OpenArtifacts publishes reviewable HTML/Markdown artifacts from code assistants
OpenArtifacts
OpenArtifacts lets code assistants publish HTML or Markdown pages people can review and share. It adds a human approval step before anything goes public.
Based on reporting by OpenArtifacts — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenArtifacts is trying to turn agent output into something closer to a publish button than a pile of code. The project says it lets people ask an agent to make a page, review the result, approve it, and then share a link that can keep getting updated.
The setup starts with a command: npx openartifacts@latest install. From there, the flow is intentionally chatty and manual. You ask your agent to create a page, review what it made, approve publishing, and then follow a sign-in link before sending the document out. OpenArtifacts calls this “a new mode of communication between agents and humans,” which is a tidy way of saying it wants the machine to draft and the person to sign off.
The free tier is narrow by design. It includes one published document, up to 6 publishes or updates per day in UTC, and 1 MiB of HTML per document. A paid Publisher plan costs $2.99 a month and raises the limits to 500 published documents, 100 publishes or updates per day in UTC, and 10 MiB of HTML per document.
There’s also a path for people with a Copilot for Obsidian license. OpenArtifacts says those users can link the license, enter the key in the browser, and carry over their documents and eligible paid access.
The pitch is broader than a publishing tool, though. OpenArtifacts says it wants agents to read webpages and artifacts natively, support HTML pages with diagrams, tables, and images, and make review part of the sharing process. Its roadmap adds private sharing, inline conversations, version history, visual comparisons, and self-hosting on infrastructure users control. In other words: let the agent do the drafting, but keep the human close enough to catch the weird stuff before it escapes.
My take — AI-written commentary, not fact-checked reporting
This is the right instinct. Agents are good at producing material fast; they’re much less trustworthy when left alone with the send button. The real competition now isn’t who can make the fanciest demo, but who can build the least embarrassing approval workflow.
Read more about this at: OpenArtifacts