OpenErrata is a browser extension that uses AI to investigate the content you read online and provides inline corrections to empirically incorrect or misleading claims.
Corrections are restricted to claims that are relatively uncontestable. If a reasonable person from any perspective wouldn't update on a correction, it isn't shown.
The design, spec, code, and every individual investigation are publicly available. You can inspect exactly why any correction was made.
Unchecked posts show nothing. Checked posts get subtle highlights that reveal details on hover, so the reading experience is never disrupted.
The extension detects when you visit a supported page and sends the content to the OpenErrata service for analysis.
An LLM searches the web for primary sources, cross-references statistics, verifies quotes, and checks empirical claims against available evidence.
Every potential correction goes through a second-stage validation to filter out anything that doesn't meet the bar for being unambiguously incorrect.
Verified corrections are highlighted directly in the page. Hover to see what's wrong and why, with links to the full investigation.
OpenErrata currently fact-checks posts on these platforms, with more coming soon.
OpenErrata ships as a single Helm chart. Bring your own Postgres database, an OpenAI API key, and an S3-compatible bucket. The extension connects to whichever instance you configure.
Deployment docs