepts them.

Previously, Active Intelligence reported the emails as built even when none were generated. Accuracy on this scenario improved from 0 out of 12 correct responses to 9 out of 12.

When you attach a PDF or other document to a conversation and reference it in a later turn, Active Intelligence can read the file's contents again. Previously, re-reading a file from an earlier turn failed silently — Active Intelligence fell back to a shorter summary of the file, losing detail, or declined with a message that it couldn't open the file. Over 30 days in production, 184 "could not read the stored file" fallbacks occurred from this cause.

Previously, an empty symbol list caused the rename request to fail before it was sent. The rename proceeds with the correct data and leaves all other Brand Kit content unchanged.

When Active Intelligence builds a widget from a single Automation's block events, the reviewer sees the send and skip totals, block type counts, and total event count that the chart actually renders.

The "View all integrations" link goes to the integrations listing page.

When you choose to resend a split Campaign, a picker lets you select which variant to send. It defaults to the winner, or shows no selection if no winner has been decided yet. The resent Campaign goes out as a regular Campaign with one email, not a split test — a note in the picker tells you this before you confirm.

The restored flow is fully localized: all supported languages see translated labels for the variant selection prompt, winner badge, and validation message.

Visibility toggles: The Image, Title, and Author elements on a Showcase block can each be shown or hidden individually using the visibility switchers in the editor panel. At least one element must remain visible.

Text truncation: A "Truncate text" switch in Showcase block settings limits Title to 2 lines and Author to 1 line, with "..." added when text is cut. Truncation is applied when you write text into the block, because email clients do not run scripts. The character limit is based on the tile's width — not a pixel-exact measurement — which is an accepted trade-off given the email's fixed content width.

Visibility toggles apply to newly added Showcase blocks. Blocks already saved in existing templates or Campaigns keep their previous settings.

The /schedule-task entry in the Active Intelligence composer slash menu shows a Tool Tip on hover explaining what Scheduled Tasks do. The tooltip reads: "Set up work that Active Intelligence runs for you on a schedule — a Monday recap, a daily check on bounce rates, a Friday list of contacts who went quiet. Describe it once and say how often; the results come back to you each time it runs."

The Scheduled Tasks chat input displays three example prompts that rotate to show you what kinds of tasks you can create:

• "Every Monday, recap last week's campaign performance."
• "Every month, draft a campaign based on content from my site."
• "What scheduled tasks should I create to help my marketing?"

These appear in the collapsed composer so you can see possibilities before you start typing.

The landing page editor top bar shows two corrected labels.

The "View as" label before the desktop/mobile device switch previously displayed as a raw translation key. It renders correctly.

A "Selection:" label has been added before the context menu that shows Auto, Element, Column, Row, and Section options.

Both labels are localized so each language can punctuate them appropriately.

The post-import performance report includes an industry average benchmark line under the open-rate and click-rate rows, displayed as a formatted percentage — for example, "Industry avg 24%" — so you can see at a glance how your historical sending stacks up against your industry.

Previously, every tool error triggered repeated calls that could never succeed, exhausting the retry budget.

Errors caused by your input or account configuration end the run with a clear result rather than a failure. Errors on read-only or idempotent operations may retry once after a short delay. Errors on write operations that may have partially applied end the run with a result noting the partial state. All other errors end the run immediately.

The filter appears in the list controls row alongside the unread toggle and sort controls. The list updates in real time: if a conversation moves out of the current filter view — for example, a conversation gets assigned while you're viewing Unassigned — it's removed from the list automatically.

The Deals tab on a contact's profile gains a sort control and three new filters — Pipeline, Stage, and Owner — alongside the existing status filter.

Sort options: Newest First, Oldest First, Recently Updated, Least Recently Updated, Value High to Low, Value Low to High, and Alphabetical A–Z / Z–A. The default order is newest-created-first.

Pipeline, Stage, and Owner options are drawn from the contact's own deals, so every option in the list returns at least one result. Stage options group under pipeline headers until you pick a pipeline, then only that pipeline's stages appear. Changing the status filter drops any Pipeline, Stage, or Owner selection that no longer applies.

When filters narrow the list, a "Showing N of M deals" count appears below the toolbar. If no deals match, a distinct empty state appears with a prompt to clear filters. When you add a deal while filters are active, a toast offers "Show all deals" to reset the view.

Clearing filters leaves the status filter alone. Filter and sort choices last for the session only — they are not saved between visits.

One value per filter is supported. Multiselect for Pipeline, Stage, and Owner, and sorting by pipeline, stage, or close date, are not included in this release.

Active Intelligence Deals CRM insights have been substantially improved across forecasting, pipeline summaries, and multi-currency handling.

Forecast improvements:

• The top-3 deal concentration figure is bounded to 100%. Previously, mismatched deal populations could produce figures like "191% of forecast depends on the top three deals."
• Forecast results can be scoped to a date window — "this month" or "Q4," for example — so the card answers the question you actually asked.
• Deal win probability is used per deal when available (0–100%), with stage position as a fallback only. Accounts without win probability configured keep the previous behavior.

Pipeline summary improvements:

• Pipeline health and stalled deals use the same scope by default. Previously, pipeline health defaulted to the first pipeline while stalled deals covered all pipelines, causing stalled counts to appear larger than the total pipeline on multi-pipeline accounts.
• Stage bottleneck detection is per-pipeline: a stage is flagged when it holds more than 40% of its own pipeline's open deals, not 40% of all open deals across the account.
• The 1,000-deal fetch cap and access-exclusion notes appear as a closing footnote rather than as a primary insight, so they no longer dominate the card.

Multi-currency handling:

All monetary totals — pipeline value, forecast, stalled value, owner performance, similar deals — are summed per currency. When deals in a scope use different currencies, the total is shown per currency rather than added together. Single-currency accounts see the same figures as before, now labeled with the currency code.

The hex input label in the color picker displays "Hex" as expected.

The operator labels for subscribed-list conditions in the segment builder now describe what each operator does, making it easier to build the segment you intend.

For accounts without a Brand Kit, Active Intelligence begins building the brand candidate set as soon as a conversation opens, so candidates are ready by the time you finish typing your first message. Accounts that already have a Brand Kit are not affected.

The Automation path view displays a Tool Tip when the Automation is paused. If you're looking at a path that isn't moving and wondering why, the Tool Tip tells you the Automation is paused so you know exactly what to address.

A new metric for anonymous pixel tracking is available in the campaign performance report, giving you an additional data point on how your Campaigns are being engaged with.

Active Intelligence's top-engaged contacts feature loads correctly on accounts with large subscriber lists, where it previously returned an error.

The updated toolbar appears in the Agent home, conversation view, Scheduled Tasks, and Campaigns. While the attach menu is open, the plus sign rotates to an × to show the menu is active. Users who have reduced motion enabled in their system settings see the open state without the rotation animation.

The process can take up to about four minutes. The dashboard header and controls — Create, refresh, and manage — stay visible throughout so you know you're in the right place.

A step-by-step progress display walks through what's being built: pulling numbers, creating visualizations, and assembling each widget, with a midpoint marker and a closing message when everything is ready.

The skeleton cards below the header show real chart shapes — area, bar, line, and pie — dimmed to signal they're not live data yet, with a slow sweep of light across them. If you have reduced-motion settings enabled, the charts appear dimmed and static instead. The quick skeleton shown on normal page loads for returning users is unchanged.

Active Intelligence uses a filtering step when processing results from API calls. When a response wraps its list inside an object — as the brand kits response does — Active Intelligence previously returned an error instead of the records you asked for. Active Intelligence now navigates nested response structures correctly, so requests that previously failed with an error return the results you'd expect.

Active Intelligence handles discount codes and deadlines more precisely when generating Campaign content. When a discount code or deadline is something only you know, Active Intelligence leaves a blank placeholder in the draft for you to fill in rather than inventing a value. The review step that checks proposals has also been strengthened: it correctly distinguishes promotional codes (including digit-leading codes like 20OFF) from units, ordinals, and other number-like strings; reads currency amounts in words and symbols across multiple languages; and recognizes offer patterns like "Buy 2 get 1 free" and "BOGO." Proposals that try to rewrite existing content, or that ask for something already provided earlier in the conversation, are rejected. The preheader suggestion is only offered when the email's preheader is currently empty.

Active Intelligence also applies your brand style to Campaign generation automatically, without requiring a Brand Kit to be saved first. When a brand has been built in the background and meets the confidence threshold, it's applied to the generated email automatically. You can override this at any time by submitting your website or choosing to describe your brand manually.

When Active Intelligence encounters specific situations while editing or importing a Campaign, it surfaces notices in plain language.

• Import notices: descriptions explain which kind of design element could not carry across during an import, and how pasted copy was interpreted.
• Image spot notices: when a picture slot could not be filled, the notice names the slot by its section number and image source.

Four in-chat notices are also shown during Campaign editing: a "no change needed" notice citing the section numbers involved, an "image not found" notice when a required picture could not be located, a "designer-ported layout" notice when a layout was rebuilt because Active Intelligence cannot edit it directly, and a "conditional content unavailable" notice when a first-name greeting was left as written.

The context selector in the Active Intelligence toolbar uses an updated icon and adds Tool Tips to its controls, completing the interactive behavior of the control.

Active Intelligence can retrieve conversion attribution data for Automations, including lifetime converted totals and conversion percentages. A converted total or percentage of 0 is reported as 0; a missing or non-numeric field is treated as not available and not reported as zero.

This capability is behind a feature flag.

Open counts, open rates, and click-to-open rates reflect the raw count that includes Apple Mail Privacy Protection opens. The other improvements from the earlier data-source change remain in place.

Previously, opens and clicks from sends that predated your selected window were included, which caused unique opens to exceed total sends in low-volume Automation emails.

Active Intelligence can report on your SMS Automation activity — account-wide or for a specific Automation — when you ask about it.

You can ask how many SMS Automations you have, request performance data for a specific Automation by name, or see metrics scoped to a single Automation. Metrics include sends, deliveries, replies, failures, opt-outs, and clicks. Automations that exist but have never sent are included in the count, so the total reflects your actual setup, not just active senders.

Reply rate can display above 100% when accumulated replies in the 30-day window exceed the send count in that same window.

When Active Intelligence looks up your top engaged Contacts, it uses the correct data source. If that lookup fails, Active Intelligence retries once via the correct source and then declines the widget with the actual failure reason — rather than substituting a different, slower query that could cause "API experiencing high load" errors and obscure what went wrong.

The button and modal title for adding a contact in the deal detail Overview contacts panel and account detail contacts card read "Add a contact", matching sentence case used across the rest of Deals CRM.

The Clarivate integration tile and its connection settings page display Clarivate's logo. The grey placeholder that appeared previously is replaced in both the integrations list and the single connection edit page.

When you mark a deal task complete and the task type requires an outcome, an outcome step opens — with an outcome picker, a "More info" input, and Cancel / Mark complete buttons — before the task is saved. Tasks whose type has no outcomes still complete in one click.

The page header on the Edit Segment tab maintains its full height when the segment builder below it is tall. Previously the header shrank on the Edit tab compared to the Contacts tab.

The Contacts tab on a segment — available both on the edit segment page and in the Active Intelligence sidebar — displays the total number of matched Contacts above the list.

The count behaves accurately at every stage:

• A loading skeleton appears while data is fetching.
• A ready but empty segment shows 0.
• While the segment is still computing or has errored, the count is hidden entirely, so a false zero is never shown.

Formatting is locale-aware. A refresh button below the count lets you recompute it against the current segment definition without leaving the page. The button is disabled while the refresh is running.

The blue check mark that previously appeared on the /schedule-task row has been removed. Hovering over the option shows a Tool Tip: "Set up work that Active Intelligence runs for you on a schedule — a Monday recap, a daily check on bounce rates, a Friday list of contacts who went quiet. Describe it once and say how often; the results come back to you each time it runs."

The Tool Tip follows your page theme, appearing dark on a dark page and light on a light page.

In the Campaigns list, scheduled Campaigns display the exact send date and time next to the relative countdown — for example, "in 3 hours (10/02/2026 20:00)". The date and time use your account's configured time zone and date/time format. Other Campaign statuses are unchanged.

Click rate, reply rate, and opt-out rate in the SMS Automations report and folder views are now calculated against delivered messages, bringing them in line with how the Broadcasts report has always worked. The numbers you see across all SMS surfaces are consistent.

The surfaces updated include the Automations report individual view, the Automations aggregate table, the SMS index metric cell, and the SMS folder-view broadcast card. Delivery rate and fail rate continue to use sends as their denominator. When deliveries is zero, engagement rates display as 0% rather than a dash.

You can ask Active Intelligence questions about SMS Campaign click counts and click-through rates. Unique clickers are counted as distinct subscribers; total clicks as distinct events. Click-through rate is calculated as unique clickers divided by delivered messages, consistent with the reply rate and opt-out rate denominators used in the Broadcasts report.

When a click-through rate can't be computed reliably, Active Intelligence returns a plain explanation with the raw clicker and click counts rather than a misleading ratio.

A few things to keep in mind: CTR is over deliveries, not sends or opens; SMS clicks aren't scanner-filtered, so they aren't directly comparable to email CTR and can't be segmented by list. Click and CTR metrics for Automations aren't yet available.

Several pages that previously redirected away when you lacked access or the URL was invalid display a clear 403 or 404 message at the same URL, so you keep context about where you were.

• Inbox deep links: Visiting a conversation URL without inbox access renders a 403 at that URL instead of dropping you on the SMS overview and losing the conversation ID.
• Campaign reports: A Campaign report for a missing or off-limits Campaign shows a 403 (permission refused) or 404 (not found) with a link to the Campaign list, rather than silently redirecting to the list and losing the Campaign ID.
• Malformed Campaign IDs: A non-numeric Campaign ID in the URL renders a not-found page in place rather than loading the overview.
• SMS with no access: Users without SMS access see a 403 in place rather than being sent to the overview.
• Conversation, insight, and Scheduled Task deep links: Direct links to conversation, insight, and Scheduled Task detail pages render a 403 at the correct URL rather than dropping the ID and redirecting to the default dashboard.

The two-column layout for the website URL input step during onboarding is the permanent experience for all users. The layout shows the prompt copy "Get marketing assets..." and the field label "Enter your business URL," with the logo displayed inline.

After completing an import review, clicking Continue opens a performance report covering six checks: open rate, click rate, bounce rate, unsubscribe rate, duplicate Contacts, and send-time opportunity. Rows are omitted when values are missing. Open and click rates include an industry benchmark comparison when one is available.

Active Intelligence appears alongside the report, summarizes your account, and offers up to three suggested prompts based on the report. Selecting a prompt or typing a message opens an Active Intelligence chat in a new tab. The Active Intelligence rail can be collapsed and expanded without affecting the onboarding flow.

Two hover animation issues in the Active Intelligence sidebar are resolved. The AI credit usage row animates on hover with bars rising from a shared bottom. The Scheduled Tasks clock no longer draws with a thicker stroke during its hover animation.

The Deals tab on a contact's profile gains a sort control and three new filters — Pipeline, Stage, and Owner — alongside the existing status filter.

Sort options: Newest First, Oldest First, Recently Updated, Least Recently Updated, Value High to Low, Value Low to High, and Alphabetical A–Z / Z–A. The default order is newest-created-first.

Pipeline, Stage, and Owner options are drawn from the contact's own deals, so every option in the list returns at least one result. Stage options group under pipeline headers until you pick a pipeline, then only that pipeline's stages appear. Changing the status filter drops any Pipeline, Stage, or Owner selection that no longer applies.

When filters narrow the list, a "Showing N of M deals" count appears below the toolbar. If no deals match, a distinct empty state appears with a prompt to clear filters. When you add a deal while filters are active, a toast offers "Show all deals" to reset the view.

Clearing filters leaves the status filter alone. Filter and sort choices last for the session only — they are not saved between visits.

One value per filter is supported. Multiselect for Pipeline, Stage, and Owner, and sorting by pipeline, stage, or close date, are not included in this release.

Several controls in Active Intelligence that previously reset on page reload now save and persist.

• Objectives save with Active Intelligence.
• Reasoning effort, auto-react, and its emoji persist across reloads.
• Routing rules persist. Active Intelligence does not apply them yet, and nothing on screen indicates otherwise.
• Turn rating (1–5 stars) is available in the Activity tab on each individual turn.
• Persona generation — generate a persona from your account's own objectives and knowledge, then apply it like a preset.
• Flow drafting — describe a goal in plain words and get a paused flow drafted, with a link to open it in the Flows builder.
• Voice clips — record or upload voice clips (AAC or MP4, up to 10 MB), play and delete them, and switch pre-recorded clips on.

Adaptive reply chips, comment replies, and relay to an Automation reset on reload.

When a Denmark SMS registration is rejected, the registration form displays the specific reason returned by the carrier above the prefilled form, so you can see exactly what needs to be corrected before resubmitting.

If the stored reason has been cleared or the registration was never rejected, no banner appears.

The SMS composer credit estimate prices messages to Denmark registrations correctly. MMS is not supported for Danish numbers, so MMS messages to Denmark show a cost of 0 and the tooltip estimate is suppressed, rather than silently falling back to the SMS rate.

Previously, following a deep link to a conversation, insight, or Scheduled Task detail page without the right access would redirect you away and lose the ID from the URL entirely. Those pages render a 403 at the correct URL, so you keep context about where you were and the link stays shareable.

Clicking the Forecasted close column header in the Deals CRM spreadsheet view sorts the list. Previously, clicks on that column were silently ignored. Saved sorts and deep links using that column also apply correctly.

Deal cards in the Deals CRM pipeline view show the correct contact avatar again. When a card displays the Contact or Contact-and-Account field, the contact's avatar leads that row — so you can identify the person at a glance without opening the deal. The avatar appears automatically whenever you've selected the Contact or Contact-and-Account field; no separate toggle is needed.

The overview dashboard tasks widget returns correctly after visiting a deal detail page in the same tab. The deal task modal's Task type dropdown also populates as expected after visiting the overview dashboard. Both issues are resolved.

An "Assigned to" dropdown in the SMS inbox conversation header lets you assign a conversation to yourself, a teammate, or clear the assignment. The dropdown loads all account users. The current assignee is shown in the trigger. If the assignment write fails, the previous assignment is restored and an error is shown.

After Active Intelligence generates a Campaign, up to three additive suggestions appear above the composer. Each one adds something the campaign is missing — a preheader, a personalization tag, a secondary call-to-action, a discount code, or a deadline — grounded in your brief or the email's own content. Suggestions never assess, shorten, or remove anything already there.

Active Intelligence classifies your campaign's purpose (promotion, newsletter, event, announcement) and only offers additions that fit: a webinar brief never receives a discount-code suggestion; a promotion never receives a summary. A check runs before any suggestion reaches you, dropping candidates that use judgmental language, invent figures, or lack a quoted anchor from your brief or email. If nothing is provably missing, the row stays empty.

Clicking a suggestion sends it as a real edit turn — the composer is left exactly as you had it. Suggestions that need your input (such as a discount code or deadline you haven't provided) open in the composer with the blank pre-selected so your first keystroke fills it in. Accepted suggestions recompute the row for the new version of the campaign.

Suggestion chips are written in English or Spanish only; campaigns in other languages receive English chips.

When you ask Active Intelligence to create a field whose name matches an existing field with different options or a different type, it no longer changes the existing field automatically. Instead, it tells you about the conflict and asks whether you want to create a new field under a different name or edit the existing one.

This matters because custom fields are shared across all Forms that use them — a silent change would have updated the field everywhere it appears. Active Intelligence asks first, so you stay in control of which fields change and which don't.

This acts as a safety net: when a rate is incoherent, Active Intelligence produces no card and no generation. A rate of exactly 100% is treated as valid.

After Active Intelligence generates a Campaign, a row of up to three additive suggestions appears above the composer. Each suggestion shows what it would add and why, drawn from your own brief or links already in the email. Clicking a suggestion sends it as a real conversation turn, triggering an edit.

The suggestion engine only proposes additions — a preheader, a discount code you asked for, a deadline line — and never assesses or removes content. It classifies the Campaign's purpose (promotion, newsletter, event, announcement) and only surfaces candidates that fit: a webinar brief never gets a discount-code suggestion; a promotion never gets a summary. When the engine finds nothing to add, it returns a neutral line rather than a verdict.

A suggestion whose material only you can provide — such as a discount code or cut-off date — opens in the composer with the blank selected, so your first keystroke replaces it.

The connection screen for importing contacts from a previous contact database has an updated headline — "Connect your account to get started" — and supporting copy that describes the migrate-and-start flow. The read-only notice, Connect, and Skip buttons are unchanged.

The date follows your account's configured time zone, locale, and datetime format — the same format used in the activity stream. If a tag's name is too long to fit in its chip, the Tool Tip shows the full name above the date. Tags with no recorded date show no Tool Tip.

The sort dropdown on the Tags tab is also translated across all supported languages. This is part of the Contact Detail Page Refresh, which promotes tags from a single row to a full tab with alphabetical sort and removable chip display.

The segment builder reads each account's actual condition limits from account configuration. Accounts with raised limits can build and save segments up to their own higher caps; accounts without overrides behave as before.

The condition cap is enforced inside the builder itself, and error messages show the actual limit number for your account. Existing segments that already exceed a limit continue to work.

When setting up a repeating Scheduled Task, two improvements make the end date easier to work with.

If you pick an end date that has already passed, a warning appears immediately — blocking save — with the message "The selected end date has already passed, therefore the schedule will never run." Previously, no error appeared until you attempted to save.

The schedule summary sentence includes the end date when one is set, reading as "Starts on [DATE] and repeats every [frequency] and ends on [DATE]." Previously, the summary gave no indication an end date had been configured.

These improvements apply to sub-daily cadences. Daily-or-longer cadences don't show the end date field.

A "Schedule report" option appears alongside the Analyze button on eligible dashboard widgets. Clicking it opens Active Intelligence in a Scheduled Tasks context and submits a weekly report prompt. On each run, Active Intelligence re-queries the widget's original request against fresh data and returns changes, drivers, and recommended actions.

This feature requires existing Scheduled Tasks access.

The platform loads the correct, current build in Safari. Blank pages and errors in these areas on Safari are resolved.

The inline segment summary on an If/Else block in the Automation Canvas shows SMS list names, so you can read your branch conditions at a glance without opening the block.

You can export Reports data to a CSV file directly from the Reports section.

The All Automations report also has a CSV export button. Clicking it starts a background export with your current filters applied and shows a success notification with a link to the import/export status page when the file is ready. The export includes Name, Status, Contacts, and Campaigns columns.

In read-only email previews — including version history, Active Intelligence Campaign draft cards, and translation review — personalization tags display as violet chips directly inside the preview content. They scroll natively with the email, scale with the surrounding text size, and show the tag's display name (for example, "First Name") rather than the raw tag syntax. Your exported HTML is unchanged.

Several pieces of text that were showing a stray "}" character instead of the intended value are fixed across 16 languages.

• The personalization chip tooltip in AI email preview reads "First name is filled in for each contact when the email is sent."
• The Automations list shows "3 labels added to an automation" / "3 labels removed from an automation" correctly.

HTML you author in a message variable's Code tab is preserved exactly when you switch to Visual mode and back. Previously, clicking inside the Visual editor after switching from Code would rewrite your markup — flattening structure to plain paragraph text and inserting extra line breaks — so switching back to Code returned a different document than you wrote.

The editor distinguishes a click (selecting a position) from an actual edit, so clicking in Visual mode no longer overwrites your Code tab content.

The Manage columns menu stays open while you toggle columns on or off, so you can make all your changes in one pass without reopening the menu each time.

The previous wording told you to "submit the prompt in the new tab" as though the prompt would already be there — but at least one major AI service no longer supports URL prefill, leaving an empty text box.

The new instructions read: "We'll copy a prompt and open your selected AI service. Paste the prompt, then copy the response back here." Prefill remains available as a convenience for services that still support it.

When a file upload is rejected because too many uploads are already in progress, a toast message appears with the file name and a prompt to try again in a moment. Previously the file disappeared silently with no explanation.

This applies to the standard file attachment component.

The Engage Settings area has been rebuilt to match the platform's design system. The hub uses a breadcrumb-plus-title header and a card-grid layout. Sub-routes show a label-and-description beside a card of rows.

Specific improvements include: the ice breakers drawer with autosave and prompt management; channel connection health banners in warning styling; reconnect reason banners in the channel drawer; Channels tab counts showing connections that need attention; and messaging rules with a disabled state. Empty and error states are present throughout where none existed before.

Settings copy that was previously missing — causing raw text keys to display — has been filled in across the ice breakers drawer, the Connect page, reconnect reason banners, the invite a teammate link, test delivery failure messages, and the delete workspace note.

SMS is visible in Settings but marked Coming soon. The delete workspace option is also shown as Coming soon.

Active Intelligence Deals insight cards consistently include the stalled-deal section. Previously, stalled-deal requests were answered with forecast data or pipeline health data in roughly 74% of cases — accounting for approximately 26% of Deals quality rejections categorized as "stalled deals section missing." Stalled-deal requests route to the correct tool even when surrounding context mentions pipeline health or forecasts.

Cancel and Send buttons in the one-to-one email composer drawer are right-aligned, with Send as the rightmost action. Save and Cancel buttons under the Notes tab and inline field editors in the Overview tab are also right-aligned. Button spacing in deal detail modals — including Delete deal, note confirmation, Add/edit task, and Add contact — is consistent with design standards.

The Add a task modal's form fields render completely — dropdown focus outlines and open-menu shadows are no longer cut off on the left. The "Hide Empty Fields", "Show Empty Fields", and "Manage Fields" buttons on the Overview tab use sentence case, matching the rest of the Deals CRM detail page.

The segment edit page includes a Contacts tab, matching the tab already available in the agent drawer. The tab shows the Contacts that match the segment's current criteria. It only loads when you open it, and any edits you've made to the segment are preserved when you switch to the tab and back.

The performance report shown after importing contacts has been reworded throughout. Headings, section titles, and row labels frame your imported data as a diagnosis rather than presenting a neutral dashboard. Metric definitions and values — including contact count, campaign count, and industry comparisons — are unchanged.

The tag dropdown in the Conditional Content modal searches your full account tag list server-side. A previously selected tag stays visible even when it falls outside the current search page.

The tag dropdown in the Conditional Content modal searches your full account tag list, showing up to 100 results at a time. A previously selected tag stays visible even when it falls outside the current search results.

Previously, the dropdown tried to load every page of tags to build a complete list. For accounts with thousands of tags this was unusably slow.