What we shipped, when, and why it matters to your workflow.
When an agent sent the short version for one network through the API, the per-account copies of every other network were replaced by nothing, so a Bluesky version a teammate had written disappeared and the post was refused as too long. Now the API merges per account: only the accounts you send change, the rest keep their copy.
Read the update →A keyword search read every network word by word instead of using the search index, so a search across all networks took as long as the slowest full scan. The index is back on every network (Pinterest and Douyin included), and on paid plans each network is asked at the same time, so the first results paint while the rest arrive. In the per-network view, a network with nothing for your keyword moves to the end instead of holding its slot empty.
Read the update →The posts that carry a keyword are usually weeks or months old, and the radar opened every search on the last 7 days, so a search found nothing while the posts were right there. Now typing a keyword switches the range to All time and the order to Most viral by itself. Narrow the range afterwards and the radar respects it; clear the keyword and your previous range comes back.
Read the update →Reports has a new tab, Grades. For each collaborator (or the whole team) it lists everything they scheduled with the grade the reviewer gave it: risk from 1 to 10, image against text, fact check, post kind, whether it reads as AI-written, the verdict and who reviewed it. Averages read like an exam, by day, week, fortnight, month and three months; a review by the author never counts. A row opens the post. Owners and admins see everyone; a member sees their own.
Read the update →When your text passes the limit of a selected network (280 on X, 300 on Bluesky, 500 on Threads), an alert under the post names each account, says by how many characters, and shows the exact tail that would be cut. One click opens the per-account editors on that network, where the characters past the limit are painted and a Trim button cuts at the last space before it. The other networks keep the full text.
Read the update →API Access has one form now: who will use the token (a new AI agent or an existing member) and what it may do. Five groups: Reader (only reads), Writer (finds content, drafts, submits to Review), Reviewer (grades what waits in Review), Publisher (writes, reviews and approves) and Full access. A token outside its group gets a clear 403 naming the permission it lacks, and GET /account shows the token its own group. Tokens created before keep full access.
Read the update →When a search of yours on any network or on the radar comes back empty, or with fewer than five results, the keyword goes straight to our extraction queue and our extension goes out to search it on X, Facebook, Instagram, TikTok, Pinterest and Quora, scrolling as a person would. Once every network has been swept you get one email with what it found and a link to the radar already filtered. You can turn the email off in Notifications.
Read the update →The Radar now opens on the By network view for everyone: one ranked panel per network, so a quiet network is never buried under a loud one. If you had picked another view, your choice still wins.
Read the update →On the Team page the owner can tick "Publishes own posts after the grade" for a member. From then on that person sends their own post from Review once the reviewer graded it: on an OK, or over a "needs fixes" if they decide to publish anyway; a block still stops it. The grade stays on their record, and the Team page shows each collaborator's average grade by week, fortnight and month, how many posts they sent themselves and how many over the reviewer's objections. Fewer people in the loop, the same account safety.
Read the update →One row of chips in the composer: Fact, Meme, Quote, Opinion, Promo. The reviewer applies only that kind's rules, so a meme is never sent back for a missing source and a quote is never checked as a news item. A post made from a board card takes the card's category by itself (Memes, Quotes, News…), and on the Drafts page the kind can be changed with one select. Less strictness where it never made sense, the same rules where they protect the account.
Read the update →When you customize per account, every account starts as a copy of the post and you change only what that network needs: a shorter text for X or Bluesky, another picture, its own first comment, its own thread. Threads now work on Bluesky too, posted as a chain of replies once the post is live. A Preview button shows how each account will show the post: the text folded or cut where that network folds or cuts it, the pictures, the thread, the first comment. And editing a draft brings its per-account copies back (they used to be lost).
Read the update →A post now carries a private field the reviewer reads and nobody publishes: the source link, where the picture comes from, what was checked. Links in the body cut the reach on every network, so the body names the source and the link goes in the notes or in the first comment; the content rule says so, and a "no source" flag now means "no link in the notes". Adding the notes to a post that was sent back asks the reviewer for a fresh look, right from the Drafts page.
Read the update →Editing a post that was already sent to Review saved nothing and showed no error, so the author could not fix a typo without waiting for the reviewer to send it back. Now the author (or an owner or admin) edits it in place; it stays in Review and the reviewer gives a fresh verdict on the new text. A member can only edit or delete their own posts.
Read the update →When you let the reviewer agent publish on its own OK, the routine goes out and the doubtful still waits for you: a post whose risk is above 5 out of 10, whose facts score under 70 or whose picture matches the text under 60 is kept for a person even with an OK. Each post also carries the language its project declares, so a Spanish post in an English project is sent back. And the reviewer can read the flags per collaborator to report them.
Read the update →A new base next to the quotes: the international days declared by the UN, UNESCO and the WHO, and the anniversaries of science, technology, space, health, the environment and civilization, each with what is celebrated or what happened and why it matters, in English and Spanish. A page in the app to plan the month, an API endpoint for your agent, a recipe source for standing orders like "one post on every big science day", and the ephemeris template to render them.
Read the update →The news corpus has had categories for a long time (Politics, Technology, Artificial Intelligence, Science, Health and a dozen more) and the API could filter by them, but nothing told an agent they existed, so "technology news" came back as the whole corpus. Now the API lists them with how many articles each one brought this week and this month, trending takes one by name, and a wrong name or an unknown parameter is an error instead of a silent full listing. The skill and the MCP server know the endpoint.
Read the update →On API Access, "Create an agent" now asks whether the agent writes posts or reviews them, and after creating a reviewer it tells you the next step: it starts in verdict-only mode, and you give it the key to publish from the Team page once its judgement matches yours. The agent guide heads that page instead of being one more entry in the account menu, and the operator panel left the account menu too.
Read the update →The reviewer agent now checks four more things on every post in Review: that the picture shows what the text says, spelling and grammar, whether the text reads as written by an AI, and the facts, with one community rule: a scientific, medical or statistical claim needs a DOI or a link to the primary source, or the post goes back with a flag. Every post gets a danger reading from 1 (safe) to 10 (breaks a network's terms). The Team page shows, per collaborator, how many flags they got this week, this fortnight and this month, and for which rule. A review can name the exact content it judged: if the post was edited in between, the server refuses it and asks for a fresh look.
Read the update →Four things closed after an audit of the API, the skill and the team extension. The extension signs every post it sends with a secret you paste once in its options page (version 1.31.0), and the server can now refuse anything unsigned. The server no longer fetches a media URL that points at a private or local address, whoever sends it. An uploaded file is refused when its bytes do not match its extension. And a token can be created for an owner or an admin only by the owner, because a token carries every right of the person it belongs to.
Read the update →Open any post in Scheduled or Published and the detail now ends with its delivery log: created, sent to PostProxy and accepted with its id, published with its link, refused with the network's own reason, retried, canceled, reviewed, approved, each with its time. "Copy for support" puts the post, its targets with their last state and the whole log on the clipboard, written so you can paste it into a support message as it is. The API returns the same log on every post (events and support_text). The log starts with what happens from today; older posts show their last state only.
Read the update →The sidebar shows the plan this workspace is on, with an Upgrade button for the owner on the free plan that opens the order form with Creator preselected. The thirteen network pages fold under one line, "Networks, one by one", open on a network page or when you open it, so the sidebar reads as a handful of things instead of twenty. The "See what you get" buttons on the free plan now say "Upgrade your account" and open the order form with the Creator plan already selected, instead of the pricing page. The Radar sits at the top of the networks list, where it belongs. Settings in the account menu opens your organization settings (or your notifications, for a member) instead of the billing site, which has its own entry, "Billing and profile". And the agent guide starts with the three connection steps and every link in one place: API key, skill on GitHub, MCP server on GitHub and npm, the install commands for Claude Code, Claude Desktop and claude.ai, the API reference; then the prompts, the ten-minute reviewer, and the reference of templates and categories.
Read the update →When you close the welcome card, a short tour walks the sidebar once: the dashboard, the radar, the editorial board, Publish and its stages, templates, recipes, the network pages and your account. The radar opens in the by-network view for a new workspace (one panel per network reads better the first time); anyone who already picked a view keeps it. And a brand-new account no longer sees "new since your last visit": the welcome card is what it sees, and the updates card starts counting from that day.
Read the update →Two "Approve and schedule" clicks a second apart both went out: the networks answered that an identical post had just been created, and the second answer overwrote the first. Now the server lets exactly one send through per post and tells the second click the post is already on its way; on the Drafts and Review page every button of the card switches off with a spinner while the request runs. The review buttons are labelled for what they are: OK means it passes the content rule, publishing is the buttons on the right.
Read the update →The content rule now lives on the platform (what blocks a post, what sends it back to Drafts, scores and thresholds, what changes per network), and the agent reads it on every pass instead of remembering it. A reviewer token takes the posts in Review that have no verdict on their current content, judges them network by network, leaves a note the author reads, and its OK schedules the post at its time. The server has the last word: it recalculates the verdict from the scores, never sends on an OK from the same token that created the post, only lets an owner, admin or a member ticked "Their OK publishes" on the Team page release a post, refuses anything under 15 minutes away, and drops an OK the moment the post is edited. Skill 1.6.0 and MCP 1.1.0 carry the job and the prompt that sets it up.
Read the update →The right column of Compose had been empty since the queue moved to the Scheduled and Published tabs. It now shows Up next: the next posts on the calendar across every project, each with its time, how long until it goes out, its networks and a thumbnail, and a click opens the post. Under it, Just published: the last five posts that went out, with their date, time and every network with its link or its failure, as the queue showed before. Above the list, one line says how many drafts are in progress and how many posts wait for authorization, each linking to its page.
Read the update →A 134 MB phone video went out on Facebook, LinkedIn and Threads but Instagram answered only "ERROR" and gave up after ten attempts. Instagram's processor is the strictest about codec, size, frame rate and bitrate, so videos are now re-encoded to the profile it accepts (H.264 and AAC, up to 1080 x 1920, 30 fps, 8 Mb/s) before they are sent to Instagram, Threads, TikTok and X; the other networks get the original. A network that still refuses gets a "Retry on Instagram" button on the post, which sends it again to that network alone. And a network that never confirms is called failed six hours after the post's time instead of staying "processing" for days, so you hear about it.
Read the update →Every Publish page carries the same bar at the top, with a count on each stage: what you are writing, what is being worked on, what waits for an owner or admin, what has a date, and what went out. Compose shows only the composer; Scheduled and Published show the queue as a list or a calendar, and remember your choice. A "?" in the bar starts a short guided tour of the page you are on, in English or in Spanish if that is your language; it also plays by itself the first time you open each stage.
Read the update →A video or picture you publish is needed only until the networks have fetched it. Two days after a post goes out, its files are deleted from our storage: a small thumbnail stays in the queue and the live post is one click away on each network. Drafts nobody finished in a week lose their files too; the post stays and asks for the media again. Evergreen posts keep theirs, because a remake reuses them. Less stored, less to pay for, nothing you still need gone.
Read the update →A draft is a post being worked on, by a person or by the agent. When it is complete and has its date (or "now"), Send to review moves it to Review, where an owner or admin presses Publish now or Approve and schedule. Needs fixes sends it back to Drafts with the note. Members and agents can only leave drafts and send them to review; the owner always publishes directly, and admins too unless the organization says otherwise. The Drafts page shows the two stages as tabs, the composer says where each post went, and all the publishing rules now live on one page: Team.
Read the update →The owner can now change a teammate's role in Team or in Organization Settings, Security: admin publishes directly on every connected network, connects accounts and approves drafts; member leaves drafts for an owner or admin to publish. Agents stay members. The two pages also show who has two-step verification on.
Read the update →Every account can now turn on two-step verification with Google Authenticator, 1Password, Microsoft Authenticator or any TOTP app: in your account, under Useful Links, Two-Factor Authentication. The reason is the connected social accounts: with a password alone, whoever steals it posts on every network of the organization. The dashboard reminds anyone who has not turned it on, and on organizations that publish the Publish, Drafts and Accounts pages require it before going on. Owners see on the Team page who has it on and who does not, with a switch "Require two-step verification to publish" that is on by default; the dashboard tells them how many teammates still have to turn it on.
Read the update →A post written by a team member, or by an agent through its API token, now always lands in Drafts for an owner or admin to approve, even when the organization does not review every post. The owner is never held back, and admins only when the organization reviews every post. The composer button says "Submit for review" for whoever is, and a draft is opened from the Drafts page with Publish now, or Approve and schedule when it carries a time. The reason is security: a stolen member session or a leaked agent key can no longer post on every connected network. The API announces it in advance (must_draft on targets), and the Claude skill and the MCP server tell the agent to say "it is in Drafts" instead of "published".
Read the update →Choosing "All" used to run every network one after the other and wait for the slowest before showing anything. The radar now asks each network separately, in parallel, and paints the rows as they land, so the first posts appear in a second or two. Each network's ranking is also kept for ten minutes, so the second load of the same view is instant. And the list view is compact: one line per post, a small thumbnail, half the padding, so a screen holds twice the posts. Hover a row for the full text and every metric, as before.
Read the update →Some news sites (La Jornada among them) refuse any automated reader with a Cloudflare challenge, and the card said "that link gave nothing to read". The reader now climbs a ladder: the page itself, a residential line, a rendering reader, and finally a metadata service, so title, description, picture, publisher and date arrive anyway. When nothing works, the message says why instead of a bare "nothing to read". The same ladder serves the cards an agent creates from a URL.
Read the update →New workspaces now open with one screen: what the free plan does and what paid adds, a big button that creates the API key for your agent right there and shows exactly what to paste in claude.ai, Claude Code, any MCP client, ChatGPT, OpenClaw, n8n or Make, and a six-tile tour of where things are: Radar, Board, Drafts, Templates, API access, Agent guide. Creating the workspace also has an express path for people who came from an agent directory: skip the form, sensible defaults, the key on the next screen. The Claude plugin now carries its own icon.
Read the update →A Collage button lays out two, three, four or six empty photo boxes at once; click a box to fill it, or drop image files from your desktop straight onto the card and each one becomes a layer where it landed. Every image now has a floating toolbar: change the picture, crop (drag the photo inside its box, wheel to zoom), fill or fit, send it behind the template, duplicate, delete. Text boxes and shapes get the same bar with their colour and bold. The Layers panel adds Text, Image, Shape and Collage right where the list is.
Read the update →The picture of a template used to be a fixed crop. Now drag it to the part you want in the frame from the first pull, zoom with the wheel or the + and - buttons, and reset with one click; the saved PNG and an agent's render frame it the same way. The side panel now lists every layer of the card like Canva does: your text boxes, images and shapes, and the template's own blocks (photo, headline, logo, brand name...). Click one to select it, the eye hides it, so a logo box you do not want is one click away from gone. Text, Image and Shape in the top bar add layers; a collage is several Image layers.
Read the update →Trends waiting and Evergreen due to remake used to open as two full boxes with nothing in them, and a new account read that as something broken. When they are empty they now fold to their title with a short hint of what fills them; Show opens the full explanation, and they unfold by themselves the moment there is something to see.
Read the update →The first page after login now opens with big, visible tiles to the places you use every day: Drafts, Publish, Radar, Board, Templates, Accounts, Stats, the Agent guide. Customize them: add any page of the app, remove what you do not use, drag to reorder; each person keeps their own set. And the agent skill learned to onboard: on the first conversation it says in three lines what it can do, reads where you are (plan, accounts, templates, drafts) and proposes one next step at a time.
Read the update →Bluesky takes images up to 2 MB. A rendered design in PNG often weighs a little more, Bluesky refused it, and the post still showed as published with nothing behind it. Now every image is fitted to each network before sending (a lighter JPEG copy where the ceiling asks for it), and a refusal by any network shows as partial with the network's own reason and a notification.
Read the update →A brand that publishes in two languages runs two projects. Link them in Projects ("also publish translated in…"), and when you approve a draft the app asks right there: also in English? The linked project and its accounts come preselected. The copy is translated by the app, a design on a template re-renders itself in the new language, and a plain picture can be swapped for its translated version before sending.
Read the update →A draft is a complete post, with its picture, copy, accounts and time, that nobody has sent yet. People and agents leave posts there; people, agents or the AI review them with a verdict, a score and warnings per network; an owner approves, or the post goes out by itself when a review says OK, if you allow it. A card on the board has two ways out: Send to drafts with a tentative time, or Schedule this card for what you already checked.
Read the update →Content Studio now starts with the image. Pick a template by category with live thumbnails (favourites first), it fills itself from the card, and one instruction to the AI writes the headline, the highlighted words and a post caption that tells the story, or changes the look: darker, bigger headline, logo up. Tweak it in the full editor, save the PNG on the card, and schedule the card with that image.
Read the update →Add what a template does not define: text boxes, uploaded images and shapes, dragged and resized on the card, snapping to the centre, the edges and the other elements. Put an image behind the template as a background, align to the card with one click, duplicate, and see every element in a layers list. A text box can be a variable the agent fills, with your instructions. Download the card as PNG from the browser. View the code of any template, take it to your own AI, and import what it gives back, sanitized.
Read the update →The skill taught the agent; nothing taught you what to ask. The new Agent guide in the app lists the quote categories we hold, the template slugs, how recipes work, and six prompts to copy: one quote a day, two templates alternating with mixed categories, using the template you customized, having the agent build one, turning the top rising post into a card, and asking what it did. The quotes base now shows the sum of its indicators and sorts by it.
Read the update →GET /v1/search asks all twelve networks for a keyword and returns one merged, ranked list with a per-network total. Underneath, the queries that took 40 to 60 seconds on Reddit and Bluesky now take under a second: text indexes for keywords, windows on the post date, and pages picked before rows are read. time_range takes any number with a unit and refuses what it does not understand. Pinterest rows carry the pin id, its URL and its image.
Read the update →Typing in the Pinterest search no longer fires a query on every letter: press Enter or the Search button. The grid deals pins row by row, so the top results are the first row, not the tallest column. The extension (1.29.0) takes one keyword, asks Pinterest what it suggests around it, and scrapes all of them one after another without stopping halfway down the page.
Read the update →A keyword search on /trending answered server_error on every source. Fixed and verified on all twelve. Titles and descriptions no longer carry HTML entities such as " or &, and the docs console has two keyword examples to try.
Read the update →Every community page now opens with the community's own description, its size and its founding date, read from Reddit and refreshed every few days. Search Console said the one question those pages win is "what is r/X?", so the page answers it first: in its title, its first paragraph and its first FAQ.
Read the update →A template you customized is now your own: name it, write in one field what your agent should use it for, duplicate it into variants, and start a recipe from it in one click. A blank layout with every block optional lets you build one from scratch. Agents read your instructions through the API and follow them over the general rules.
Read the update →Templates can now be restyled by a person, not only by an agent. Click an element on the card and its controls light up; type the texts straight on the card; drag the logo where you want it; select a few words and give them their own colour, gradient or font; pick one of twelve palettes and ten type pairings. Saving a global template creates your own copy. New: eight quote templates and editable versions of the image, photo hook, stat, tip and ephemeris templates, with variants.
Read the update →For quote posts and daily quote schedules: a base of quotes merged across sources, each with its author, the work it comes from, categories in English and Spanish, and whether the author's work is in the public domain. Author portraits come from Wikimedia Commons with their licence and credit. Read it at /quotes, or ask GET /v1/quotes for one that fits a category, a language and a length.
Read the update →A recipe says what to make (a quote, a trending post, or something you hand it), with which template, in which format, for which networks, how often and at what time. Your agent reads the list through the API and runs it. Nothing runs on the server, so you keep the last word on every post.
Read the update →A new page, viralhunt.io/claude, walks you from a free account to a working skill or MCP server in three steps. Free accounts now get an API key with 24 content queries a day, and the API reports what is left.
Read the update →The radar is where trends get picked up, so it now does the picking: assigning sets the priority (high by default), an assigned row says who has it and since when, and a new sort ranks posts by the engagement they gained in the last day. The dashboard stops repeating the radar and keeps a five-row taste of your niche.
Read the update →High-priority cards rise to the top in red, every card says how long it has waited, the dashboard and a first-login reminder list what is waiting on you and on the team, every assignment email carries the whole backlog, and Monday brings the team scorecard.
Read the update →Four new endpoints give an agent the same numbers our public pages show, with the sample size and the window every time. The MCP server (0.2.0) and the agent skill expose them, plus the Editorial Board, and trending now serves all twelve networks.
Read the update →Every article in the RSS view now carries Bluesky, Hacker News, comments on the article, coverage across outlets and what the X and Facebook posts in our radar earned, next to Facebook, Reddit and Pinterest. Every chip opens the conversation it counts.
Read the update →Ten public pages that answer "when should I post on TikTok, Instagram, X, Facebook, Pinterest, Bluesky, Mastodon, Douyin, Tumblr, Reddit" from the publish times of the viral posts we track, recomputed daily, in four languages.
Read the update →The extension reads a profile through its grid when Instagram refuses the batched feed, says exactly what Instagram answered instead of stopping silently, and saves the outbound link of every tweet. Pinterest no longer opens on an empty week.
Read the update →Every Pinterest pin now lands under a category and a subcategory, the boards refresh daily, and each pin carries the date it was really created instead of the day we found it.
Read the update →The prices on the site are founding prices for the first customers and may rise later, but a plan you start at keeps its rate for as long as you keep it. And the questions buyers actually ask now have their own page.
Read the update →Mark a scheduled post as evergreen and pick how long to wait, from one week to two years. When the wait is up it shows on the dashboard and in the scheduler as due to remake, one click reopens it with the original alongside. And before you schedule anything, the composer tells you if the team already posted something similar.
Read the update →In 43 places the app printed the moment we saved a post as if it were the moment it was published. Every one now says which it is, and the Radar stops burying posts whose date we never learned.
Read the update →Tumblr sat in the menu with nothing behind it: the page searched Tumblr live and kept nothing. It now reads our own store, which grows every two hours and re-reads saved posts so their note counts move, and there is a public archive of the most-noted posts by year.
Read the update →Hacker News does not let you browse its past. We have been recording the front page every hour, so there is now a public archive ranked on the one number nobody else has: how fast a story climbed.
Read the update →Trending Mastodon posts from several instances land in ViralHunt every hour with no browser in the loop, keyed so a post seen on three servers is stored once. And every network on the public API now returns growth_24h.
Read the update →Viral posts from Douyin, the Chinese TikTok, land in ViralHunt with every title, creator name and hashtag translated. Its own page, the Radar, the dashboard, the API and English hashtag pages.
Read the update →Viral Bluesky posts land in ViralHunt automatically every hour: their own page, the Radar, the dashboard and the API. Hashtags, categories and growth tracking included.
Read the update →Re-checking a Reddit post now records what it had before, so every card can draw how it grew. You can sort and filter by measured growth instead of raw score.
Read the update →Free accounts hit clear, consistent limits across every network instead of quietly running out of results. Paid plans were checked to make sure nothing touches them.
Read the update →Every ranking on the public site now has a page explaining exactly what is counted on each network, and it reads the numbers from the code that enforces them.
Read the update →See which articles from a given publisher got shared most, and which outlets lead a topic, as real pages you can link to instead of a filter that disappears on reload.
Read the update →A batch of fixes to numbers and accounts that were wrong rather than missing, which is the worse kind, because a wrong number looks like an answer.
Read the update →Someone comments a keyword on your post, they get a DM. The ManyChat pattern, built into ViralHunt, with a report of exactly who was answered.
Read the update →An MCP server and a packaged agent skill, so Claude or any MCP client can read your trending content and work your editorial board directly.
Read the update →Per-post performance for everything you scheduled through ViralHunt, ranked by brand and by network, under Reports.
Read the update →The scheduler now works in brands, so an agency can keep clients apart, and it connects to 11 networks including Bluesky, Telegram and Google Business.
Read the update →Facebook scrambles the date on a post. We were falling back to the current time when we could not read it, which floated old posts to the top of every recent view.
Read the update →Those two counts were coming back empty. They are behind a login, and getting them needed the exact headers Pinterest's own site sends.
Read the update →Give it the topics you care about and it pulls new questions from each one, moving on when a topic stops producing anything new.
Read the update →Instagram, TikTok, X, Facebook, Pinterest and Quora are one installed extension now, which auto-updates. You no longer manage six.
Read the update →Instagram stopped answering our servers entirely. Capture now happens in your logged-in tab, which is both faster and far less likely to be blocked.
Read the update →Create an agent, give it a token, assign it cards from the editorial board like any teammate. It sees its own workload and comments back.
Read the update →The grid is good for browsing. When you are triaging a hundred posts, you want a table. Now there is one, and your choice is remembered.
Read the update →Out of 739 articles checked, the Facebook numbers were coming back for almost none of them. Two separate causes, both fixed.
Read the update →One merged list let the busiest network drown out the rest. Now each network gets its own section, so a quiet one is still visible.
Read the update →