What ErmisAI does
How ErmisAI turns monitored international coverage into localized drafts your editors approve before they reach your CMS.
ErmisAI watches international news sources, groups the articles that cover the same event into one story, and writes a source story in your newsroom's article language. Your desk then composes the article, reviews it, and approves it.
Nothing leaves the workspace on its own. Every story that reaches your CMS was written because a person ran compose, and released because a person pressed Approve.
These pages describe what is on screen today. If a capability is limited to one role, gated by your plan, or not built, it says so in the same breath.
The chain, end to end
ErmisAI collects
ErmisAI fetches the feeds your newsroom watches. You pick those during setup — from a catalog of 1,427 international sources, or by adding your own RSS feed. See Choosing and adding sources.
Collection runs on a schedule of roughly ten minutes, so your feed keeps moving whether or not anyone has the tab open. Refresh sources on the feed forces a fresh pass for your newsroom.
ErmisAI groups overlapping coverage
Articles that look like the same event are grouped into a single story. Grouping is done by comparing headlines and article text, not by asking a model to decide which articles belong together. A single story therefore stands on several outlets, and the number of distinct sources behind it is the largest input to its confidence score.
ErmisAI writes a source story
Each group is written up as one source story in your newsroom's configured article language, translated from the source material where needed. Output language is a setting. It is never inherited from whatever language the sources happened to be in.
Alongside the text, the story carries a confidence score, a category, the list of outlets behind it, any conflicts found between them, and a sensitive flag.
The story lands on your feed as an empty draft
This is the step people expect to be different from what it is. A new story arrives with no article text. The card shows the source headline, the confidence signal, and the line "Needs first article in english" — the language name is inserted in lower case and is your newsroom's article language, not always English. Below that: "The source story is available, but no article in english has been saved yet." Its pipeline badge reads "Ready to compose".
There is nothing to edit yet, because nothing has been written yet. What you are looking at is the source story, not a draft of your article.
A person composes the article
Open the story and use the compose action — "Compose the first article" on a story that has none, or "Generate article" once you are in compose mode. This is the only way an article comes into existence. The refinement chat is blocked until compose has run once, because there is no saved article for it to refine.
After the first article is saved, the same workspace handles rewrites, tighter leads, headline options, and cleanup. See Composing and refining a story.
The author submits it
"Submit for review" moves the story from Draft to In review. It is refused if the article is still the untouched placeholder. While a story is in review its text is locked on the author's side.
An editor decides
Owners and admins open Editorial and work the board. Four decisions: Save, Approve, Reject, Send back to compose. Approve is what makes a story deliverable. Reject makes the text editable again so the author can revise and resubmit. Send back to compose returns the story to Draft so it can be regenerated.
If you are the only active member of your newsroom, you review and approve your own work. That is expected, and the workspace says so. See Editorial review and approval.
Approved stories go out
Approved stories are released to your in-app feed and become available to whichever delivery surfaces your plan includes. Releases are capped per UTC day.
What ErmisAI will not do
These boundaries are drawn from the product's own published AI policy and security pages.
| Boundary | What it means for you |
|---|---|
| Not an autopublisher | There is no auto-publish path, no confidence threshold that publishes, and no "publish on approval" integration that pushes into WordPress. ErmisAI never writes into your CMS; the WordPress connector pulls. |
| Not a wire | ErmisAI does not deliver source material. It groups coverage, drafts from it, and leaves the publishing decision with your desk. |
| Not a CMS | Approved stories export to supported surfaces. Your CMS remains your CMS. |
| Not a rights-clearing system | A subscription is software, not a content licence. Licensing, syndication, and attribution terms are separate agreements. |
| Not image, audio, or video generation | Text only. There is no media generation anywhere in the product. |
| Not a guarantee of correctness | The security page is blunt about this, under "What we do not claim": "The product generates AI articles. These may contain factual errors, misattributions, or incomplete coverage. Editorial review before publication is a design assumption, not an optional step." |
Where you work
Four items in the sidebar. What you see depends on your role, not on your job title.
| Screen | What it is for | Who can open it |
|---|---|---|
| Feed | Triage. Every story your newsroom has, with its signal, its state, and whether an article exists yet. Read more | Everyone |
| Editorial | The Editorial board and the Editorial review screen, where stories are approved, rejected, or sent back. Read more | Owners and admins |
| Monitoring | Alert rules, monitored source health, and delivery channels. Read more | Owners and admins |
| Settings | Newsroom, Team, Billing, Usage, Integrations, Notifications, Account. Read more | Everyone sees Notifications and Account; the rest is gated by role and plan |
There are exactly three roles: owner, admin, and member. Titles such as Editor-in-Chief or Reporter are free text you type about yourself — they grant nothing. Billing and Integrations are owner-only; an admin cannot open either. The full matrix is in Team, roles and permissions.
Two status systems, side by side
Every story carries two independent statuses, and they can disagree without anything being wrong.
Pipeline describes how far ErmisAI got with the source material: Collecting, Clustering, Ready to compose, Review, Published. In practice, once an article has been composed the pipeline badge follows the editorial outcome — Review after submit or reject, Published after a human approves. An uncomposed story always shows Ready to compose. The pipeline's own high-confidence "published" signal is internal and never reaches a card, so a Published badge on a card always means a person pressed Approve — nothing is auto-published. A new draft always materialises in editorial Draft with placeholder text until someone composes it.
Editorial describes where your desk is with it: Draft, In review, Rejected, Published. Editorial Published is only ever set by a human pressing Approve, and it is terminal.
A card reading "Ready to compose" next to "In review" is legitimate: the source story came in flagged for review before anyone wrote anything.
Alongside the two states sits the signal score — a 40–99 number with one of four labels: Verified signal, Strong signal, Developing signal, Low signal. It is a heuristic built from how many distinct sources carry the story, how those sources are tiered, how closely their headlines agree, and how fresh the newest article is. It gates nothing — no delivery, no plan feature, no review requirement. It is a hint about how much scrutiny a draft deserves. See Confidence, sources and conflicts.
A story can also arrive marked sensitive. That raises its editorial priority and shows a Sensitive badge on the review surfaces. It does not add an approver or block delivery — see Sensitive coverage and editorial limits.
Getting approved stories out
Four delivery surfaces work today:
- In-app feed — every plan. Approved stories are released into the feed and counted against a daily cap that resets at 00:00 UTC: 20 on Free, 100 on Plus and Business Plus, 500 on Pro and Business Pro, unlimited on Enterprise. Over-cap stories are not shown at all; they appear only as a Queued count and an amber banner.
- Outbound webhooks — Plus and above. Signed payloads sent to an endpoint you configure.
- CMS export API — Pro, Business Pro, and Enterprise. Your system pulls approved stories with a tenant API key.
- WordPress pull API — Pro, Business Pro, and Enterprise. The same export, reached through the WordPress connector.
Email delivery is named in the monitoring screen but is not implemented; digest-mode alert rules deliver in-app instead. External pulls draw from the same released pool as the feed, so a busy polling loop consumes your daily cap. See Getting approved stories out and API keys, webhooks and WordPress.
Languages
ErmisAI has two independent language settings.
Interface language is the language of the product itself — ten locales: English, Greek, Polish, Italian, Spanish, Portuguese, Swedish, Danish, Norwegian, and Finnish.
Article language is what your stories are written in. It is a newsroom setting. Source material is translated into it regardless of what language the sources used, and it is never inferred from the sources.
A handful of labels stay English in every locale — the four signal descriptors, the breaking pill, and the notifications the review workflow generates. See Interface language and article language.
Plans and limits
Plans are priced in EUR per month, from Free at €0 to Business Pro at €239, with Enterprise handled by contract. The Free plan needs no payment details.
What a plan buys is a monthly AI capacity denominated in euros, a source cap, a daily delivery cap, a number of monitoring rules, and a set of delivery surfaces. When a newsroom uses its whole monthly capacity, billable AI requests stop until the next billing period. A lapsed payment reverts the whole workspace to Free limits.
Every number is listed in Plan and quota reference; the payment side is in Billing and AI capacity.
Where to go next
Create your account
Sign in, sign up, and the waitlist and paused-signup states.
Set up your newsroom
The four-step wizard, and what the first question actually decides.
Your first story, end to end
One worked example, naming the exact button at every step.
The feed
What each card shows, how the filters work, and why some stories are not on screen.
Pipeline phases and editorial states
The two status systems in full, and which labels you will never see.
Glossary
The vocabulary the product uses on screen, with the distinctions that trip people up.
Troubleshooting
The messages users actually see, what caused them, and what to do next.
Engineers, operators, and anyone integrating a CMS should start in the platform documentation instead.