Changelog

​

September 2026

​

Images: GPT Image 2.5 by default, the brand image profile, and article images in one step

• As of 2026-09-12, GPT Image 2.5 draws the pictures: every image surface (the Media Library’s generate action, the Design Studio chat, the Generate Image and Image Set workflow nodes, the Visual Editor, article images, the agents, the Claude Code plugin and the API) now defaults to OpenAI’s gpt-image-2.5-flare, released 2026-09-08: better than the previous default at about half the latency, with stronger preservation of a referenced subject on edits. A failure on that lane renders the same request on Gemini 3.1 and the result says which model drew it; Gemini 3.1 and GPT 5.4 stay selectable, gpt-image-2.5-sunburst is the premium pick for an edit that must hold the subject exactly, and the six GPT Image quality settings (low, medium, high, xhigh, max, auto) are accepted where a quality is asked for. Each image is charged to the credit ledger at what it actually cost
• The brand image profile: a branded generation used to be told the brand name, three colors, two fonts and 200 characters of voice. It is now handed the whole imagery side of the brand guide as plain sentences (personality and adjectives, the full palette with hex codes and gradients, typography for any text in the image, photography, illustration and icon style and rules, logo rules, the AI prompt rules and an avoid list) plus the primary logo and up to three Media Library images tagged brand_reference as reference input, with the prompt told which attached picture is which. Results carry brand_reference_ids and name any reference that could not be fetched; GET /api/olympus/marketing/brand/image-profile (brand_image_profile_get) returns the profile exactly as a model receives it. The flux, flux-pro and recraft models take the written profile only
• Mark approved brand imagery with one tag: add brand_reference to a Media Library image under Marketing > Media (or with tags on the media API) and it rides along with the logo on every branded generation from then on. The three most recently tagged images are used; tag a newer one to promote it
• Article images: a content item gets a hero and one image under each major section where a picture adds information, from one step that reads the stored body, the item’s grounding and the brand image profile, renders the hero first and the sections with the hero as a reference so the set matches, registers each in the Media Library (tags content_image and content:<item id>), places each under its heading and sets the featured image with its alt text. Available as Generate images in the marketing content editor (it opens with what the run would do and how many generations are left), as GET and POST /api/olympus/marketing/content/{id}/images and their dashboard twins, as the content_images_generate tool for the department, the Claude Code plugin and the VS Code extension, and as a Generate images switch (off by default) with a Max images cap on the Create Marketing Content workflow node, which the shipped Weekly Trending Topic Blog Draft template turns on so the Monday draft arrives with its pictures and the count in its email. Each image draws from the monthly image allowance; a run refuses with 402 quota_exhausted before planning when nothing is left, stops with a partial result and a warning when the allowance runs out part-way, and refunds a render that fails. Alt text is written for every image and checked by the pre-publish check. The plugin’s content play now illustrates every long-form piece this way after the draft and before the quality gate, with the allowance read first. See Media Library and AI Images, Images for the piece and Article Images

​

Content: grounded drafts, a pre-publish check, real internal links, and a calendar that shows what will publish

• As of 2026-09-11, Foundation first: the Content department reads the account’s customer avatars, journey maps, before/after grids, brand guide and existing content before any outline or draft, and every draft opens with one header line: For: avatar | Stage: journey stage | Promise: before/after item | Keyword: target keyword | Links: n internal links. No usable avatar means no draft; the department names the missing fields and points at the Customer Avatars page instead of writing generic copy. The brand file it reads now carries ai_forbidden_phrases, copy_dos, copy_donts and the reading level, a forbidden phrase is a hard fail it rewrites rather than saves, and long-form copy runs through a four-rule rubric (sourced claims, no filler openers, one job per section, a call to action matched to the stage). The target keyword travels with the item in its target_keyword column (mirrored to settings.target_keyword for older readers), which the editor’s new Target keyword field, the pre-publish check and rank tracking all read
• As of 2026-09-11, Who this is for, recorded on the item: a content item now carries who it is written for as five typed fields, avatar_id, journey_id, journey_stage, before_after_grid_id and target_keyword, and every reader returns them with customer_avatar, customer_journey and before_after_grid as { id, name }. The marketing editor’s sidebar gains a Who this is for panel (avatar, journey and stage, before/after grid, target keyword, with a one-line summary and Clear), the New dialog offers the avatar and stage so a row starts with its audience, and the AI assistant on the item reads the five as its context instead of asking for the audience in every chat. content_create, content_update, the Olympus and dashboard content routes and the Create Marketing Content workflow node accept the same five; an id from another account is a 400 invalid_reference that writes nothing, and the Olympus list filters on avatar_id, journey_id, before_after_grid_id and journey_stage so the avatar-by-stage coverage matrix is one call per cell. The Content department records the avatar, stage, grid and keyword it drafted for on the row and reads its header line back from them; the workflow node warns on the run log when a draft is written for nobody while the account has avatars; and the Claude Code plugin and VS Code extension write the five with the draft and treat an echo without them as a write that did not land. See Who this is for
• A pre-publish SEO check on every surface: one deterministic check runs on the marketing editor (a Pre-publish check section above Status with a score, errors under Fix before publishing and warnings under Worth a look, each row jumping to its field), on GET /api/olympus/marketing/content/:id/seo-check and its dashboard twin, on the content_seo_check MCP tool, and inside content_publish_to_site, which returns the findings as warnings[] plus a structured seo_check. It checks meta title and description lengths, the placeholder slug, an empty or thin body, multiple H1s, keyword placement in the title, slug, H1 and first 100 words, heading order, alt text on the hero and every inline image, internal-link count, figures with no source link, and the brand guide’s forbidden phrases. It never blocks: errors ask Publish anyway? in the editor and come back as warnings on the API
• Internal links from real URLs: content_list and content_get carry url (and url_source) for every published, site-linked item, the department’s data file carries the same, and GET /api/olympus/marketing/content/site-links?project_id= (content_site_links) lists every published post and page on a project with its live URL and explains every omission in notes. The department links at least two existing pieces by URL from that data and says Links: 0 or Links: 1 when fewer exist, rather than inventing a path
• Scheduling that executes: a scheduled content item now lands on the CMS scheduler that runs, whether scheduled from the editor’s new Schedule dialog (with the deploy targets the scheduler executes), PUT /api/marketing/content/{id}, the Olympus schedule routes or the content_schedule tool. An item not linked to a site collection cannot be scheduled and answers 409 not_linked with the link step named; a linked item with no site entry gets one created as a draft so nothing goes live early; and when the schedule runs, the item is stamped published and its mirror date cleared. On the dashboard save, only a changed future time is a schedule request, null never cancels, and cancel_scheduled_publish: true does
• The content calendar: Marketing > Content > Calendar is a month grid and a list of what is scheduled, published, failed or merely recorded (a date on an item that nothing executes, said plainly), with the account’s cadence in the header: published in the last 30 days, scheduled, failed publishes needing attention, and the target cadence read from the department’s memory when one is written there. Fed by GET /api/marketing/content/calendar?from=&to=
• Resume a department turn: talk_to_department waits about 110 seconds while a department turn may run for 1200; a timeout now returns session_id, turn_id and whatever had arrived, and GET /api/olympus/marketing/ai/turns/:turnId (department_turn_get) polls the turn under the same API key, with since_seq paging, a stale flag for a dead worker, and the finished shape (response, tool_calls, data_updates) once it completes
• Fixed on the way: the Trending Topics to Blog workflow node saved empty article bodies since April because it posted body where the route read content; both names are accepted and the node fails when the saved article is empty. Site posts born from the New Content modal published at the untitled-content slug; the slug is now derived from the first real title. Unpublish in the editor set the status back to published; it sticks. The sidebar said Live on your site before any deploy; it is keyed on the recorded deploy outcome. content_create and content_update declare every parameter the routes read, so featured_image_alt, meta_keywords, canonical_url, settings and the CMS link no longer fall off between an agent and the row. See The Content Department, Pre-Publish SEO Check and Scheduling and the Content Calendar

​

Webflow: the content-source switch, and every write pair asks on both halves

• As of 2026-09-10, 111 actions: project_cms_provider_set joins the Webflow registry as its one Hiveku-side action. It switches a project’s content source between the Hiveku CMS and the Webflow CMS with the same bound-site check as the Settings page, so a chat or an MCP client can finish the step that used to need the dashboard. The workflow palette gains the matching node under a new Project family (95 generated nodes beside the sixteen featured ones) and the MCP server its 111th webflow_* tool
• The content source is not switched for you: a Webflow-hosted site keeps the Hiveku CMS as its content source until you switch it. The Content department and the AI now say so and will not publish a content item to the site until the source is Webflow CMS; they offer the switch, which asks before it runs. Page SEO, schema markup, assets, custom code and publishing work either way. The CMS capabilities read now carries external_platform and project_type beside provider, because provider names the content source and not the host
• 32 confirmation gates, direction irrelevant: the gate rule now covers a write that cannot be undone through the API, one that changes what a visitor sees or what is collected about them, and one that starts or stops delivery of customer data to a third party, and where the OFF half of a pair asks, the ON half asks too. Thirteen writes joined the nineteen already gated (the deletes, order fulfil and refund, robots.txt replace, webhook update and the site publish): item publish and unpublish, the site and page script lists, the Google tag, the Hiveku snippet, llms.txt, robots.txt merge, redirect create, well-known create, webhook create and the content-source switch. Nothing was ungated. The workflow nodes carry the checkbox, the MCP tools the confirm argument, and the Claude Code plugin asks before each one
• One notice for page wording: the single-language notice reads the same on every surface: “These words are edited in the Webflow Designer. Page titles, slugs, SEO fields and CMS content stay editable in Hiveku. Webflow only opens page wording to other tools on a site that publishes in a second language.” See The Webflow AI Chat, Webflow as the Content Source and Webflow for Developers

​

Webflow: keep the site in Webflow, run it from Hiveku

• Connect once: authorize a Webflow Data Client app over OAuth and pick which sites to share on Webflow’s own consent screen, or paste a site API token. The connection is an account default that every project inherits, with a per-project override for agencies running several Webflow accounts in one workspace. Tokens are encrypted at rest and never echoed back, and a pasted token is validated against Webflow before anything is stored
• A site token is the lesser option, and we say which parts it loses: it covers CMS, assets, forms, comments, ecommerce, page metadata and publishing, but Webflow refuses it custom code and webhook registration. No webhooks means no inbound events and no form routing, so fifteen actions answer 412 oauth_required on a site token rather than failing upstream as an unexplained 403
• The Webflow workspace: a full screen at /<accountId>/dashboard/<projectId>/webflow with 27 routes across sixteen rail areas — overview, pages and a page editor, CMS collections, fields and items, assets, SEO, forms, comments, components, ecommerce, custom code, redirects and robots, webhooks, departments, activity, publish and settings. No area is hidden by what your plan or credential can do; every page explains its own limit instead
• A chat that drives the site: a collapsible AI pane on the left of the workspace, locked to Webflow mode because the site has no Hiveku code to write. Ask it to draft posts into a collection, rewrite every page’s SEO title and description, add a Google tag or publish, and it calls the same operations the panels call, behind the same gates
• 111 actions, one registry, three surfaces: the workspace panels, 111 workflow action nodes and the webflow_* MCP tools all run the same registry over the Webflow Data API v2, nine of them served by Webflow’s beta namespace. A person, an automation and the AI therefore pass the same permission, confirmation and locale gates and land in the same audit log
• Webflow as the content source: switch a project’s content source and New Content, the Content department, the AI and the CMS panel write Webflow collections instead of Hiveku’s own. Saving an entry stages it in Webflow; Publish makes it live. The SEO department’s page-SEO apply reports applied_via: webflow with publish_required: true, so a report never claims a change is live before the site is published
• Forms, comments and events flow back in: an inbound receiver verifies each delivery — Webflow’s HMAC signature plus a per-connection URL secret on an OAuth connection, the URL secret alone on a site token — records it once per site and event, routes form submissions into the Hiveku Forms ledger, the CRM contact and notifications, and feeds 13 workflow trigger nodes covering site publishes, CMS item lifecycle, page creation and deletion, page metadata changes, new comments and form submissions
• Analytics without a Designer trip: install the Hiveku analytics snippet on a connected site from the Custom code panel, on an OAuth connection. It is registered as an inline script applied site-wide and starts counting after the next site publish; a site-token connection gets the tag to paste into Webflow instead
• Kept honest in the background: four crons watch the connection — token health, webhook health, webhook URL reconciliation and event retention — and a connection made before webhook registration existed can be brought up to date from Settings without reconnecting
• The limit, stated plainly: Webflow’s Data API cannot write page or component body text in a site’s primary locale, cannot create, delete or rename pages, and cannot touch layout, classes, styles, interactions or components. Those stay in the Webflow Designer, and the workspace says so on the page rather than hiding the control. What is editable from Hiveku: everything bound to a CMS collection in either locale, page titles and slugs, SEO and Open Graph, JSON-LD, assets and alt text, custom code and scripts, form submissions, comments, ecommerce, and publishing. Page titles and SEO are editable in the primary locale; adding a secondary locale unlocks page and component body text in that locale only
• Enterprise-plan sites additionally get 301 redirects, robots.txt, llms.txt, files under /.well-known/ and the Webflow site activity log; on other plans those actions decline with the plan named
• See Webflow Overview, Connecting Webflow, The Webflow Workspace, Webflow as the Content Source, The Webflow AI Chat, Webflow Automations and Webflow for Developers

​

Creative Studio: brand-aware images, a media toolkit, and fonts that survive export

• Brand by default: every AI image generation now applies the account’s active brand guide unless you opt out, in chat, with the Apply Brand Context setting on the Generate Image, Image Set and Video workflow nodes, or with use_brand: false on the API. The result records whether the brand actually reached the model; with no active guide it says the image is unbranded, and a brand guide that cannot be read refuses the generation before anything is spent instead of billing an unbranded render
• Image allowance read: before a batch, an assistant can read what the account has used, its limit, what remains and when the period resets, from the same counter the next generation is gated on. Generations and upscales draw from it; imports, transforms and design exports do not
• Import, transform, upscale: import an outside image by URL as a real copy in your storage (25 MB, images only); crop, resize or re-encode an existing asset for free; AI-upscale up to 4x within a 32 megapixel cap for one image generation plus $0.03 per output megapixel. Every operation creates a new asset linked to the original; an asset’s bytes, dimensions and generation metadata can no longer be rewritten, only its title, alt text, caption, description, tags and folder
• Custom brand fonts render in exports: PNG and MP4 exports, save-to-library and the final storyboard render now carry the brand kit’s custom fonts. A font that fails to load falls back and the export names it in a warning, never a failed render
• The inbox hears the studio: a comment pinned on a design files an item in the Ops Inbox (one per thread), and a storyboard video reaching completed or failed files one item per run, each with a link to the design
• Creative to a scheduled post: an assistant working through the Claude Code plugin or the API can attach a finished design export or library asset to an existing or scheduled post, replacing its media list wholesale with a type per file; publishing and published posts stay locked
• Six wide-format templates: two 1200 x 600 email headers, a 1200 x 630 Open Graph card, a 1280 x 720 YouTube thumbnail, a 1584 x 396 LinkedIn banner and a 1500 x 500 X header, brand-substituted and grouped under their sizes in the Templates library
• Concurrent edits cannot clobber each other: the canvas carries a version on every save, whether you, a teammate or the AI designer wrote it. A save built on an older version is refused; the editor asks whether to keep your version or load theirs, and the AI designer re-reads and re-applies your request instead of overwriting. Clip requests now report the length that was rendered next to the length that was asked for, since several video models snap to fixed steps. See Media Library and AI Images, Brand fonts in exports and Sized templates

​

File uploads on website forms

• Nothing to wire: add <input type="file" name="resume"> to any form on a deployed site and the file uploads to Hiveku when the visitor picks it, with a status line under the input and a submit button that waits for the upload. No upload endpoint, no storage to provision, no per-project database
• Files travel with the lead: download links in Analytics → Forms, real attachments on the notification email (up to 5 files or 25 MB per email), an Attachments section on the CRM contact plus document links on the form activity card, and {{trigger.attachments}} in workflows with an Attach form files toggle on the Send Email node
• Limits and controls: 25 MB per file by default, adjustable per project from 1 MB to 100 MB (files over 10 MB upload in parts automatically), 5 files per submission by default, documents and images only, with the file type verified against its contents before anything is stored. Per project you can switch uploads off, narrow the accepted types, raise or lower the size limit and set retention (one year by default, 30 days to one year) under Forms → Notifications → Uploads. Opt a form or input out with data-hiveku-uploads="off"
• Malware scanning: every uploaded file is scanned after it lands. A file becomes downloadable once its scan is clean, usually within minutes; the notification email waits up to 10 minutes for the verdict before going out without the file; a file the scanner could not read is delivered with a “Could not be scanned” note, and a flagged file is removed and shown as “Removed: malware detected”
• Open attachments from the iOS app: tapping a form attachment on a contact’s timeline opens it through a short-lived link, and the contact screen lists every file the contact sent
• Agents and tools: the coding agent, the Claude Code plugin and the VS Code extension know the rule (a file field is a named file input, nothing else) and gain marketing_form_attachments_list, marketing_form_attachment_download_url and the form upload settings tools, on the coding agent, the marketing agent, the Claude Code plugin and the VS Code extension alike. See File uploads

​

Working branches in the editor

• Branch selector: pick the branch you are working on from the top left of the editor. The file tree, live preview, AI chat and Database panel follow it, and edits from a chat turn land on that branch, never on main
• Uncommitted and committed: saves land on the branch immediately; Commit changes records a named point. Deploys, pull request merges, branch creation and branch deletion commit your unsaved changes first so nothing is lost
• Environments: Development and Staging can serve a branch; Production always serves main. The Deploy button knows which branch each tier ships, and the Environments strip on the Branches tab shows what each tier is serving and since when
• Pull requests: open, review with a per-file diff, and merge into main. Merges are strict: any conflict means nothing is merged, and the conflicting files are listed so you can resolve them on the branch
• Branch previews start automatically when you switch and stop after an hour idle; a Stop preview action lives in the branch menu
• One database per project: the Database panel says so plainly on a branch. Content and visual editing stay on main for now
• Claude Code and VS Code: the file tools take a branch parameter, project_vcs_commit with no files commits a branch’s pending changes, project_vcs_diff_file returns one file’s diff between two branches, and project_vcs_env_bindings tells deploy_site what ships. The VS Code extension gains pull request file review and reopen, binding-aware deploy labels, a branch item in the status bar, and a command to stop a branch preview. See Branches, environments and pull requests

​

August 2026

​

Accounts can no longer be deleted out from under the people who live in them

• Home accounts explained: every login is anchored to one account, set when the profile is first created. Being invited elsewhere adds access without moving it, so a personal workspace you signed up with can still be your home account long after your real work moved on. See Your Home Account
• Deleting an account is now blocked when it is the home account of someone who also belongs to other accounts. Previously that deletion removed the person entirely and took their access to every other account with them, and removing them from the team first did not prevent it
• Profiles can be moved to a different account on request, with one condition: the person must already be a member of the account they are moving to. Their memberships are unchanged by the move. Email support@hiveku.com with the person’s email address and the destination account
• Deleting or offboarding an account now revokes its MCP credentials immediately, so editor and CLI access stops at offboarding rather than lingering until the final purge

​

PPC depth tools on every platform

• Google Ads extended lane: geo, radius, and language targeting (including the presence setting that stops local budgets leaking to out-of-area searchers), conversion-action create/update with the tag snippets to install, image and YouTube asset uploads, rule-based remarketing lists, shared negative keyword lists, and full Performance Max asset-group support — one atomic create that refuses to build a group that can never serve, per-group servability diagnostics, and a performance read with Google’s per-channel split of where PMax delivery actually went
• Microsoft Advertising family: campaign build-out, the full targeting-criterion surface (locations, radius, location intent, dayparting, device, demographics), audience association, UET tags and conversion goals with a plain-language readiness verdict, sitelink and callout extensions, shared negative lists, and three account reports — search terms with a wasted-spend summary, quality score with the weak components named, and impression share split into budget-limited vs rank-limited headroom
• Meta and LinkedIn depth: Meta breakdown insights, creative and audience management, lead retrieval, Advantage+ support, and confirm-gated audience edits; LinkedIn’s demographics report (performance by job title, seniority, industry), campaign groups, creative lifecycle, conversion rules and API events, matched audiences, and ABM company segments
• Disapproval reads on all four platforms return the same shape, with policy reasons where the platform provides them — a disapproved ad spends nothing and silently starves its campaign, so these are the first reads to run when traffic drops
• Budget guardrail per ad connection: set a monthly spend target and a daily sweep watches it — an alert files to the agent inbox at a configurable percent of target (default 85), and an opt-in hard line auto-pauses live campaigns
• Creations land inert by design — paused on Google, Meta, and Microsoft, drafts on LinkedIn — and anything that spends, changes live delivery, or is irreversible previews first and executes only on an explicit confirm. See PPC Depth Tools

​

SEO delivery: live rank tracking, page intelligence, and the Tag Manager pipeline

• Live rank tracker: track a keyword and the first check runs within minutes — Google and Bing positions, local-pack tracking by business name and city, and AI-engine citation tracking across ChatGPT, Perplexity, AI Overviews, Claude, and Gemini, with per-keyword history views
• Page intelligence: Core Web Vitals with real-user field data and a Lighthouse lab run in one call (works on competitors too), a Knowledge Graph entity check, a five-section conversion audit of any landing page with named fixes, and a citation audit that diffs the business’s public listings against its canonical name, address, and phone
• Analysis suite with honest data-source notes: content decay, keyword cannibalization, internal-link graph, E-E-A-T scores, and 30-day ranking forecasts — each names exactly where its data comes from and when it’s empty, instead of implying analysis that didn’t happen
• Search Console, full scope: the canonical multi-dimension analytics query, time series, period comparison with winners and losers, URL inspection, an index-coverage substitute, and sitemap submit/delete — connections authorized read-only before late August need a one-time reconnect for sitemap writes
• Google Tag Manager pipeline: create a tag as a workspace draft, snapshot it into a version, and publish — with the publish step confirm-gated because it changes the client’s live site. Closes the classic failure of conversion tags that sat unpublished recording nothing. See SEO Delivery Tools

​

Client reports and the two-key approval rail

• Client report lifecycle: create a scheduled, shareable marketing or social report, regenerate its stored numbers on demand, fetch the public share link, email it to the client behind a recipient-list preview and confirm, or download the marketing report as a PDF of the exact stored numbers
• Report sections cover leads, deals won, traffic, spend, SEO, rankings movement, search-term wins, listings, AEO, local, and a work log showing the client what was done
• Implement-with-AI rail: an SEO finding filed as a task can be handed to the coder agent headlessly — it edits the website project, typechecks, test-builds, and stages the production deploy for approval. Two keys stand between a finding and the live site: one confirm to dispatch the work, a separate approval to ship it
• Approval rail and agent inbox: list, inspect, approve (confirm-gated, single-use tokens), or reject staged agent actions; the agent inbox is the staged alert queue where guardrail findings and health sweeps land, with resolve-or-dismiss lifecycle. See Client Reporting & Approvals

​

Standing marketing playbooks

• Thirteen ready-made workflow templates for recurring hygiene: weekly search-terms negatives mining on Google and Microsoft, monthly impression-share review, lost-backlink alerts, monthly tech-audit regression diffs, content-decay refresh briefs, monthly AEO visibility, a weekly Core Web Vitals watch, daily review-SLA escalation, weekly Business Profile post drafts, and event-driven responses to ad disapprovals, rank drops, and new negative reviews
• Every template is staged-to-approval: negatives and review replies land in the ops inbox as drafts, posts land in the composer approval queue, and read-only loops just alert — nothing is applied, posted, or published automatically
• Three named delivery playbooks — the weekly PPC optimization pass, the monthly SEO retainer, and the monthly AEO pass — give the agent the ordered tool sequence, expectations, and gotchas for each recurring deliverable. See Standing Playbooks

​

Call tracking: keyword-level attribution and every-platform conversion send-back

• Keyword and ad-group attribution: calls from Google Ads clicks now resolve to the exact campaign, ad group, keyword, and ad automatically — shown on the CRM contact timeline, in new by-keyword and by-ad-group breakdowns on Analytics → Calls and the Attribution dashboard, and in a Calls-by-keyword table on each PPC campaign page
• Conversion uploads to Microsoft Advertising and Meta join Google Ads and GA4: one toggle sends qualifying calls to every connected ad platform, with per-call upload status chips on the call drawer so a disconnected account can never silently stop your conversions. Default policy is now Answered calls only
• Exact PPC call counts: campaign call numbers join on the platform’s own campaign ID, so renames and UTM changes no longer zero them out
• Smarter number pools: numbers recycle after a two-day post-call hold instead of a month, repeat callers within 90 days still attribute to their original campaign (with a repeat-caller badge), and search-engine crawlers no longer consume tracking numbers or see swapped ones
• Automatic monitoring: admins are emailed when a pool runs out of numbers or when a site’s tracking script goes missing — each email names the problem and the fix
• Honest reporting: unattributable calls show as Unmatched instead of inflating Direct traffic, truncated views say so, and the calls CSV export carries the full attribution columns
• Cost-per-lead report: a new Reports tab computing spend against credited leads under five attribution models (first-touch, last-touch, 50/50 split, W-shaped, qualified), down to ad-group and keyword depth, with tracked calls counted as touches alongside forms
• Conversation keywords: define the terms that matter and every call transcript is scanned automatically - chips on the call log, trend charts on Analytics, and workflow triggers that fire when a call mentions them
• Local numbers for every visitor: pools now show visitors a tracking number with an area code local to them, and the Pools page shows each pool’s area-code spread
• Route by caller location: numbers can send callers to different destinations by their area code, and ring-group round robin now takes true turns
• Form and deal conversions upload to Microsoft Advertising and Meta alongside Google Ads
• See Phone Tracking and the new Migrate from CallRail guide

​

Client portal and payment pages overhaul

• Payment pages: shareable “pay us any amount” checkout at /pay-to/your-slug. Fixed, preset, or open amounts, reorderable payer fields, optional tax with tax-exemption claims, a customizable thank-you page or redirect, a branded payer confirmation email with test-send, CRM automations (auto-link the payer’s company, optionally create a deal), an outbound webhook with a test button, and per-page views/conversion/revenue analytics. Every payment lands as a normal invoice and payment record. See Payment Pages
• Your own billing domain: connect billing.youragency.com from Commerce > Settings > DNS with two DNS records; SSL issues automatically and every share link and portal email switches over. Domains are re-checked hourly and certificates self-repair. See Billing Custom Domain
• Portal settings redesigned: a dedicated page with sections for contacts, permissions, appearance, emails, and advanced controls, live previews of the portal and its emails, and every field showing the value the portal actually renders
• Appearance controls: brand color, header style, welcome copy with merge tags, an editable help block, support contacts, and Terms/Privacy links in the portal footer
• Members area: clients see who at their company can use the portal, edit contact details, and portal admins add or remove members. Changing a sign-in email requires verifying both the old and the new address (on by default)
• Saved-card editing: clients update a card’s expiry date and billing address themselves; the card number and security code are never stored or editable (on by default)
• Contracts in the portal: clients open a contract, see signer status, sign when it is their turn, and download the completed PDF

​

Domains: buy, transfer and manage

• New Settings > Domains page. Every domain your account uses in one place, across all sites and environments. Domains you connected previously appear automatically
• Buy a domain without leaving Hiveku. Live availability across .com, .net, .org, .co, .io, .shop and .agency, filter by extension, sort by price, and pick a term from 1 to 10 years
• Registration and renewal prices are both shown before you pay. A promotional first year is common across the industry, so we fetch the real renewal figure rather than assuming it matches, and say so plainly when a particular name’s renewal price cannot be read
• Transfer a domain in from GoDaddy, Namecheap or anywhere else. We read the registry first and tell you which situation you are in: a domain whose DNS is hosted elsewhere transfers with no effect on your website or email at all, and one using its registrar’s own nameservers has its DNS set up here first, because those nameservers stop answering once that registrar is no longer in charge
• Transfers are blocked, not warned about, when proceeding would cost you something. A domain renewed in the last 45 days gains no time from a transfer while still costing a full year, so we refuse rather than let you pay for nothing
• Your authorisation code is encrypted the moment you paste it and never shown again
• DNS report on any domain, connected to Hiveku or not. Where its DNS actually lives, whether its email records are intact, and whether a certificate authority restriction would block HTTPS. Read-only, changes nothing, and names its own blind spots rather than implying it saw everything
• Automatic renewals with reminders at 35, 28, 21, 14, 7 and 3 days before expiry. Expiry takes a website and its email down on the same day, so the ladder starts early
• Hiveku is a reseller, and the page says so. Domains are registered through our accredited registrar partner, who appears in your public WHOIS record, and the emails you receive from them are genuine
• See Domains, Buying a Domain and Transferring a Domain In

​

Sensitive (write-only) environment variables

• New Sensitive toggle on any environment variable. Once saved, the value is hidden everywhere in Hiveku and cannot be revealed again, by you, by teammates, by Hiveku staff, by AI agents and the MCP tools, or by the VS Code extension and hiveku-sync
• Builds and deployments still receive the real value. Only the ability to read it back through Hiveku goes away
• Values cannot be un-hidden, because nobody can retrieve them to show you. Use Replace value to set a new one, or untick Sensitive while supplying a replacement to make the variable readable again
• The AI agent can create, update, delete and mark variables sensitive, but cannot read a sensitive value. It reports such a variable as configured rather than missing, so it will not loop trying to set one you already have
• hiveku-sync pull-env and the VS Code extension list which variables they skipped, so a local .env.local is never silently incomplete
• Sensitive values are stripped from stored build logs, which are readable through the dashboard and the deployment-log tools
• Marking a variable sensitive does not retroactively un-leak it. Rotate anything that may already be exposed. See Environment Variables

​

Environment Variables editor rebuilt

• Renaming a variable, or changing which environments it applies to, no longer leaves a duplicate behind. Previously the old entry survived and reappeared on reload
• Edits are staged and applied together when you click Save changes, with a summary of what will be added, updated and removed. Deletions show an Undo link and are not applied until you save
• Deleting a variable used to report success even when the delete had failed. Deletion is now applied atomically with the rest of your changes
• Variable names are normalised to uppercase as you type, so my key and MY_KEY can no longer become two separate entries that look identical
• Duplicate names, reserved platform names, and invalid formats are flagged inline on the offending row before you save, instead of failing after a round trip
• New Import for pasting a .env file. Handles comments, export prefixes, quoted values and values containing =, and reports unparseable lines instead of skipping them silently
• Concurrent edits are detected. If a teammate or the AI agent changed a variable while you had the page open, saving stops and names the conflict rather than overwriting their work
• All of a project’s variables are capped at 60KB, with a warning before you save rather than an opaque error after
• The editor now has its own Environment Variables page. Project Settings shows a read only summary with a Manage link, and never displays values

​

Security fixes

• Branch and pull-request previews were being given production environment values. They now correctly receive development values. If a variable is scoped only to Production, add a Development or All Environments value for it so previews still work
• Real .env files are no longer included when a project is exported to GitHub. .env.example templates are still exported

---

​

August 2026

​

Number porting

• Bring your existing numbers to Hiveku — a full-page wizard at Communications → Porting → New port request runs a free portability check with a per-number verdict (portable, FastPort eligible, SMS-capable, or not portable with the reason) before anything is committed. Bulk-paste tolerant of spreadsheet exports, up to 100 numbers per request
• Guided port-away paths for CallRail, GoHighLevel, and Twilio — step-by-step instructions, a copy-paste support-ticket template, and per-provider gotchas — plus a generic path for any other carrier
• Letter of Authorization e-sign — Hiveku prefills the LOA, the authorized person signs it in the browser via an emailed link (or on the spot), and the signed copy attaches to the port automatically. Already have a signed LOA? Upload it instead
• Every port gets an order page with a PORT-XXXXXX reference tracking the full pipeline — filed → sent to carrier → transfer date confirmed → transferred — plus a carrier-requirements checklist before you send, plain-language exception fixes with one-click resubmit, mid-flight carrier-detail updates (for providers that supply the account number and PIN after filing), and a carrier comment thread
• Client handoff for agencies — a secure no-login link where the account holder enters the account number and PIN, uploads a recent bill, and signs the authorization, so you never need the client’s provider login
• Account admins get in-app and email updates at every step, and freshly ported numbers land on the Numbers tab marked Needs setup — assign E911, activate, and choose where calls ring. See Number Porting

​

Call tracking: repeat callers, faster recycling, and the swap tester

• Repeat-caller matching — a caller who comes back within 90 days is matched to their original campaign and badged as a repeat caller on the CRM timeline, even after their tracking number was recycled to a new visitor
• Numbers recycle far faster: a number is held just two days after a call instead of a month, so a small pool now covers much more traffic
• Watch it swap opens your live site in a popup with an on-page verdict banner proving the swap; Test my live site runs full diagnostics — script detected, swappable numbers found, a real test assignment, and live pool availability
• Google Ads attribution now resolves down to campaign, ad group and keyword on each call record
• Optional GA4 send-back posts a phone_call event for qualifying tracked calls to your Analytics property, joined to the visitor’s real session
• Search engine crawlers always see your original number so indexed listings stay consistent, and account admins are emailed when a pool runs at capacity. See Phone Tracking

​

Listings & local presence

• New Marketing > Listings page — one health-and-action card per Google Business Profile location: a 0–100 Listing Score with a full breakdown, live profile data (hours, categories, attributes, contact info), warnings straight from Google (unverified, duplicate, marked closed), and a daily score history so a fix visibly moves the needle
• Every failing check comes with the action that fixes it: inline edits validated with Google before they apply, Fix with AI proposals applied only after you approve, and tracked tasks for name and address edits — those trigger Google re-verification, so they are deliberately not one-click
• Attributes audit — every attribute Google offers for your location’s category versus the ones you’ve set, with the top gaps named. Often the cheapest local-visibility win available
• Directory coverage with honest provenance: every claim is labeled synced, detected via scan, confirmed, or self-reported — never a fake sync. Includes an on-demand Google Maps presence scan (claimed status, public-data drift, possible duplicates; re-runs monthly on its own) and a guided checklist for Apple Business Connect, Bing, Yelp, and Facebook with per-field NAP copy helpers
• No more false alarms: a connected-but-unconfigured profile says “finish setup” instead of erroring, a client marked as having no Business Profile is skipped in reports, health checks, and AI analysis, and a failed sync always shows the last saved data with its date
• Listing health now flows into client reports (new listings section) and the weekly SEO digest, which calls out low scores, unverified locations, and Google-flagged duplicates. See Listings

​

Advertising: attribution columns & audience sync

• The PPC Dashboard’s campaign table now shows what happened on your side after the click: Identified visitors, ICP Match, Calls attributed via call tracking, Contacts, Deals (with won count), and Sales — click a count to see exactly who is behind it
• Audiences built under Marketing > Audiences can now sync to Google, Meta, TikTok, LinkedIn, and Vibe as customer-match / custom audiences over your existing ad connections — adds and removals stay in sync, the platform’s matched count reports back, and identifying data is hashed before upload so raw emails and phone numbers never leave Hiveku
• When a platform gates an upload (Meta’s Custom Audience terms, Google’s Customer Match eligibility), the sync row shows the reason and what to do about it. See Advertising Campaigns

​

Client portal expansion

• Portal lists (invoices, payments, estimates, contracts) are now full data tables — search, status filter, sortable columns, pagination — with an Export CSV button that downloads exactly what the current search and filter show. Invoices and estimates also get Download PDF
• New Payments tab in the portal — a searchable, exportable history of every payment a client has made. Turn it on with the See their payment history toggle under Commerce → Settings → Client portal
• Helpdesk live chat inside the portal — the launcher opens already signed in as the client (no name-and-email form), and each conversation opens a ticket linked to their contact record. Toggle: Show the live-chat widget; requires the helpdesk to be enabled on your account
• Manage client contacts without leaving portal settings — add, edit, bulk-send sign-in links, and remove contacts from the Client contacts card. Changing a contact’s email signs them out of all active portal sessions
• Invite to portal — a customizable welcome email (edit the subject and body under portal settings → Emails), sent from your own verified domain, optionally granting company-wide billing access in the same click
• Open as — view the portal exactly as a client sees it, read-only: payments and changes are blocked, an amber “Viewing as” banner stays on screen, and the session expires after 60 minutes. See Client Portal

​

Payment pages

• New Commerce → Payment pages: a hosted checkout at a public link that takes a payment from anyone — no invoice to send first. Every payment still creates a paid invoice and a receipt, so it lands in your reports like any other
• Links are served from your own host — your verified custom domain when you have one — and default to an unguessable name you can shorten to something printable whenever you want to put it on paper
• Control what a payer can enter (quick-pick amounts, fixed-amounts-only, or a min/max range) and what the form asks for — name, company, phone, a reference field, type-the-email-twice — with reorderable field order and tax-exemption claims
• A customizable payer confirmation email with merge tags and a Send a test to me button, sent from your verified sending address when your commerce email domain is set up
• CRM automations on payment: every payer becomes a contact without overwriting data you have curated, and you can optionally create and link their company and open a deal in a pipeline and stage you choose
• A webhook on every successful payment with a write-only bearer secret and a Send a test event button, plus per-page views, payments, conversion, and collected totals. Double-charge protection and card-testing rate limits are built in. See Payment Pages

​

Call recap emails

• Every completed call now sends a recap email — and it knows who handled the call. A call your AI agent handled arrives as an AI call recap; a call a teammate placed or answered arrives as a Call recap naming that person
• Recaps now reach the teammate who actually took the call, not just the account’s owners and admins
• When the call was recorded, the recap includes a Listen to the recording button that opens the call in your dashboard with the recording ready to play. Recordings are never attached to the email and never exposed through public links
• Calls land on the matching CRM contact’s activity timeline, attributed to the teammate (or AI agent) who handled them
• Configure the email, in-app notification, transcript inclusion, and recipients under AI Agent > Call recaps & notifications — and use Simulate call to send yourself a test recap. See the Communications Overview and AI Voice Agent

​

Rich MMS everywhere

• The SMS composer gained an emoji picker, GIF search, and image attachments (up to 10 per message) — in the SMS inbox, on CRM contact pages, in replies to SMS helpdesk tickets, and in sequence SMS steps
• Scheduled sends and sequence steps now deliver their attached media. Previously they silently sent the text and dropped the media. Media-only messages with no text are also now valid
• Size limits are type-aware with a clear error instead of a silent failure: still images up to 5 MB (carriers resize these in transit), animated GIFs, video and audio capped at 600 KB, because carriers strip oversized media while still delivering the text — the reason a GIF could show as delivered yet arrive blank. The GIF picker now picks a rendition small enough to deliver on its own
• Inbound texts appear in the open inbox in real time — no refresh needed
• The Sales and Helpdesk AI agents can now text in the conversations they work. The recipient is locked to the contact or ticket — an agent can never text an arbitrary number — and every agent send passes the same opt-out, registration and sending-cap gates as a manual send. See SMS Messaging

​

SMS opt-out handling

• A text starting with STOP, UNSUBSCRIBE, CANCEL, END, QUIT or STOPALL opts the sender out automatically, and every send path — manual, bulk, scheduled, automation, and AI agent — refuses opted-out recipients from that moment. START or UNSTOP opts back in, and confirmations are sent automatically. Nothing to implement
• New Do not text list on the Compliance tab showing every opted-out number. Add a number by hand when someone asks you to stop by phone or email — the obligation is the same, and the list is your record of honoring it. Removal is only appropriate with documented consent to resume texting
• A bare STOP no longer rings your whole team’s notifications — it is fully auto-handled, and the message still lands in the thread and audit trail. “STOP texting me” is a human talking and still notifies
• A newly purchased number can receive STOP texts meant for its previous owner — numbers are recycled, and strangers may still have old threads with them. These opt-outs are honored automatically and are nothing to worry about. See Compliance and Recording

​

Multiple mobile devices per person

• Sign in to the mobile app on as many phones as you like — each phone now registers as its own device, and all of them ring on incoming calls
• New Mobile devices panel on the Extensions tab listing every phone signed in to the app: owner, device model, OS and app version, the extension it rings, when it was last seen, and its push notification health
• Stop ringing this device revokes a lost phone or a former employee’s device instantly — incoming calls stop reaching it, and it only rings again if someone signs back in to the app on it. See Extensions and Devices

​

Reports Hub & Client Reports (beta)

• The reports page at Marketing > ROI Reports is now a full Reports hub with four tabs — Overview (campaign ROI attribution), Client reports, Journeys, and Custom (AI-built sheets and report sections) — each with its own shareable URL
• New client reports: scheduled cross-channel reports covering paid ads, SEO, local listings, social, email, calls, and website traffic — branded with your account name and logo, every number paired with a prior-period comparison, shared as a single public link that needs no login
• Weekly, monthly, or manual-only schedules; saved recipients receive every scheduled run by email, and a Send now button handles one-offs. Reports are snapshots — they update on schedule or on Regenerate, never silently in between
• Channels you haven’t connected simply don’t appear in a client report, and one channel’s data failing never blanks the rest — the report renders with a short note listing what’s missing
• The campaign ROI Overview gained custom date ranges, asset-type filter chips (Email, PPC, Social, Content, Sequence, Cold email), a Solid-matches-only toggle, CSV export, and an amber warning when ad spend data is more than 3 days stale
• Rolling out in beta — see the Reports Hub docs

​

AI chat: send follow-ups while the AI works

• Prompts sent while the AI is busy now queue and run in order, one at a time, instead of being dropped or forcing you to wait for the turn to finish
• A queue strip above the message input shows what is waiting. Expand it to reorder prompts by dragging, inspect their text and attachments, remove one, or clear all with confirmation
• Interrupt & run next stops the current turn and starts the next queued prompt immediately. Stopping a turn pauses the queue until you click Resume
• A queued prompt that fails to start is tried again automatically, then parked visibly with a Retry button after 3 attempts. Queued prompts are never silently deleted
• Stopping a turn mid-edit now discards that turn’s unsaved changes cleanly, so a cancelled prompt never leaves half-applied edits in your project
• The queue runs on the server, so queued prompts keep running in order even if you close the tab. See Message Queue

​

Undo this turn, rebuilt

• Undo this turn now reliably reverts everything an AI turn changed. Restore points reflect your project as of the moment you sent the prompt, so an undo can no longer include the turn’s own edits
• Turns that added pages or files are now fully undoable: files the turn created are removed, not just modified files restored
• A new confirmation dialog lists exactly which files will be restored or removed, then shows live progress with named phases and elapsed time. You can send it to the background, and closing it never cancels the restore
• Before anything changes, your current state is pinned as a safety checkpoint, so every undo is itself reversible from checkpoint history
• The outcome summary now tells you what was restored, what already matched, and which files were removed, and a failed preview refresh is reported instead of showing success
• Fixed the preview reload storm after an undo, and replaced the browser confirmation popup that Chrome could silently suppress, which made the button appear dead. See Undo an AI Turn

​

37 new workflow nodes across five new categories

• Five new node categories join the editor: Creative Studio, Web Research, Voice, Accounting & Billing, and Health & Safety Checks — the Add Node picker now spans 350+ nodes in ~30 categories
• Creative Studio generates AI images (single or sets of up to 10), short AI video clips, and royalty-free stock photo searches inside a workflow, feeding straight into CMS or social steps
• Web Research brings live search, single-page scrape, site crawl (up to 50 pages), LLM extraction with a JSON schema, and fast sitemap-based URL mapping
• The CMS family can now read as well as write: Read Entry, Search Entries, List Collections, Delete Entry (draft-aware), and Bulk Import of up to 200 entries in one call — so workflows can dedupe-check before writing
• Accounting action nodes pair with the billing triggers — record bill and invoice payments (with idempotency keys so a replayed run can’t book twice), AR/AP aging, and P&L summaries — while Voice nodes give read-only access to calls, transcripts, AI summaries, and numbers
• Health & Safety Checks make self-monitoring automations possible: test an integration’s auth, diagnose a broken deploy or missing conversion tracking, snapshot a checkpoint before risky steps, and audit account drift. See the new Workflow Node Reference

​

Merge variables never leak into customer messages

• An unresolved {{variable}} in any workflow node is now dropped instead of being sent as literal curly-brace text — no more {{trigger.phone}} showing up in a CRM record or an outbound email
• A field whose entire value fails to resolve is omitted: a create leaves it unset, and an update leaves the existing value alone instead of clearing it
• Every miss is flagged on the affected step in the run inspector (“N merge variables resolved to nothing”), so a typo’d tag surfaces in Run History rather than in someone’s inbox
• Where an empty fallback would be dangerous, the node now refuses to run instead — for example, a phone call node errors rather than dialing with the default caller ID

​

SMS triggers: scoping, full message text, and MMS

• The SMS Received trigger’s “To this number” filter is now enforced — a workflow scoped to one of your numbers fires only for texts sent to that number, instead of every number on the account
• The trigger payload now carries the full message text (previously truncated to a short preview), the sender’s matched CRM contact, and MMS attachments via {{trigger.media}}
• If a scoped trigger can’t determine the receiving number, it stays quiet rather than firing for the wrong number

​

CMS: real publishing, scheduling, and a rebuilt editor

• Every entry now shows one derived status — Draft, Published, Scheduled, or Archived — computed from the same rules your site uses to decide what renders, so the badge never disagrees with the live site. The entry sidebar shows exactly the actions valid for that state: Publish, Schedule, Unpublish, Archive, Publish draft, Restore
• Scheduled publishing that actually publishes: pick a date and time, and the entry goes live then — with an automatic production deploy by default (staging, dev preview, or no-deploy are one click away). Several entries scheduled together share a single deploy, and failed attempts surface their error instead of vanishing. See Scheduled Publishing
• Edits autosave into a private draft as you type — nothing reaches the live entry until you Save or Publish, and a published entry with pending draft edits says so with an amber dot plus Publish draft / Discard draft actions
• Publishing is honest about deploys: the sidebar’s Deploy state section shows Production, Staging and Dev, and an amber “Published changes aren’t live yet” card offers a one-click Deploy. Manual publishing never deploys behind your back
• The panel itself was rebuilt: a collections rail with groups and live entry counts, a real entries table (status filter chips, a column picker with drag-to-reorder, 25/50/100 page sizes, a tree view for hierarchical collections), and bulk Publish / Unpublish / Archive / Set field / Delete with per-row results — one bad row never sinks the batch
• New collections start from one-click presets — Blog, Products, Team Members, Testimonials, FAQs, Case Studies, with reference targets and publish fields included — or from a visual schema builder with per-field settings, drag-to-reorder, and nesting up to four levels. See the updated CMS docs

​

Section Library

• Browse 1,700+ professionally designed page sections — heroes, feature grids, pricing tables, testimonials, headers, footers — in the website editor’s new Components pane, every one with a preview thumbnail
• Click a section in your live preview and the catalog narrows to designs in that section’s category — click a hero, see heroes
• Use in chat writes a precise swap instruction into the AI chat that preserves your existing copy; Replace section now swaps the design in place instantly while you’re still blocking out a page
• Sections arrive as normal, editable code in your project — restyle them, rewrite them, or hand them back to the AI like anything else you built. See Section Library

​

Social share cards on every site

• Every Hiveku site now ships a branded 1200 x 630 share card by default, so links shared on Facebook, LinkedIn, X, and Slack get a proper preview instead of a cropped logo or a gray box
• The default is deliberately shy: it only appears when your project has no custom card, and a page-specific or site-wide image you set always wins
• Pages that customize other social fields (a share title, a description) no longer lose the image — the site default fills in wherever a page doesn’t bring its own
• Static sites get share-image, canonical, and Twitter-card tags by default too
• Changed your card but the old preview keeps showing? That’s the platform’s scrape cache — Social Share Cards covers how to force a refresh

​

Design Studio: AI video generation

• Brief in, MP4 out: describe the video you want and the AI plans a storyboard of scenes — AI clips, stills, and canvas layouts — then generates, voices, scores, and stitches it
• Nothing renders and nothing is charged until you approve: the storyboard card itemizes every scene and puts the exact price on the button — Approve & generate — $4.80. Approval is always a human action
• A Video tab inside the design editor holds the whole flow — approval, live per-scene progress, and the finished video — no separate page, no refreshing
• Revise one scene without regenerating the video: hover its tile, describe the change, confirm the cost. The new scene is seeded from the current frame so it stays continuous
• Your logo and all text are composited as crisp overlay layers, never baked into AI footage — no melted lettering
• Narrator voices are brand assets: audition and approve voices under Branding → Voiceovers, set a default narrator, and videos can only speak in voices your account approved. See AI Video

​

Social composer: first comments and link previews

• Add an optional First comment to any post — it publishes automatically as the post’s first comment the moment the post goes live, with nobody online. Facebook and LinkedIn show posts containing outbound links to fewer people, so moving the link into the first comment usually reaches more of your audience
• Works on Facebook, Instagram, LinkedIn (personal profiles and company Pages) and X. Instagram also gets its own per-platform first-comment field, the classic home for a hashtag block that would otherwise clutter the caption
• Paste a URL anywhere in the post body and the composer now detects it and shows an editable link preview card — adjust the title, description and image before publishing. Closing the card removes the link from the post entirely, so a hidden URL can no longer publish silently
• Facebook and LinkedIn render their own link cards from the URL, so the card you preview is the card your audience sees
• Account chips in the composer now carry a platform badge on each avatar, so the same brand’s Facebook Page and Instagram profile are distinguishable at a glance
• See the Composer

​

Social: safer scheduling, post stats, and comment replies

• A post scheduled for later no longer risks going out early: once a date is set, Save & Schedule becomes the primary button, and Publish Now asks for confirmation — naming the scheduled time — before discarding the schedule
• The posts list gains a Stats column: impressions, likes, comments and shares for every published post, summed across the platforms it went to. The first sync runs about 30 minutes after publishing
• Google Business Profile posts show a dash rather than waiting forever — Google discontinued per-post statistics
• Facebook impressions require a permission that Pages connected before August 2026 didn’t grant. Reconnect the Page under Marketing > Social > Accounts (one click, history preserved) to enable them; likes, comments and shares flow either way
• Reply to Facebook and Instagram comments straight from the Engagement inbox — previously LinkedIn only. Replies thread under the comment on the platform exactly as if you’d replied natively, and your own replies don’t echo back into the inbox
• See Analytics and the Engagement inbox

​

Social: schedule in your client’s timezone

• New Timezone setting under Settings > General. Set it and every scheduled-post time you pick means that zone — an operator in Chicago scheduling “3:00 PM” for a Sydney client gets Sydney’s 3:00 PM
• The composer names the zone right under the picker and shows the equivalent in your own time, so there’s never a question of whose clock you’re on. The calendar, queue and suggested best-time chips all render in the same zone
• Leave it unset and nothing changes — times mean each viewer’s own timezone, exactly as before
• Daylight saving is handled correctly: a post set for 9:00 AM stays 9:00 AM on the wall clock, even when a DST transition falls between scheduling and publishing
• See the Composer

​

Outbound pipeline stage automation

• Cold-email board columns are now configurable rules — define what puts a lead in each stage (reply category, lead tags, the Interested flag, or reply classification), optionally scoped to specific campaigns
• Entering a stage can create CRM records automatically: a contact (created or updated, never duplicated), the company, and a deal with a configurable pipeline, stage, value, currency, name template, and owner
• Each automation runs once per lead per stage — a lead re-entering a column never creates a duplicate deal, and the deal is never modified or deleted afterward
• A live “N leads match” preview shows a rule’s reach as you edit, and switching automation on above 200 matching leads asks for confirmation with the exact count before anything is created
• Dragging a card by hand pins it so rules won’t move it back, and asks before running the target stage’s CRM automation — pausing the sequence is now an explicit per-stage choice instead of a silent side effect of every drag
• Board columns now show exact lead counts with per-column paging, plus an always-present Unassigned column so leads matching no rule stay visible — see Outbound Lead Pipeline

​

Form capture: spam protection and duplicate prevention

• Every captured form submission is now scored for bot fingerprints the moment it arrives — filled honeypots, inhuman submission speed, burst rates, link-only bodies, disposable email domains. Suspicious submissions are filed under the Spam filter in Analytics → Forms and never emailed to you, never added to your CRM, and never fire a workflow
• Nothing is ever deleted. A spam-filed submission stays visible and is one click from restored — Not spam returns it to your inbox and delivers the notification email it was originally denied, within a few minutes
• Optional Google reCAPTCHA on every form: toggle Form spam protection on under Hosting → Integrations, per environment. It is invisible — no checkbox, no puzzle, and Google is not contacted until a visitor focuses a form field. No keys or Google account needed; it runs on Hiveku’s platform configuration
• Pick how aggressive to be, per project: Lenient files only certain spam, Balanced is the default, Strict files borderline submissions too — because a false positive on a high-value quote form costs more than one junk email does
• One form under sustained attack can require a visible “I am not a robot” checkbox on just that form, from the Forms tab’s per-form settings, leaving the rest of your forms friction-free
• One physical submission is now exactly one lead. Captures of the same submit seen through different channels are linked into a single row, a single notification email, and a single CRM contact — and URL variants like a trailing slash no longer split one form into duplicate records in the Forms tab. See Form Submissions

​

Sensitive (write-only) environment variables

• Mark any environment variable Sensitive to make it write-only. After you save, the value is hidden everywhere in Hiveku and cannot be revealed again — not by teammates, Hiveku staff, AI agents, or the VS Code extension. Builds and deployments still receive the real value
• Leave the blank field alone on later saves to keep the stored value — blank never means clear for a sensitive variable, so renaming it, changing its environments, or editing other rows never requires re-entering anything
• To change a sensitive value, click Replace value and supply a new one; a sensitive variable can never be saved empty. To make it readable again, untick Sensitive while supplying a replacement at the same time
• Branch and pull-request previews now receive development values, never production ones. If a variable is scoped only to Production, add a Development or All Environments value so previews still work
• See Environment Variables

​

Mobile app on multiple phones

• Sign in to the Hiveku app on as many phones as you like — each phone now registers as its own device with its own extension, and incoming calls ring all of them
• Every phone identifies itself, so you can tell devices apart: model, OS and app version, the extension it rings, when it was last seen, and its push notification health
• New Mobile devices panel on Communications → Extensions lists every phone signed in to the mobile app across your team
• Stop ringing this device silences a single phone — the lever for a lost phone or a former employee’s device. It only starts ringing again if someone signs in to the app on it
• Needs app version 1.2.8 or later. Phones on older builds keep working on a shared extension and show an “older app build” badge until they update
• See Hiveku Mobile App and Extensions and Devices

​

Mobile app experience overhaul

• The app feels faster everywhere: buttons respond when pressed, screens show skeleton previews while loading, lists keep their results on screen while you filter, and detail screens open instantly from data the app already has
• Platform-native polish: haptic feedback, cached images, long-press to copy, system share sheets, and home-screen shortcuts — press and hold the app icon for Dial, New chat or Inbox
• Links to app.hiveku.com now open directly in the app, already signed in — no browser detour, no second login
• Android notifications get one channel per category, so silencing one type no longer silences the rest, and the app icon shows an unread badge on both platforms
• New global Search across contacts, companies, deals, projects and tickets, plus a customizable tab bar — drag to reorder, Chat and More stay pinned at the ends
• iPad layouts, VoiceOver and TalkBack labels across the app, and most updates now install themselves in the background when you open the app — no store update needed. See Hiveku Mobile App

​

July 2026

​

Call tracking and dynamic number insertion

• Attach a tracking number pool to a website project and every visitor sees their own number — swapped into tel: links and any configured plain-text numbers, rendered in the same visual format as the number it replaces, extensions preserved
• Assignments are sticky: a visitor keeps their number for about 30 minutes of browsing, extended after they call, so page navigation and quick return visits show the same number
• Every tracked call ties back to the visitor’s session — UTM source, medium and campaign, landing page, referrer, and ad click IDs from Google, Facebook, Microsoft and TikTok
• New Calls view on each project’s Analytics tab: answered/missed/duration KPIs, per-campaign call counts, and a call log with inline recording playback
• Missed-call SMS autoresponder texts the caller the moment an inbound call is missed or goes to voicemail, with merge variables and a live segment counter
• Opt-in Google Ads offline conversion uploads push attributed calls back to Google Ads so Smart Bidding learns which campaigns make the phone ring — with answered-only and minimum-duration policies and an optional conversion value. See Phone Tracking

​

Workflow automation overhaul

• The Form Submitted trigger now fires for every form submission path — embedded forms, hosted forms, and forms captured on deployed sites all reach your workflows, with spam and duplicate submissions filtered out before the trigger fires
• AI Agent nodes can route to a department agent (with your brand voice, memory, and tools) instead of a bare model, return structured JSON you can branch on with a Condition node, and run specialist sub-agents whose findings feed the main prompt
• New Test Run button in the editor — watch each node light up on the canvas with its output as it executes; nodes that would send, publish, or charge are simulated and show exactly what they would have done. Enabling a workflow now validates it first and names any broken node
• Merge variables got a real editor: type {{ in any text field for autocomplete, and inserted references render as pills that show whether they resolve. Unresolved variables are listed per step in the run inspector
• Fairer auto-pause: manual runs, Test Runs, and replays never count toward the failure circuit breaker, so iterating on a half-built node can’t pause a workflow whose live traffic is healthy. Auto-pause now emails account owners with the reason and a resume link, and resuming replays every trigger event that arrived while paused, oldest first
• New daily AI budget: each workflow may execute at most 200 AI-agent node runs per day (25 per run). Crossing it pauses the workflow and emails the owners, so a mis-wired loop can’t burn AI credit overnight

​

May 2026

​

GitHub source-of-truth model

• When GitHub is connected to a project, GitHub is now the canonical source of code — the Hiveku database is a working buffer reconciled against GitHub at every deploy
• AI chat sessions auto-commit to GitHub at session end. A 10-second debounce batches back-to-back chat turns into a single hiveku[bot] commit, prefixed [hiveku] Sync from Hiveku
• Bot commits never trigger an auto-deploy. The auto-deploy toggles continue to fire only for pushes made outside Hiveku (your laptop, VSCode, external collaborators, GitHub Actions, etc.)
• Every deploy starts with a pre-deploy sync. If the database and GitHub have BOTH modified the same file since the last sync, the deploy is paused with a conflict notice — you pick which version wins, then retry. Previously these conflicts could silently overwrite work
• External MCP agents (Claude Code, Cursor, etc.) gain a project_commit invariant — tool descriptions now state that committing is required for persistence on GitHub-connected projects, with explicit warnings about the auto-overwrite risk
• See GitHub Integration for the full model and Deployments for how the pre-deploy sync gate works

​

Native Mobile (Expo) project type

• New project type — pick Native Mobile in the new-project dialog to scaffold a real Expo Router app with Supabase auth, items CRUD, NativeWind, and protected route groups; Supabase database is provisioned automatically
• Per-project Fly.io preview running Metro + React Native Web inside an iPhone/Android phone-frame chrome (11 device presets, landscape mode, accent color picker — selection persists per-project in localStorage)
• AI agent trained on Expo Router + NativeWind conventions; knows the difference between an OTA-deliverable JS edit and a native-dependency change that requires a Build & Ship
• Mobile Credentials page — connect your Apple Developer (.p8 ASC API key) and Google Play (service account JSON) accounts; live preflight validation via signed JWTs, encrypted storage in AWS Secrets Manager
• One-click Build & Ship — kicks off parallel iOS + Android EAS builds, auto-creates the iOS App Store Connect record on first build, and auto-submits to TestFlight + Play Console Internal track
• Build cancel button, email notifications on success/failure, decoded Apple/Google error messages with Fix with AI deeplink (~60% of common failures resolvable end-to-end)
• OTA Update channels (Preview + Production) with atomic Promote button; bundle size warnings (amber at 75 MB, red at 100 MB) with per-package breakdown
• App Store Assets generator — one form turns app name + tagline + accent color into a 1024×1024 icon, splash screen, and store-listing copy (subtitle, description, keywords)
• GitHub integration parity with web projects — Build & Ship tags the commit, AI edits auto-commit, .env excluded from sync
• See the Mobile Quickstart to ship your first build in ~60 minutes

​

Design Studio — Layered Motion Graphics

• New animated design surface at /dashboard/marketing/design/ — every text, shape, image, and CTA stays an editable Fabric layer while motion plays
• Animation is a property of layers (not a baked MP4), so dragging the headline mid-playback, retyping any word, or swapping a color via the picker keeps motion working
• AI design agent generates the static frame using your brand kit, then animates each layer with coherent strategies — Cascade Fade-Up, Kinetic, Hero Scale + Fade, or Slideshow
• 14 enter/exit presets + 4 loop motions (pulse, wiggle, breathe, rotate-slow) shared across the AI tool, the per-layer Style panel editor, and the MP4 export compositor
• 18 layered animated starter templates spanning Instagram / Reel / TikTok / YouTube / email-banner surfaces — fully editable from frame zero, no template lock-in
• Per-layer animation editor in the right Style panel with debounced sliders (250ms), auto-pause on selection, and snapshot-and-restore runtime that prevents drift
• Canvas-level Animation panel with one-click strategy buttons that preserve existing loop motions when re-strategizing
• MP4 export via Remotion render worker; PNG export client-side for static designs
• See the Design Studio docs for full coverage

---

​

April 2026

​

Hiveku CMS (Webflow-style structured content)

• New manifest-driven CMS — declare collections in hiveku.cms.json with typed fields, references, and validation
• CMS panel in /v3 with a Webflow-style entry list, form editor, version history, and bulk import
• File-backed storage: MDX for blog-like content, JSON for structured data — all in content/
• AI-driven scaffolding: agents read your existing site and migrate hardcoded data into the CMS
• 13 supported field types including reference for linking collections (blog → author, product → category)

​

Email Marketing Agent

• New Email Marketing department with a full ESP (audiences, campaigns, templates, sequences, deliverability)
• AI coach “Lumen” embedded on every email page — drafts newsletters, builds audiences, triages deliverability
• Default memory, 4 skills, 3 rules, and the coach persona auto-seeded on first enable
• Cross-agent delegation: content, branding, and ppc agents can hand work to the email team

​

Sales AI Co-pilot

• AI co-pilot layered on top of CRM — drafts replies, summarizes meetings, surfaces follow-ups
• Approval-first: every draft and suggestion stages in a pending actions queue
• Reads meeting transcripts, calendar, and contact activity for full context
• New how-tos for AI SDR, sales sequences, and reminders

​

Calendar & Booking

• Calendly-style scheduling built into Hiveku
• Meeting types, availability rules, public booking page, embeddable widget
• Bookings flow into CRM and trigger workflows automatically
• Team calendars with multi-member availability sync

​

Communications (Voice & SMS)

• Production phone system with inbound/outbound calls, IVR, voicemail, call recording
• Phone number provisioning, extensions, ring groups, call queues
• AI agent routing — calls can be handled by an AI agent with handoff to human
• SMS sending via Twilio integration
• E911 compliance built in

​

Commerce

• Invoices, estimates, contracts (with e-signature), subscriptions, payments
• Client portal for customers to view and pay invoices
• AR aging, MRR/ARR reports, payment trend dashboards
• Product catalog and order management coming soon

​

Better Stack Integration

• Added uptime monitoring for app.hiveku.com
• Incident alerting via phone and email

​

Documentation Launch

• Launched public documentation on Mintlify
• API reference for Builder, Marketing, CRM, and Workflows
• 100+ how-to guides covering every major feature

---

​

Recent Highlights

​

AI Builder Enhancements

• Multi-model support via OpenRouter (GPT-5.4, Claude, Gemini, Grok)
• AI mode switching (Coder, Designer, Full-Stack)
• Context condensation for long conversations
• Model failover with automatic retry

​

Visual Editor

• Click-to-edit WYSIWYG editing
• Component library with drag-and-drop sections
• Figma import with multi-page support

​

Marketing Suite

• Social media scheduling with LinkedIn integration
• Outbound email with lead management and sequences
• Content management with review workflows
• SEO tools with Google Search Console integration

​

Multi-Runtime Deployment

• Support for Next.js, Python, Go, PHP, Ruby, .NET, Rust
• AWS ECS containers with automatic SSL
• CloudFront CDN
• Custom domain support

​

Database Management

• Supabase PostgreSQL provisioning
• SQL editor with direct query execution
• Automated backups to S3
• Point-in-time restore via checkpoints

​

Workflow Automation

• Visual workflow builder
• Webhook, schedule, and database triggers
• Dead letter queue for failed executions
• Version history and rollback