Monir Newsroom

By Monir Tech Solutions

0+ active installationsVersion 1.4.3

AI newsroom for WordPress. Drafts and fact-checks news articles from your RSS sources — human approval before anything publishes. Not an autoblog.

Description

Monir Newsroom turns WordPress into a complete AI-powered newsroom. RSS sources are monitored for new stories, clustered by topic, drafted by an AI reporter, fact-checked against the originals, copy-edited, and held in a review queue for human approval before publishing. Editors get a clean Editor’s Desk surface for triage; admins get spending caps and provider-health visibility.

The plugin is built around two non-negotiable principles: nothing is published without passing the fact-check gate (by default an editor approves every draft; a beat can opt in to auto-publishing only the drafts the Fact-Checker approved), and every published article carries a clearly-labelled AI disclosure block (the badge cannot be hidden). Bring-your-own-API-keys economics — you pay AI providers directly at their published rates, and the plugin adds zero markup on usage. Multi-provider routing means a credit issue at one provider doesn’t stop your newsroom; the router automatically falls back to a healthy alternative.

Features

  • Multi-provider AI routing (Anthropic Claude, OpenAI GPT, Google Gemini)
  • Bring-your-own-API-keys — no plugin markup on AI usage
  • Beat Reporter agent for clustered RSS sources
  • Fact-Checker verifies every claim before draft is queued
  • Copy Editor polishes prose without changing factual content
  • Mandatory AI disclosure on every published article — schema.org markup + visible badge
  • Human-in-the-loop — every draft waits for approval at the Editor’s Desk unless the beat opts in to auto-publishing fact-checked drafts
  • Cost transparency — per-call audit, daily / monthly spending caps with 80% alert
  • Provider health monitoring — automatic skip on credit / auth errors, traffic-light status indicators
  • Configurable beats — Politics, Tech, Business, Sports, Entertainment, Science, Health + custom topics
  • Source attribution — every claim links back to its original source
  • Photo Editor — Pexels stock photos with a bundled placeholder fallback, attribution kept on every image

Safety architecture

Two architectural rules keep the editorial trust intact. First, the disclosure floor: every published article gets a visible “AI-assisted” badge plus schema.org metadata. The badge’s placement and styling are configurable; its presence is not. Second, the fact-check gate: drafts arrive at the Editor’s Desk and stay there until an editor approves and publishes, and nothing the Fact-Checker did not approve can be published without an editor’s explicit override. A beat may opt in to auto-publishing, and even then only approved drafts go out. These together mean readers always know what they’re reading and editors always get final say.

Monir Newsroom Pro

A separate add-on plugin, sold at https://monirtechsolutions.com/plugins/monir-newsroom/, adds translation into your publication language, AI-generated feature images (Google Imagen 3, OpenAI DALL-E 3), Wikimedia Commons portraits for stories about public figures, and the Insights page with cost analytics. Everything in this plugin works without it, including queueing approved drafts for the publishing times you set on a beat.

External services

AI providers (your own API keys; nothing is sent until you add one)

Draft generation, fact-checking and copy editing send the RSS source text, the draft and the prompt to the provider you configured at Newsroom → Settings → Providers. Each call is triggered by a beat run you scheduled, a manual draft, a fact-check, or a Test Connection click. Keys are stored encrypted on your site; the plugin adds no proxy and no markup, and you pay the provider directly.

  • Anthropic (api.anthropic.com) — terms: https://www.anthropic.com/legal/commercial-terms — privacy: https://www.anthropic.com/legal/privacy
  • OpenAI (api.openai.com) — terms: https://openai.com/policies/terms-of-use/ — privacy: https://openai.com/policies/privacy-policy/
  • Google Gemini API (generativelanguage.googleapis.com) — terms: https://ai.google.dev/gemini-api/terms — privacy: https://policies.google.com/privacy

Stock photos: Pexels (optional, your own API key)

When a beat’s image strategy includes stock photos and you have entered a Pexels key at Newsroom → Settings → Image Sources, the plugin sends the article’s search keywords to api.pexels.com and downloads the chosen photo into your media library with its attribution. Terms: https://www.pexels.com/terms-of-service/ — privacy: https://www.pexels.com/privacy-policy/

Starter news sources (optional, loaded only when you pick a beat)

The plugin ships a short list of public RSS feeds in data/starter-sources.json. Nothing is downloaded from any publisher until you pick a beat in the setup wizard (which loads only that beat’s feeds) or add feeds on the Sources page. After that, your own server polls each feed you leave enabled, on the schedule you set, to collect headlines and links; the AI agents read the articles your beat selects. You can disable or delete any feed at any time. Every publisher below links to its own terms of use and privacy policy; each feed, terms and privacy URL was last checked on 2026-11-20. Content you publish from these feeds is subject to the publisher’s terms.

World

  • BBC News — World — feed: https://feeds.bbci.co.uk/news/world/rss.xml — terms: https://www.bbc.co.uk/usingthebbc/terms-of-use/ — privacy: https://www.bbc.co.uk/usingthebbc/privacy-policy/
  • UN News — feed: https://news.un.org/feed/subscribe/en/news/all/rss.xml — terms: https://www.un.org/en/about-us/terms-of-use — privacy: https://www.un.org/en/about-us/privacy-notice
  • Al Jazeera English — feed: https://www.aljazeera.com/xml/rss/all.xml — terms: https://www.aljazeera.com/terms-and-conditions/ — privacy: https://privacy.aljazeera.net/
  • The Independent — World — feed: https://www.independent.co.uk/news/world/rss — terms: https://www.independent.co.uk/service/legal-terms-and-conditions-a6184181.html — privacy: https://www.independent.co.uk/service/privacy-policy-a6184186.html

Politics

  • BBC News — Politics — feed: https://feeds.bbci.co.uk/news/politics/rss.xml — terms: https://www.bbc.co.uk/usingthebbc/terms-of-use/ — privacy: https://www.bbc.co.uk/usingthebbc/privacy-policy/
  • The Independent — UK Politics — feed: https://www.independent.co.uk/news/uk/politics/rss — terms: https://www.independent.co.uk/service/legal-terms-and-conditions-a6184181.html — privacy: https://www.independent.co.uk/service/privacy-policy-a6184186.html
  • ProPublica — feed: https://www.propublica.org/feeds/propublica/main — terms: https://www.propublica.org/legal — privacy: https://www.propublica.org/legal

Business

  • BBC News — Business — feed: https://feeds.bbci.co.uk/news/business/rss.xml — terms: https://www.bbc.co.uk/usingthebbc/terms-of-use/ — privacy: https://www.bbc.co.uk/usingthebbc/privacy-policy/
  • CNBC — Business — feed: https://www.cnbc.com/id/10001147/device/rss/rss.html — terms: https://www.nbcuniversal.com/terms — privacy: https://www.nbcuniversal.com/privacy
  • MarketWatch — Top Stories — feed: https://feeds.content.dowjones.io/public/rss/mw_topstories — terms: https://www.dowjones.com/terms-of-use/ — privacy: https://www.dowjones.com/privacy-notice/
  • Financial Times — Home — feed: https://www.ft.com/rss/home — terms: https://help.ft.com/legal-privacy/terms-and-conditions/ — privacy: https://help.ft.com/legal-privacy/privacy-policy/

Tech

  • BBC News — Technology — feed: https://feeds.bbci.co.uk/news/technology/rss.xml — terms: https://www.bbc.co.uk/usingthebbc/terms-of-use/ — privacy: https://www.bbc.co.uk/usingthebbc/privacy-policy/
  • Wired — feed: https://www.wired.com/feed/rss — terms: https://www.condenast.com/user-agreement — privacy: https://www.condenast.com/privacy-policy
  • Ars Technica — feed: https://feeds.arstechnica.com/arstechnica/index — terms: https://www.condenast.com/user-agreement — privacy: https://www.condenast.com/privacy-policy
  • The Verge — feed: https://www.theverge.com/rss/index.xml — terms: https://www.voxmedia.com/legal/terms-of-use — privacy: https://www.voxmedia.com/legal/privacy-notice
  • TechCrunch — feed: https://techcrunch.com/feed/ — terms: https://techcrunch.com/terms-of-service/ — privacy: https://techcrunch.com/privacy-policy/
  • MIT Technology Review — feed: https://www.technologyreview.com/feed/ — terms: https://www.technologyreview.com/terms-of-service/ — privacy: https://www.technologyreview.com/privacy/

Science

  • BBC News — Science & Environment — feed: https://feeds.bbci.co.uk/news/science_and_environment/rss.xml — terms: https://www.bbc.co.uk/usingthebbc/terms-of-use/ — privacy: https://www.bbc.co.uk/usingthebbc/privacy-policy/
  • Nature — News — feed: https://www.nature.com/nature.rss — terms: https://www.nature.com/info/terms-and-conditions — privacy: https://www.nature.com/info/privacy
  • ScienceDaily — feed: https://www.sciencedaily.com/rss/all.xml — terms: https://www.sciencedaily.com/terms.htm — privacy: https://www.sciencedaily.com/privacy.htm
  • The Conversation (US) — feed: https://theconversation.com/us/articles.atom — terms: https://theconversation.com/us/terms-and-conditions — privacy: https://theconversation.com/us/privacy-policy

Health

  • BBC News — Health — feed: https://feeds.bbci.co.uk/news/health/rss.xml — terms: https://www.bbc.co.uk/usingthebbc/terms-of-use/ — privacy: https://www.bbc.co.uk/usingthebbc/privacy-policy/
  • STAT News — feed: https://www.statnews.com/feed/ — terms: https://www.statnews.com/terms/ — privacy: https://www.statnews.com/privacy/
  • WHO — News — feed: https://www.who.int/rss-feeds/news-english.xml — terms: https://www.who.int/about/policies/terms-of-use — privacy: https://www.who.int/about/policies/privacy
  • ScienceDaily — Health & Medicine — feed: https://www.sciencedaily.com/rss/health_medicine.xml — terms: https://www.sciencedaily.com/terms.htm — privacy: https://www.sciencedaily.com/privacy.htm

Sports

  • BBC Sport — feed: https://feeds.bbci.co.uk/sport/rss.xml — terms: https://www.bbc.co.uk/usingthebbc/terms-of-use/ — privacy: https://www.bbc.co.uk/usingthebbc/privacy-policy/
  • The Independent — Sport — feed: https://www.independent.co.uk/sport/rss — terms: https://www.independent.co.uk/service/legal-terms-and-conditions-a6184181.html — privacy: https://www.independent.co.uk/service/privacy-policy-a6184186.html
  • Sky Sports — News — feed: https://www.skysports.com/rss/12040 — terms: https://www.sky.com/help/articles/skycom-terms-and-conditions — privacy: https://www.sky.com/help/articles/sky-privacy-and-cookies-notice

Entertainment

  • BBC News — Entertainment & Arts — feed: https://feeds.bbci.co.uk/news/entertainment_and_arts/rss.xml — terms: https://www.bbc.co.uk/usingthebbc/terms-of-use/ — privacy: https://www.bbc.co.uk/usingthebbc/privacy-policy/
  • The Hollywood Reporter — feed: https://www.hollywoodreporter.com/feed/ — terms: https://pmc.com/terms-of-use/ — privacy: https://pmc.com/privacy-policy/
  • Variety — feed: https://variety.com/feed/ — terms: https://pmc.com/terms-of-use/ — privacy: https://pmc.com/privacy-policy/

Installation

  1. Upload the plugin folder to /wp-content/plugins/ or install through the WordPress plugin directory.
  2. Activate Monir Newsroom from the Plugins screen.
  3. Go to Newsroom → Settings → Providers and add your API key for at least one of: Anthropic, OpenAI, Google Gemini.
  4. Add RSS sources at Newsroom → Sources.
  5. Configure beats at Newsroom → Beats — choose preset beats or create custom ones.
  6. Drafts arrive in Newsroom → Editor’s Desk for human review and approval.

FAQ

Do I need to pay for AI usage separately?

Yes. Monir Newsroom uses a bring-your-own-API-key model — you obtain API keys directly from Anthropic, OpenAI, and/or Google, and you pay those providers at their published rates. The plugin adds zero markup. This keeps costs transparent and means your API spend is governed by the same billing dashboard you’d use for any direct API project.

Which AI providers are supported?

Anthropic Claude, OpenAI GPT, and Google Gemini. You can configure as many providers as you have keys for; the router falls back automatically when one is unavailable. Feature images come from Pexels stock photos (optional key) or a bundled placeholder.

Will my content be sent to AI providers?

Yes. The plugin sends RSS source content, generated drafts, and supporting context to whichever AI provider you configure. Choose providers you trust with your editorial pipeline. Provider settings include direct links to each company’s data-handling policies, and the External services section below lists every service with its terms and privacy policy. Self-hosted models are not currently supported.

Can it auto-publish?

Not by default. Every beat starts in draft mode: drafts stay as WordPress drafts until an editor clicks Approve and Publish at the Editor’s Desk. A beat can be switched to auto-publish, which publishes a draft the moment the Fact-Checker approves it; drafts with concerns or a rejected verdict always wait for an editor. A beat can also queue the drafts an editor approves for the publishing times set on the beat; approval is still the editor’s click.

What if my AI provider’s credits run out?

The Provider Health monitor catches credit and auth errors immediately. A “fast-fail” rule marks the provider as failed after a single occurrence (no point retrying when a 401 won’t fix itself), and the router cascades to the next available provider. The System Status panel shows which provider failed, the verbatim error message, and a direct link to the provider’s billing page. Beat dispatches that would have used the failed provider are skipped pre-flight rather than burning calls into a known-broken endpoint.

Can I cap my AI spending?

Yes. Settings → Routing exposes per-day and per-month spending caps in USD. When spend hits 80% of either cap an alert renders in the admin notice area; when it hits 100% the router stops dispatching new work. Caps reset on the day / month boundary. The Newsroom Dashboard shows today’s spend.

Can it translate articles?

Not on its own: drafts are written in the language of their sources. Translation into a publication language (Bengali, Hindi, Arabic, French, Spanish, German, Portuguese and more) is part of the Monir Newsroom Pro add-on.

Is the AI disclosure block customizable?

Placement and styling are customizable — you can choose where in the article the disclosure renders, edit the text, and apply your theme’s CSS. The disclosure itself cannot be removed. This is a deliberate architectural rule, not a setting toggle. Schema.org metadata for AI authorship is also always present.

What happens if I deactivate the plugin?

Articles that have already been published stay published with their disclosure blocks intact. Draft articles in the Editor’s Desk queue stay as drafts and remain accessible if you reactivate. If you uninstall (delete the plugin), your data is preserved by default — beats, sources, configuration, wire items, pipeline drafts, agent-run history, and per-article editorial metadata (fact-check status, image attribution, AI disclosure tags) all survive a delete-and-reinstall cycle. To explicitly wipe data on uninstall, enable “Delete plugin data on uninstall” at Newsroom → Settings → System Status (Danger zone) BEFORE removing the plugin. Published article post content is always preserved.

Where do I report bugs?

Support: https://monirtechsolutions.com/plugins/monir-newsroom/ — include your WordPress version, PHP version, plugin version (Newsroom → Settings → System Status), and a minimal reproduction.

Screenshots

Changelog

1.4.3 — 2026-11-11

Background runs that stopped part-way are fixed. No new settings are required; one is added.

  • Every background run keeps to a time budget. A source poll, a beat dispatch and each pipeline step now work for at most the run budget (default 50 seconds, set on Settings → System Status, 30 to 180) and hand what is left to a follow-up action a minute later. No single AI call runs longer than 45 seconds. On a site where the web server, not PHP, ended long requests, the Fact Checker and Copy Editor never ran after a draft; they now run as their own scheduled actions and finish.
  • Each step after the draft is its own scheduled action. Fact-check, copy-edit, image and post creation are separate actions, and each queues the next itself. Before it starts, a step also queues its own retry; a step the host ends is retried by it (attempt 2, then 3 of 3) and then marked failed with the reason “did not finish”, with a Retry step button. The Pipeline page shows which step is next and when it is due, and has a Retry step button for any failed step.
  • Relevance verdicts are kept. A cluster a beat has judged is not sent to the model again unless it has gained items, so a continuation picks up where the last run stopped without repeating its calls. The Beats page shows “running now: checking cluster 3 of 12” while a dispatch works and “In progress” when a run has handed the rest to a continuation.
  • Every AI call is logged when it goes out, not only when it returns. A call whose process was ended before the provider answered shows as cut off on the Activity page instead of leaving no trace, and System Status counts cut-off calls over the last day.
  • The draft call is sized to the beat’s target length. The Reporter asks for a limit sized to the beat’s target length instead of a fixed 4,096.
  • System Status shows the web server and PHP SAPI, and on LiteSpeed explains that the server’s connection timeout can end a background run, with the .htaccess exemption LiteSpeed documents. The plugin never writes .htaccess.
  • Source polling and clustering fit the budget too. Feeds are polled least-recently-polled first with a 10-second timeout each, and clustering runs in batches of 50; whatever is left waits for the next run.
  • The Pipeline page prints a dash instead of “0000-00-00 ago” for a draft without a timestamp.
  • The AI disclosure label is in the post’s language. A translated post showed Bengali disclosure text behind an English “AI disclosure:” label. The label now comes from the same per-language strings as the text (Bengali, Hindi, Urdu, Spanish, French, Arabic); an English post is unchanged. The Bengali disclosure text and label were reviewed by a native speaker.
  • The dek no longer repeats the first paragraph. The Reporter is told that the line under the headline is a different sentence from the lede, and a dek that still restates the first paragraph is left out of the body (the excerpt keeps it).
  • The Activity page’s live view finishes its running rows. A call shown as “Running” stayed that way until the page was reloaded. Each poll now asks after the rows it still shows as running and replaces each one that has finished with its OK, error or cut-off row.
  • Beats draft again. In the first 1.4.3 build the query that picks a beat’s clusters did not run on MySQL, so no beat made a relevance call or drafted and every run read “No fresh clusters”. The query is fixed, a cluster the beat has not judged yet is always considered, and a run that finds nothing shows Idle (or Active when the beat drafted in the last day) instead of Failing.
  • Drafts are not cut short. A draft may now use about five tokens a word of its target length (3,000 for 500 words), so the article is no longer cut off. A draft that would not fit in what is left of a run waits for the next run, which starts with it, and an answer that hits the output limit is reported as cut off rather than as unreadable. Beat cards count drafts waiting for review, so a beat that has drafted shows Active or Idle, not “Wire warming up”.
  • Drafts become posts again. In 1.4.3 as first built, Action Scheduler reported each step’s own action as still scheduled, so no step could queue the next: drafts moved one step per ten-minute sweep, used up their retries on steps that had finished, and none became a post. Each step now queues the next one itself, and drafts left stranded resume on the next source-monitor run from attempt 1.
  • The image step stays inside the run budget. Every outbound request a pipeline step makes now ends 5 seconds before the run budget does; the chosen image is downloaded with a 15-second timeout instead of WordPress’s 300; a photo search that would start with too little time left continues in a fresh run; and an image that cannot be downloaded falls back to the placeholder.
  • The Image Prompt Writer runs only when its answer is used: an AI image provider with its key, a Pexels key, or the add-on’s photo sources (Wikimedia Commons in Pro). A beat on the bundled placeholder makes no call.
  • Beat cards after an update. A beat whose last run failed on an earlier build of the plugin reads “Waiting for next run” until it runs on this one, and drafts still in the pipeline count as drafts. With nothing in flight the live pill always shows the card’s label. The Activity live view asks for a fresh answer on every poll, and its script is versioned by file, so a cached copy from an earlier build is not kept.
  • Translation calls cost less. The Anthropic provider can cache the unchanging start of a system prompt (5-minute prompt caching); an add-on’s translation calls use it, and the recorded cost prices cache writes and reads.
  • Word counts in any script. The Editor’s Desk counted only Latin letters, so a Bengali post showed a handful of words; it now counts the words of any language.
  • The placeholder image’s caption and alt text are in the post’s language. A Bengali post with the bundled placeholder showed the English caption “Stock illustration”; it now reads in the post’s language, as the AI disclosure does.
  • Paused beats are listed. The Beats page showed only active beats in plain view; paused beats sat in a closed fold under the standard beats, which read “All 8 standard beats are configured”. Paused beats now have their own section after the active ones, each with Resume and Edit.
  • Every source is listed. The Sources page showed the first 25 sources and no way to reach the rest, so a site with more could not see, enable or delete them. It now lists every source, grouped by beat, with a count of sources, enabled and disabled.
  • A source or beat says who turned it off, and why. Every change of a source’s Enable/Disable or a beat’s Pause/Resume, by any button, the beat form or an upgrade routine, is noted on the row (“Disabled 2026-11-16 03:30 UTC by admin: with the Disable button on the Sources page.”) and listed on the Activity page. Nothing in the plugin turns a source off or pauses a beat on its own. Adds two columns to the sources and beats tables.
  • A feed can be added once. Adding a feed URL that is already a source (on the Sources page, a beat’s form or the setup wizard) is refused with a message naming the source that has it. The check on a beat’s form and in the wizard never matched before, so a feed could be added twice.
  • A run counts the time its request already spent. When Action Scheduler runs several actions in one request, the web server’s timeout counts them all; a run that starts late in a request now has that much less budget, so it hands its work to a follow-up action instead of being cut off.
  • The Activity page shows prompt-cache tokens. Under each call’s input tokens: how many were read from the provider’s prompt cache (billed at a tenth of the input price) and how many were written to it (billed at 1.25 times). Adds two columns to the agent-runs table.
  • Editor’s Desk no longer shows “model output needed repair (code_fence)”. A note that an AI answer needed a small formatting fix is collapsed under Details and worded for a reviewer.
  • A draft can be held for an editor. An add-on that finds something a human must look at (Monir Newsroom Pro’s translation check, when a critical mismatch remains) holds the draft: auto-publish and scheduled slots wait, the Editor’s Desk card shows a ⚠ badge, and approving the draft by hand releases it.

1.4.2 — 2026-11-09

Changes asked for by the wordpress.org plugin review of 1.4.1. No new features.

  • Requires WordPress 6.9 and PHP 8.2. The bundled Action Scheduler is updated from 3.9.3 to 4.2.0, which needs WordPress 6.9. The PHP floor rises from 8.1 to 8.2.
  • Every database query is prepared. All reads and writes go through $wpdb->prepare(), with table and column names passed as identifiers; nothing is interpolated into SQL. Behaviour is unchanged.
  • Input is sanitized where it is read; output is escaped where it is printed. Filter and form values are read one key at a time, sanitized and checked against an allow-list; status badges, the setup-wizard byline, the first-draft e-mail and the JSON-LD disclosure are escaped at the point of output. Badge labels are now translatable.
  • Starter sources ship as JSON. data/starter-sources.opml is replaced by data/starter-sources.json with the same feeds, terms and privacy links. Medical Xpress is removed from the Health starter set.
  • Starter feeds that did not answer reliably are gone. Removed NPR, CBC and ESPN starter feeds whose links did not respond reliably; added WHO — News and ScienceDaily — Health & Medicine to the Health beat so every beat still has feeds from at least three publishers.
  • Only plain file types in the download. The plugin ZIP contains php, js, css, txt, md, json, xml, png, svg, jpg and pot files only.
  • A declined cluster is not a failure. When the Reporter declines to draft a cluster because its sources are too thin, the beat card now says Active with “no draft (cluster declined)”, the Recent activity row says the same, and the Dashboard no longer announces a failing beat. Failing is reserved for real errors.
  • Plugin version on every Newsroom page. A small version badge sits next to each page title and the admin footer reads “Monir Newsroom 1.4.2” on the plugin’s own screens.
  • Pro line corrected. The Settings footer no longer lists scheduled publishing as a Pro feature; scheduling has been free since 1.4.1.

1.4.1 — 2026-11-04

Fixes from the first live test of 1.4.0 on a 37-feed site.

  • Picking a beat loads only that beat’s starter sources. Choosing World in the setup wizard loads the six World feeds and moves on; it no longer loads all 38 feeds and then asks you to untick the sports and entertainment ones. A custom beat loads the starter feeds that match your description plus any feed URLs you add, and the wizard warns when they cover fewer than three publishers (a story is only drafted when three different outlets cover it). The separate “Load starter sources” button and the source picker are gone; edit feeds on the Sources page.
  • Drafts no longer merge unrelated stories. Three different UN News items were being written up as one article. Clustering ignores boilerplate lead-ins such as “The UN …”, needs at least two shared entities before they count, and the three-source rule now counts distinct publishers (BBC feeds on bbc.co.uk, bbci.co.uk and bbc.com are one publisher), not three items from one feed. The drafting prompt tells the model to write only the event most of the reports share and never to combine separate events.
  • Article bodies are fetched when a draft is written, not at ingest. 1.4.0 queued one background job per wire item (930 on the test site). Bodies are now fetched for the few items in a cluster right before drafting, 8 seconds each, falling back to the feed summary; the leftover queued jobs from 1.4.0 are cancelled on update.
  • “Next run” is right while a beat is running. The Beats page said “not scheduled” and the Sources page said “in 57 years” while the source monitor was mid-run; the three pages now share one helper that reports running, due, scheduled or none.
  • Editor’s Desk fallback note. A repaired model response (a JSON code fence stripped) was reported as “used a fallback provider — primary was ( )”. Provider fallbacks and JSON repairs are now listed separately, on the Editor’s Desk, the draft cost panel and System Status.
  • API key mask. Saved keys showed one dot per character plus the last four (a 108-character value ending in “AA”); they now show a fixed-width mask such as sk-ant-••••QAAA.
  • Setup wizard. Dispatching the first run completes the wizard, so other admin pages stop redirecting back to it and nobody is told they “skipped setup” after finishing; the Skip link is hidden on the summary step; the agent list on step 5 no longer renders “1. 1”.
  • Beat card menu. The ⋮ menu on a bottom-row beat card opened off the page; it opens upward.
  • Placeholder feature image. The bundled placeholder was a 1×1 pixel; it is now a neutral 1200×675 image. One copy is added to the Media Library and shared by every article that falls back to it, instead of a copy per article. On update, articles still pointing at a 1×1 copy are switched to the shared image and the 1×1 copies are deleted.
  • Database. wire_items gains a source_domain column with an index, filled on update in batches; no data is removed.
  • Related reports now cluster. Six reports of one sports retirement from three outlets were written up as six separate drafts: each feed summary named the player once, and the headlines were editorial (“A bitter end for…”, “Six tournaments, 11 goals…”). A name that the headline also carries now counts even when the summary mentions it once, and two reports are one story when the same person or organisation is the subject of both headlines and the texts share three or more topic words. Unrelated items that merely mention the same name (a club match report about the same player) stay apart, and the boilerplate merges fixed above stay fixed.
  • Setup wizard remembers the beat you picked. Steps 2 to 6 all describe that beat: the source count on step 2 is that beat’s, step 3 shows the connected provider with a masked key (change it behind “Use a different provider or key”), and step 6 waits for the first draft of the run it started, not an older draft. A custom beat created in the wizard keeps exactly the sources it matched plus the feeds you added.
  • Scheduled publishing works without the add-on. A beat set to “queue at scheduled times” now queues each approved draft for its next publishing slot; a beat with no slots leaves the draft as a draft and says so.
  • Routing tab. Lists the four agents that make AI calls (Reporter, Fact Checker, Copy Editor, Image Prompt Writer) by name instead of ten slugs, six of which never call a provider.
  • Activity page. The feed starts with the latest runs instead of the oldest fifty, the status filter is applied by the server, internal bookkeeping rows and connection tests are hidden behind a “Show internal steps” checkbox, agents are shown by name, and times are in your site’s timezone with the zone named once in the column header.
  • Pipeline page. Describes each step in plain words (Completed, Held for fact-check, Failed, In progress, Stuck); add-ons can add their own row actions and labels.
  • Fewer paid calls per draft. The Image Prompt Writer no longer asks the model for an AI illustration prompt when nothing on the site can use one (no AI image provider; or a beat set to placeholder only, or with no Pexels key).
  • Dates and times. Every absolute time in the admin uses one format with the timezone named; the Dashboard’s draft list no longer shows 1970 for unpublished drafts.
  • Wording. “(free tier)” on the Usage page is “(no-cost tier)”; settings that an add-on can extend say “when an add-on provides one”; leftover version notices (v0.3.0, v0.6.1, v1.2.0) and the one-time Routing redesign notice are gone.
  • Routing never defaults to the most expensive model. Saving the Routing tab in Manual mode keeps the choices for agents the tab does not list, and no default or Auto quality below Thorough picks an Opus model: Balanced is Sonnet for drafting and fact-checking and Haiku for everything else.
  • Every outlet reaches the Reporter. The five items a draft is written from are taken one per outlet first (by publisher domain), so a story covered by three outlets cites three, even when one outlet published five times.
  • No duplicate items at ingest. A feed that republishes a story with an edited summary no longer stores it again: the feed’s own id is kept, and the same URL, or the same headline within 48 hours, is skipped.
  • Editor’s Desk shows “7 reports · 3 outlets” on a cluster instead of “7 sources”.
  • Activity agent filter has one entry per agent (the two reporter names are one “Reporter”).

1.4.0 — 2026-10-27

  • Global prefix changed: every function, class, constant, hook, option, transient, post/user meta key, database table, AJAX/REST name, script and style handle, admin menu slug and the WP-CLI command now use the monirnr_ / MONIRNR_ / monirnr- prefix instead of mnr_ / MNR_ / mnr- (WordPress.org requires a distinctive prefix of at least four characters). The REST namespace is now monirnr/v1 and the WP-CLI command is wp monirnr. CSS class names and DOM ids are unchanged.
  • Free and Pro are two plugins: translation, AI-generated feature images (Imagen 3, DALL-E 3), Wikimedia Commons photos, the Insights page and scheduled publishing moved to the separate Monir Newsroom Pro add-on. (Scheduled publishing returned to the free plugin in 1.4.1.) This plugin no longer bundles the Freemius SDK, has no licence tab, no free-tier caps on beats, sources or articles, and no locked settings: every setting it shows is yours to use. Uninstall runs through uninstall.php again. A beat in the “queue at scheduled times” output mode keeps approved drafts as drafts, with a notice, until the add-on is active to schedule them.
  • No data migration from 1.3.x: tables, options and meta written by 1.3.x are not read by 1.4.0 and are not migrated. Updating an existing install behaves like a fresh install: run the setup wizard again and re-enter API keys. 1.4.0 only touches monirnr_ data, so leftover mnr_ tables and options from 1.3.x are not removed; delete the 1.3.x plugin with “Delete plugin data on uninstall” enabled before installing 1.4.0 if you want them gone.

1.3.1 — 2026-06-07

  • Freemius platform requirement: migrated uninstall logic from standard uninstall.php to Freemius after_uninstall hook. Freemius needs to intercept the uninstall lifecycle for analytics + license cleanup; standard uninstall.php bypassed the SDK initialization.
  • No functional change to uninstall behavior — same tables dropped, same options cleared, same default-preserve gate via monirnr_delete_on_uninstall option.

1.3.0 — 2026-06-02

  • Plugin Check cleanup: Resolved all 165 production-code errors across 5 cleanup sessions (NonceVerification, SQL, EscapeOutput, i18n, Resource Offloading) to align with WordPress.org plugin submission requirements.
  • Chart.js bundled locally: Chart.js v4.4.0 now ships with the plugin at assets/vendor/chartjs/ instead of loading from jsDelivr CDN. Removes external resource dependency per WP.org plugin guidelines. Cost Dashboard rendering unchanged.
  • i18n improvements: Added translator context comments to 24 printf/_n() sites; refactored 2 sites for proper xgettext extraction.
  • Security hardening: Wrapped exception messages in esc_html() to prevent unescaped output in error contexts.
  • Internal code quality: 230 phpcs annotations added across canonical WordPress patterns (table-name interpolation, custom-table queries, helper-escaped output). Zero behavior changes.
  • Bug fix: Repaired 10 phpcs:ignore directives that used Unicode em-dash separator (Plugin Check silently failed to recognize them).

1.2.0

  • New. Custom beats on Free tier. Wizard Step 2 now leads with a description textarea so operators can describe a custom focus (“Electric vehicle industry news”, “Boston restaurant openings”, etc.) and the source-matcher picks curated feeds that fit. The 8 standard beats (Politics, Tech, Sports, etc.) demote to an expander beneath. Free tier allows 2 active beats (mix custom + standard); Pro removes the cap.
  • New. Curated 144-source wire library. Sources span Bangladesh, India, US, EU, and global newsrooms across politics, tech, business, sports, science, health, and finance. Source Monitor polls them on the v0.6.0 cron cadence; the rare-token IDF-lite matcher introduced in this release routes them to the right beat based on the description.
  • New. Per-beat publication language column on wp_monirnr_beats. On Free, beats are silently locked to en (Translator is a Pro feature). On Pro, the column tracks the operator’s per-beat language choice; the per-beat picker UI lands in v1.3. Pre-v1.2.0 the site-wide monirnr_publication_language option could carry a stale value across plugin resets, producing English bodies with non-English image captions — fixed by snapshotting the gated value at draft-create time and reading it back in Photo Editor’s attribution path.
  • New. Wizard Step 3 hard gate on Connect AI. The Save & continue button starts disabled and only enables after a successful Test Connection for the currently-entered key. Changing the provider or editing the key re-disables Save until a fresh test passes. Backend mirror gate at MONIRNR_Setup_Wizard::handle_save_provider rejects POSTs that lack a test_connection_<provider> success row within the last 30 minutes; failed bypass attempts redirect back to Step 3 with monirnr_wizard_notice=test_required. Wizard test handler now records the test to wp_monirnr_agent_runs so the gate has an audit trail.
  • New. Defensive Sources-block guard in MONIRNR_Post_Creator. When the Beat Reporter’s draft response truncates mid-JSON (Gemini 2.5 Flash counts hidden reasoning tokens against maxOutputTokens, eating the visible budget before the sources field is written), the wp_post_content was previously saved without a Sources block at all despite monirnr_source_urls postmeta carrying valid URLs. Post Creator now detects the absence and synthesises a Sources block from the postmeta URLs using MONIRNR_Source_Names::for_url() for friendly outlet labels. Logged to wp_monirnr_agent_runs with agent_name='post_creator' and event marker defensive_sources_block_appended so operators can monitor frequency.
  • New. Beats list status pill recognizes more terminal states. waiting_for_sources and no_matches (recent_drafts=0) render as “Wire warming up” amber; skipped_provider renders as “Provider issue” deep red with a 3px left-border accent. failed + recent_drafts>0 de-escalates to “Active (intermittent)” amber. The rollup uses a 24h wp_posts recent-drafts count to distinguish “terminal failure” from “intermittent run inside a productive day.”
  • Fixed. Source attribution rendered “Com” (or “Co”, “Net”) as the outlet name in the article-foot Sources <ul> for any 2-segment-TLD host (.com.au, .co.uk, .com.bd, .com.br, .co.in). Root cause: MONIRNR_Source_Names::for_url() walked the subdomain chain by mutating $parts and the SLD fallback read $parts[0] after the mutation, landing on the TLD fragment. Fix preserves the original parts array; SLD is read from the unmutated leftmost segment. Added reneweconomy.com.au, theage.com.au, smh.com.au, abc.net.au to the KNOWN outlet map while fixing the fallback.
  • Fixed. wp_monirnr_beats.last_run_at timezone-unification. The v0.13.0 writer flip to LOCAL had drifted out of sync with every reader (REST API explicitly appended ' UTC'; admin views relied on PHP’s default UTC; agent_runs.created_at was always UTC). Operators saw “X hours ago” deltas off by gmt_offset on the Beats card, Dashboard widget, and Pipeline page. Writer flipped back to UTC (current_time('mysql', true)); 3 admin-view readers now append ' UTC' so strtotime() is unambiguous; back-stamp migration maybe_run_v120_tz_unification_migration runs once on upgrade to convert pre-flip LOCAL rows to UTC (DATE_SUB by gmt_offset minutes; supports fractional zones).
  • Fixed. Free-tier cap visible on the Beats list’s Standard beats grid. Pre-v1.2.0 every Standard beat card rendered an “Enable” button regardless of cap state; only the server-side handler rejected the click with a notice. Now the grid computes is_feature_enabled('beats_unlimited') + get_active_beats_count() and renders a Pro-lock pill + disabled button when at the 2-beat Free cap. Counter above the grid reads “X / Y active beats — Pro removes the cap.”
  • Fixed. REST /beat-status/{id} no longer flags waiting_for_sources and safety_skip as failed:true. Both are informational non-failure states; the pre-v1.2.0 classifier put anything outside success + no_matches into the failure bucket, which short-circuited the pipeline-visibility JS to “Failed” red regardless of the labelmap entry. Now classified as non-failure; pill renders “Wire warming up” / “Skipping” correctly.
  • Fixed. Wizard Step 5 → Step 6 UX streamline. Pre-v1.2.0 the “Start beat now” submit redirected back to Step 5’s post-dispatch state card with four buttons (Continue to summary / Run again now / Update preferences / View Editor’s Desk + Continue) over the same “wire warming up” copy Step 6 already rendered. Customers stalled. Now redirects directly to Step 6 which presents the clean three-button set (Editor’s Desk / Adjust beat / Finish). Step 5’s post-dispatch render is preserved as dead-code fallback for v1.2.1 cleanup.
  • Fixed. Soft-reset tooling clears all provider state. The c:\tmp\v120-soft-reset-v2.php operator tool previously preserved API keys + wizard pick across resets to save re-entry, but a stale invalid Anthropic key carried between cycles concealed the test_connection bypass that Bug 13 fixed. Now clears every monirnr_*_api_key, monirnr_default_provider, monirnr_routing_mode, plus LIKE-pattern sweeps for monirnr_provider_% / monirnr_routing_% / monirnr_route_% / monirnr_agent_provider_% / monirnr_agent_model_%. Truncates wp_monirnr_provider_health. Tooling fix, not plugin code.
  • Known limitation. Beat Reporter draft calls to Gemini 2.5 Flash may truncate mid-JSON on token-heavy clusters because Gemini 2.5 counts hidden reasoning tokens against maxOutputTokens. Affected drafts are skipped automatically (the cluster fails with malformed_draft_json and Beat Reporter moves on); the defensive Sources-block guard above recovers attribution when the body is salvageable but the sources field was cut off. A focused fix (provider thinkingConfig parameter, max-tokens bump to 8192+, or prompt restructure) is scheduled for v1.2.1.
  • Known limitation. Custom beats covering niche topics may take 24-72 hours to produce their first draft on fresh installs, depending on wire content. The Section 8.1 safety floor requires min_cluster_size distinct sources (default 3) reporting the same story before drafting. Operators can lower the threshold per beat via Settings → Beats → Edit, or switch the beat from source_filter_mode=auto to curated source picks for faster matching.
  • Verified: tools/verify-v1-2-0.php — 294+ scenarios A-AA + nested v1.1.0 chain green.

1.1.0

Provider/tier coherence release. On Free, the wizard-chosen AI provider becomes authoritative — no silent fallback to a different paid provider when the chosen key fails. New Pipeline-page recovery card surfaces beats stuck at provider resolution. Includes one-time activator migrations to widen last_run_status and re-encrypt any plaintext API keys.

1.0.5

Hotfix: New AI Draft page now resolves any active provider instead of being hardcoded to Anthropic. Display-only fix; draft generation was already provider-agnostic.

1.0.4

Wizard marquee redesign: Step 2 consolidates source loading + beat picking; Step 4 shows a live pipeline stage list; Step 5 ends on the actual produced article preview. One-time step-state migration for mid-wizard users.

1.0.3

Display-string fix: Beats page no longer renders the misleading “Image source: Global (None)” label on Free first-runs.

1.0.2

Foundation hotfix: Free-tier first-run articles now sideload the bundled placeholder image. Optional Pexels API key slot added to wizard Step 3.

1.0.1

Hotfix: two routing bugs that caused beats to silently fail on fresh installs — the v0.17.1 free-tier routing override now applies to both preflight and dispatch paths, and the wizard’s API-key save now goes through the canonical encrypt-and-persist path so provider classes can decrypt. New Test Connection button on wizard Step 4.

1.0.0

First public release. WordPress.org submission preparation: refreshed tagline, placeholder plugin-page assets, submission documentation, GPL-compatibility audit, and Plugin Check pass. Minimum required: WordPress 6.5, PHP 8.1.

0.16.0 – 0.17.1

  • Wizard Done-state detection + Free-tier routing honors wizard pick + Action Scheduler self-heal heartbeat (v0.17.0 – v0.17.1).
  • Newsroom Dashboard rebuild + Insights page (Cost Dashboard + Usage merged) + fast-fail health escalation (v0.15.14 forward).
  • Provider health monitoring + status panel relocation + verbatim API error display + billing-page action buttons (v0.16.12 – v0.16.13).
  • Submission-ready readme.txt + .pot translations + screenshot references (v0.16.2).
  • Build script for distributing free and premium variants from one codebase + license infrastructure groundwork (v0.16.0 – v0.16.1).

0.10.0 – 0.15.13

  • Cost Dashboard with date-range filters, charts, breakdowns, and CSV export (v0.15.0 – v0.15.5).
  • Stock photo cascade with Pexels (v0.15.6 – v0.15.7).
  • AI image generation: Image Prompt Writer agent + Photo Editor cascade with Gemini Imagen 3 / DALL-E 3 + per-beat AI provider override (v0.15.8 – v0.15.10).
  • Routing tab redesigned around explicit Auto/Manual mode (v0.15.11).
  • Editor’s Desk UI for human approval workflow (v0.14.0).
  • Translator agent with publication-language support (v0.12.0).
  • Beat Reporter agent + clustering + Copy Editor agent (v0.7.0 – v0.10.0).