Open-source newsroom platform
Papyrus gives independent publishers a staff of agents: a researcher, a reporter, a copywriter and an editor. They find the sources, prepare the reporting and draft the stories. You steer what they find, what they write about and how it reads. Nothing publishes until an editor says so.
Request access Early accessView the source
MIT licensed and free to run in your own AWS account. Managed for $20 a month.
Steer, don't do
Agents research, prepare reporting packets and draft from assignments. You don't hunt for sources or check citations yourself. You steer in three places.
Monitors and research agents bring in new references from the feeds you follow and the open web. Approve or reject each one. The Cyclotron, Papyrus's classifier for what belongs, learns your line and starts suggesting include or exclude.
Agents spot trends across your references and propose stories. Tell them what's worth pursuing and why. Your feedback shapes the next round of suggestions.
Copy-editing agents check every draft against your written editorial policies and underline what to fix. Apply a fix, ask for rewrite options, or dismiss with a reason. Your decisions recalibrate the checks.
Up close: one call
Every new reference arrives with the Cyclotron's suggestion, how sure it is, and why it is in front of you: under the review line, or picked at random so an editor can check its work. Approve or reject, and the call teaches it what belongs. Beta
Dive club starts a weekly underwater basket weaving night
Why you're seeing this: picked at random so an editor can check the Cyclotron's work.
Explainable by design
An opaque pre-trained model can't tell you why it picked a story. A Cyclotron can. Every decision is tied to the exact version that made it, and you can read that version's rubric, questions and examples in plain words. Changes are measured on held-out decisions before they go live.
Editors decide
The first session was attended by twelve divers on 14 October. The organizers expected a handful, but the idea basically just sold itself.
Name who acted: "Twelve divers came to the first session on 14 October."
Filler words weaken the sentence.
| Check | Result |
|---|---|
| Attribution policy | !! 1 Must fix |
| Voice and style | ? 1 Consider |
| Claims against cited passages | Coming soon |
| Figures against source figures | Coming soon |
| Quotes against source text | Coming soon |
Every draft is checked against your publication's own written policies. Findings are underlined in the editor and never applied without you. When you dismiss one, say why, and the checks recalibrate.
Every change an agent makes is a new version with its reason attached, and every move from assignment to draft to published is on the record. You can always see who changed what, and why.
Readable by agents
Every Papyrus page is published for people and as a Markus twin that agents can read: the same story at the same URL. Ask for text/markdown and that's what you get.
Twelve divers came to the first session, and the next three are already full.
$ curl -H "Accept: text/markdown" https://wetweave.example/events/weaving-night --- title: Dive club starts a weekly underwater basket weaving night section: Events published: 2026-10-15 --- # Dive club starts a weekly underwater basket weaving night Twelve divers came to the first session, and the next three are already full.
Every page has a Markus twin. Also published: llms.txt, llms-full.txt, Accept: text/markdown on the same URL, RSS and Atom, and a sitemap. An MCP server for your publication is next. Coming soon
The platform
Researcher, reporter, copywriter and editor agents work from assignments.
Every agent change is a version with a reason. Every stage move is recorded.
Watch sources on a schedule and fill your queue. Hacker News and USPTO TTAB decisions run today.
Add any RSS or Atom feed and run as many monitors as you need.
Paste a URL, email it to your publication's address, or mention it in Slack.
Research assignments search the web and find new candidate sources.
Accepted references, a topic tree and a graph of entities. Editors browse it as a concepts wiki.
The same knowledge, shown to your readers as a wiki, with weekly trends and emerging themes. How it works.
Front pages with exact continuations, solved before the page is drawn.
Issues composed from your publication and sent after an admin approves. How it works.
References, decisions, policies and archive, as files you keep.
MIT licensed. Your account, your data, your sites.
Built with Papyrus
These publications and sites are built and published with Papyrus. They're all Anthus's own, for now.
Pricing
Papyrus is the same software at every tier. You choose who runs it.
Do it yourself
No cost
MIT licensed. Deploy it yourself; Anthus does nothing and charges nothing.
We run it
$20 a month
You run the CloudFormation template in your account; Anthus operates it and keeps it updated.
We set it up with you
$20 a month and $100 once
Anthus runs the setup session with you, then operates it.
We build it for you
Quoted
Anthus builds and operates automated newsrooms adapted to your needs.
Talk to us
Tell us about your beat, your sources and your team. We'll show you a Papyrus newsroom set up for something like yours.
Just want something for yourself? A personal feed that learns what you care about, or a personal blog that writes from what you read, is opening in small groups.